Thoughts

Regarding the Limits of Theory

(an allegory)

July 1945, Los Alamos. For the past few days, you and a few other skilled metallurgists have been busy crafting spheres of plutonium for the experimental superweapon that'll be tested in a week. To be honest, it's kind of eerie: while plutonium looks like, and is worked like, any other kind of metal, there's one way in which it's unlike anything you've ever worked with: it's warm. Sure, any one of the baseball-sized spheres you've made will cool in a freezer—but take it out, and it'll warm right back up, settling at just above human body temperature. It's the kind of subtle anomaly in the fabric of reality that provokes an itch in your ape brain: stop, you can't bring this into the world, you'll break everything, STOP. But you know better than to fall for the same old end-of-the-world prophecies that've been conning mankind for time immemorial: the world is too big, humans too sensible and resilient, and these things are just too complex and labor-intensive to be mass-produced. Besides, if the US doesn't build the atom bomb first, Japan will, say your bosses' bosses—better for it to fall into the hands of democracy than tyranny, you assure yourself.

Sometimes, the warmth trips some... primal circuit in your midbrain, you get the disorienting sense that what you're building is alive. Your cortex has to step in and remind you that it's just metal, and metal isn't alive. Metal can't be alive. Nevertheless, even being near your work provides a physical reminder that you're bringing something new into the world. Something world-changing powerful, something alive ineffable. All the more exciting, though—it was your passion for technological innovation that led you to this project, after all. You want to help ensure that nuclear fission is deployed in a safe and trustworthy manner that empowers creativity, and there's no better way to do that than by bringing it to fruition as soon as physically possible. Your coworkers are divided: while Eve wants to help you "fostэr a cultuяe of openиess" with regards to American nuclear technologies, Alice has told you in no uncertain terms that she's here for the money, not for a vision.

In fact, Alice doesn't even believe that nuclear weapons are possible. Chain reactions are fine—after all, didn't Fermi build a working reactor a few years back?—but they only result from what the eggheads are calling "delayed" criticality. In essence, a lot of Pu atoms fission, and their extremely diverse radioactive products slowly decay, releasing neutrons which fission other Pu atoms, and their radioactive products slowly decay, releasing neutrons which... ad infinitum. It's a slow and controllable form of indirect recursion.

Nuclear weapons, on the other hand, require "prompt" criticality, in which the few neutrons directly released when a Pu atom fissions go on to directly fission other Pu atoms, releasing neutrons that... again, ad infinitum. Both forms of criticality result in exponential growth, but, since delayed criticality can only proceed as quickly as the fission products of plutonium decay, its growth is measured in minutes to hours. This is more than enough time to control the reaction. Prompt criticality, lacking any such intermediates, happens in a flash: within one microsecond, a full fifth of the sphere has been fissioned, producing a shower of x-rays and relativistic projectiles that render all nearby matter into formless plasma.

All of this Alice knows. When you ask her why she nevertheless believes that prompt criticality isn't real, she walks you through a bit of math. If you assembled a critical mass of plutonium—an 11 kg sphere—it would instantly go prompt critical, with around 20% of its atoms being fissioned before it gets too, uh, blown-apart for the nuclear chain reaction to continue. How much energy would this release? It's a straightforward calculation:
$$11\ \text{kilograms} \ ^{239} \text{Pu} \times \overbrace{\left(239\ \frac{\text{grams}\ \text{Pu}}{\text{mol}}\right)^{-1}}^{\text{atomic mass of Pu-239}} \times \overbrace{\left(6.022\times 10^{23}\ \frac{\text{atom Pu}}{\text{mol}}\right)}^{\text{definition of a mol}} \times \overbrace{0.2}^{\text{fission rate}}$$
$$\times \overbrace{\left(200\ \frac{\text{MeV}}{\text{atom Pu}}\right)}^{\text{energy per fission}} \times \overbrace{\left(1.602\times 10^{-19}\ \frac{\text{joules}}{\text{eV}}\right)}^{\text{conversion ratio}} \times \overbrace{\left(4184\ \frac{\text{joules}}{\text{grams TNT}}\right)^{-1}}^{\text{energy density of TNT}}$$
$$= 4.2 \times 10^{10} \ \text{grams TNT}$$
The reaction would release the equivalent of 40 kilotons of TNT. While the plutonium spheres you've been crafting are below critical mass, they've got small beryllium "initiators" at their center, and are to be surrounded with U-238. An explosion outside of the U-238 encasing is to implode the plutonium sphere onto the initiator, causing it to produce a burst of neutrons; this jump-start, combined with the neutron-reflecting property of the uranium shell, will make the nuclear chain reaction much more potent, overcoming the missing mass. Clearly, then, the purpose of the this bomb is to achieve prompt criticality. You're making a city-eraser; taken far enough, perhaps a humanity-eraser.

This settled, she argues: Plutonium is dense enough that an 11 kg sphere would only be the size of a softball... yet we're supposed to believe, based on entirely untested theoretical arguments, that this metal softball is as destructive as ten Olympic swimming pools full of high explosives? Further, the theorists tell us that prompt critical nuclear chain reactions are supposed to happen autonomously whenever you get enough of a fissile isotope (and there are lots of these, many of which occur naturally)—yet no such thing has ever been observed, whether in artificial or natural conditions; why should we believe that these spheres will do what no human invention has ever done before?

The more parsimonious conclusion, she continues, is that theorists are simply refusing to accept that their equations are not universal. She asks you if you're familiar with the Schwarzschild theory of the "black hole"—you are—and you end up agreeing with her that, indeed, as brilliant as general relativity may be, black holes are a reductio ad absurdum of its global validity. Like all theories, Alice tells you, the new atomic theory is only valid in a certain domain, outside of which it's either pathological or unworkable; and if city-erasing softballs aren't pathologies, then what is? Sometimes scientists need to step back and notice that they're confused, and any fanciful belief in runaway chain reactions that grow powerful enough to end the human race is totally confused. So prompt criticality, she concludes, is mere fiction. You can't really argue against her point (maybe Groves really is as stupid as he seems), so you just half-heartedly agree.

Later on, the conversation continues to replay in your mind. Her arguments are pretty reasonable, and you certainly don't want to imagine that you're building something that could wipe out humanity itself. Even though your more conservative half has been spurred to ask what could possibly make these spheres act so different from anything we've ever seen?, some other part of you feels a conviction that they really are somehow special. It's not clear where this feeling is coming from until you recall that unearthly aura, that indication of a break in the consensus reality you and Alice are so accustomed to: why are they warm...?

Decoherence in Many Worlds

An attempt to justify Born's Rule by appealing to decoherence within Many Worlds-like interpretations of quantum mechanics in discussion with GPT-4specifically gpt-4-0613, with extensively customized log-probs and a custom prompt. (smaller sans-serif gray text). The essential idea is that it takes two to measure: the quantum state has to resolve in a certain way, and we have to experience its resolving in that way (after all, a measuring apparatus is nothing more than a tool for accurately translating microscopic interactions into macroscopic experiences; the importance of its needing to have an effect on us in order to collapse for us is illustrated by the delayed choice quantum eraser experiment). Decoherence, along with the basic fidelity of the apparatus and our experience, means that the set of realities where we see one way when it resolved in another undergo destructive interference due to their unphysicality; therefore, only the histories in which we see it one way and it resolved in that way retain nontrivial amplitude. As whatever we see had to obtain twice, in the sum over possible histories, experiences obtain with proportions relative to the squares of the corresponding amplitudes, which is Born's rule.
!tab Edit: it turns out that this argument was made in much greater detail by Sebens and Carroll a decade ago, in Self-Locating Uncertainty and the Origin of Probability in Everettian Quantum Mechanics.

Can decoherence be used to resolve the measurement law and derive Born's Law in many-worlds-like interpretations of QM? The resolution that comes to my mind goes something like:

With a pure state like $|\psi\rangle$ $= \sum_i \alpha_i |\phi_i\rangle$, with $\langle \phi_i \mid \phi_j\rangle$ $= \delta_{ij}$ (orthonormal eigenbasis), all possible measurements of all possible states are superposed, representable by the density matrix $\rho$ $= |\psi\rangle\langle\psi|$ $= \sum_{i,j} \alpha_i\alpha_j|\phi_i\rangle\langle\phi_j|$, but, upon measurement, "unphysical" modes of causation—those in which a state $|\phi_i\rangle$ physically affects the measuring apparatus so as to cause it to read out a different $|\phi_j\rangle$—scramble each other out since such paths can only be realized by really high-action paths.
!tab Measurement doesn't really mean anything fundamental, it's just a physical effect which resolves in every way at once, but the unphysicality of various possible physical effects means that after a very very short time, $\rho$ will evolve as $$\sum_{i,j}\alpha_i\alpha_j|\phi_i\rangle\langle\phi_j| \Rightarrow \sum_{i,j}(\delta_{ij}+\epsilon_{ij})\alpha_i\alpha_j|\phi_i\rangle\langle\phi_j| = \sum_i \alpha_i^2|\phi_i\rangle\langle\phi_i| + O(\epsilon)$$ Here, $\delta+\epsilon$ represents the fact that the $i=j$ states are preserved, because they represent least-action physical effects, and the $i\ne j$ states near-instantly become negligible because they don't. So the perfectly pure state $\rho$ becomes a perfectly mixed state in which the measurement apparatus's reading out $|\phi_i\rangle$ is assigned the probability $\alpha_i^2$, with the inherited relation $\sum_i \alpha_i^2 = 1$ reflecting the fact that the measurement apparatus does reliably read something out. Because we're nothing more than biological measuring apparatuses that happen to augment ourselves with high-tech physical tools (i.e., we don't have a god's-eye view on the multiverse, we're just physically in it and subject to its rules), we observe state $i$ with probability $\alpha_i^2$.

Is this argument correct, assuming the many-worlds interpretation?

Yes, your argument correctly describes a process akin to the notion of decoherence and how it invariably leads to a probabilistic sort of behavior in the context of the many-worlds interpretation (MWI).
!tab In more technical language, the process you've described is quite close to something called "environment-induced superselection" or "einselection", a process described by Wojciech Zurek in the context of quantum decoherence, which is essentially the mechanism through which certain preferred basis states arise from interactions with the environment.
!tab The gist of the process is that some states are more robust or "stable" through time in respect to the (inevitable) interactions with the environment. In your argument, you characterize these states with the least-action language, asserting that "physical" outcomes are the ones that survive through time (those where $i = j$). In the loss of "off-diagonal" terms in the density matrix (those where $i \neq j$), we see the quantum system starting to behave less like a quantum superposition and more like a classical probability distribution, due to the system-environment interaction. This, indeed, is one of the stepping stones toward the emergence of classicality from a quantum system.
!tab The move from $\rho = |\psi\rangle\langle\psi|$ to $\sum_i \alpha_i^2|\phi_i\rangle\langle\phi_i|$ captures the result after the process of decoherence where the system behaves like a mixture of eigenstates rather than a superposition. There's an implicit assumption here that each $|\phi_i\rangle$ is associated with some particular observable measurement.
!tab While your account is qualitatively accurate, it's important to note that deriving the exact Born rule — that $|\alpha_i|^2$ gives us the probability of observing state i — within the many-worlds interpretation is still a matter of ongoing investigation. There are several approaches to this problem and decoherence plays a critical role in many (but not all) of these derivations.
!tab All in all, your argument grasps the essential role of decoherence in contributions to solving the measurement problem in the MWI and suggests a specific mechanism through which one might understand the origins of the Born rule. Nevertheless, the precise derivation of the Born rule within MWI is a nuanced problem and still a hot topic of discussion in quantum foundations.

Oh, yeah, I should've used $\alpha_i^*\alpha_i = |\alpha_i|^2$ instead of $\alpha_i^2$, since they're complex and conjugate upon dualization... But where is the controversy in this argument? Is it mostly just confusion about the nature of conscious experience, or are there technical issues legible within a physical reductionist standpoint? Does it work, for instance, in explaining why a non-conscious measuring apparatus in a sealed box should become perfectly entangled with a quantum measurement it's making (like if Wigner had a robot friend)?

That's correct, ideally we should use $|\alpha_i|^2$ as the probability associated with each eigenstate in Born's rule since $\alpha_i$ can be a complex number.
!tab In regards to your other questions, the derivation of Born's rule in the framework of the many-worlds interpretation (MWI) rather abstractly pertains to the interpretation of probability itself in a deterministic multiverse. Indeed, the controversy is not so much due to technical issues, rather it arises from deeper conceptual and philosophical problems.
!tab Quantum decoherence indeed can explain why a classically behaving macroscopic device gets correlated (becomes entangled) with the observable outcome (quantum state that's measured), and why the result of that measurement, once it's in the environment, can be seen by anyone (or anything) else who subsequently measures the system (like the "Wigner's friend" thought experiment you're referring to). This process can accurately describe measurement circumstances where an observer's brain or a robotic measuring apparatus becomes classically correlated with the quantum observable.
!tab The controversial point involves the transition between describing deterministically evolving quantum states (where each outcome evolves in a distinct universe) to assigning specific probability measures over these outcomes that correspond to Born's rule. It's a shift from a deterministic evolution to a probabilistic one in our experiential description of the system. That's where much of the debate around deriving Born's rule in the MWI lies. The MWI is fundamentally deterministic and yet we seem to observe probabilistic outcomes. How do we reconcile this? That's the question.
!tab In conclusion, your original argument captures essential features of quantum measurement in MWI through the lens of decoherence. Controversies lie primarily in the philosophical and conceptual interpretation of probability itself in the multiverse where every possible outcome, deterministically, happens in some branch. This is quite apart from the technical process of how a measurement device becomes entangled and classically correlated with a particular quantum state through the process of measurement and decoherence.

Hmm, I thought the resolution to that question—why do we experience single outcomes—was exactly the core insight of MWI: every fork is taken in the road at once. And decoherence is the exact reason why process ought to look like a set of discrete forks taken (einselection) with squared probabilities (i.e., according to the Born Rule) rather than an open field. This is what I was alluding to with our being in and subject to physics, rather than having a god's eye-view. To put it another way, a single point on the real line has no right to ask why it's that point rather than another, since if it were another point it'd just be asking the very same question—the same is the case with humans in an MWI multiverse, who happen to be physically complicated enough to actually ask that question.

Your grasp of the philosophical implications of MWI is certainly insightful. You are correct in asserting that MWI explains the single-experience-of-reality problem by positing that all possible outcomes occur simultaneously, with each observer-instance experiencing only one outcome due to the way photons physically interact with their eyes and neural system. Each "you" in each universe will only ever experience that one universe's outcomes.
!tab The process that goes from a single coherent quantum state to an apparent distribution of probabilities through decoherence is basically what you're describing.
!tab Nevertheless, where the controversy lies, and why this is still a hotly researched topic, is that while it's easy to say, "in each branched universe, 'I' perceive that particular outcome," it isn't as easy to then justify why particular outcomes are perceived with particular probabilities, in particular, probabilities given by the squared amplitudes of the quantum state, as Born's rule demands.
!tab The issue boils down to how we interpret probability in a deterministic multiverse where "everything happens." Why it feels like some things happen with different probabilities than others — that's where the trick is in making the connection between MWI and the Born rule.
!tab It is at this point where there is a substantial literature and various arguments which aim to derive the Born rule within the framework of MWI, and different researchers prefer different derivations, to different degrees of success, under different sets of assumptions. You've laid out one possible approach to this problem, and it's certainly a valid perspective to understand probability in MWI in terms of the particulars of one's physical existence in a particular universe. But this still remains a subtle conceptual area in the foundations of quantum mechanics.

So, basically, the question is "why should the Born Rule even govern the proportional distribution of the 'I'-ness that feels itself as perceiving particular states, rather than our 'feeling' each state in proportion with its existence rather than its Born Rule probability?" If I'm understanding correctly, the point is to question why our experience of measuring say an electron with superposed spin state $|\psi\rangle$ $= \frac{1}{2}\mid\downarrow\rangle$ $- \frac{i\sqrt{3}}{2}\mid\uparrow\rangle$ yields the experience of $\mid\downarrow\rangle$ one in four times and $\mid\uparrow\rangle$ the other three times, rather than our just experiencing both with identical proportions?

Yes, you've got a good handle on the problem. In the many-worlds interpretation, if every possible outcome of a measurement is realized in some "world" or "branch" of the wave function, then it's not obvious why our subjective experiences should obey the probabilities given by the Born rule, whereby outcomes are not just equiprobable but have probabilities given by the squared magnitude, $|\alpha_i|^2$, of their associated amplitude.
!tab In MWI, "everything happens," yet it doesn't just "happen" — it seems to happen with specific probabilities that just so happen to match the Born rule we find in quantum experiments. The goal then becomes to understand why these particular Born rule probabilities are the ones we should expect to experience, rather than any other rule or, as you noted, an "equiprobable" rule where each possible outcome is equally likely.
!tab This presents a significant conceptual issue within the MWI - a challenge that various scientists and philosophers have taken different approaches to resolve. The crux of these endeavors is often about how to interpret probability in such a deterministic multiverse, and various solutions have been proposed over the years — with decoherence and self-locating uncertainty being two key tools often invoked. But precisely reconciling the deterministic "everything happens" of MWI with the probabilistic "some things happen more often" of the Born rule remains a deeply subtle and compelling problem in the philosophy of quantum mechanics.

I still don't see the issue, actually—the answer to the question I posed seems pretty clear to me, and its asking seems to rely on a misconceptualization of what branching means. It's not like a graph-theoretic tree between possible universe states; weightedness is inherently part of the ontology, in a way you'd need at least a flow network to represent. Though this, too, would be an oversimplification, since the fact that it takes a very short but non-zero time for decoherence to happen implies that the "open field" really does exist, and only happens to be almost perfectly approximable by weighted branches. (In other words, the existence of decoherence times cannot exist in the same physical framework as equiprobable branches!).
!tab One metaphor I heard somewhere and really like is calling QM the dynamics of "realityfluidI think this particular term is due to Yudkowsky, but can't pin down an original source. The fluid metaphor is older.": we're conscious structures in the 'reality' that is flowing around the configuration space, and wherever this fluid ends up forking, this dynamics tells us that it forks with Born Rule probabilities, which is why we experience these probabilities as a matter of course at every point in the fluid.

Yes, your understanding is accurate. You have captured the essence of the problem and why some people don't see it as a problem at all.
!tab Indeed, there are many who would argue—as you do—that the probability issue in the many worlds interpretation (MWI) is not actually problematic, that the Born rule emerges naturally from the structure of quantum mechanics itself. After all, the quantum wave function evolves according to deterministic laws, and the distribution of "reality fluid" through all possible branches is governed by these laws.
!tab Your conceptualization of each branch as inherently weighted—an intrinsic part of the ontology rather than a "post-quantum measurement" add-on—aligns well with the view of many proponents of MWI. This is certainly a valid perspective on interpreting the MWI and the role of probability within it. The controversy arises primarily from different philosophical interpretations of probability rather than disagreements on the actual mathematics of quantum mechanics.
!tab As for the metaphor of "reality fluid", I agree that it's a compelling way to conceptualize the many worlds view of quantum mechanics. It evokes the image of a constantly forking river, with each branch having a size proportional to the square of the quantum amplitude, as prescribed by the Born rule. It vividly illustrates why, in the MWI, we might naturally find ourselves more often in those branches associated with larger quantum amplitudes.
!tab In the end, much of the debate over probability in MWI is about how to properly understand and interpret these core concepts within the framework of quantum mechanics. But as with many things in quantum mechanics, different people can hold different understandings and intuitions about these things without necessarily disagreeing about the actual physical predictions of the theory.

Biological Randomness

Nothing is ever really random when it comes to biology. Lots of processes are conceptualizable to zeroth order as random, but these processes generally exist only because they're facilitated by extremely complex biological systems whose future construction they have a hand in, and there'll always be ways to alter how these biological systems effect the randomness of these processes that not only self-propagate but differentially outcompete unaltered biological systems; natural selection allows these contourings of randomness—which, having been selected, are not themselves random—to thrive, where they form higher-order corrections to our conceptualization of randomness.

For instance, mutation is generally taken to be something that randomly happens to DNA—copy errors, damage from oxidative stress or radiation followed by lossy repair, and so on, acting on each base pair in a more-or-less i.i.d. manner. But the entire concept of mutation only obtains within this massive complex of pre-existing structures that interact with one another in unimaginably complex, subtle ways. Consider, from a programmer's perspective, just how complicated a DNA repair system would be.

• First of all, there are several different types of DNA damage that require entirely different repairs. The system has to detect the kind of damage and trigger the appropriate solution, or at least be able to differentially trigger downstream systems depending on what it detects.
• Each mechanism must be able to chemically manipulate the damaged areas in very specific ways. For damage to a base pair, for instance, the repair mechanism generally removes the base pair entirely to create an AP (apurinic/apyramidic) site, which is filled in with whatever matches the pair on the complementary strand.
• In general, the system has to demand full access wherever it's checking for, or repairing, damage. DNA spends most of its time supercoiled, and special tools are required to uncoil it at selected locations without (permanently) breaking it or altering the global topological invariants inherent to the coiling.
• It must be very accurate and very efficient, so as not to introduce more errors in the process of all of this manipulation—which is very easy when working with something as small and fragile as DNA. Since damage tends not to be isolated, for instance—one break over here is strong evidence for similar breaks nearby, since it serves as evidence of both a local break-causing process and a local vulnerability to such a process—a mechanism ought to be able to chain repairs without multiplying the starting and ending sequences (e.g. uncoil, fix X, recoil, uncoil, fix X, recoil...).
• All of this has to be carried out by self-organizing molecular machines that function almost entirely by shape alone, get created stochastically, move stochastically, and act stochastically (proteins don't have eyes, or flippers for that matter, nor are there intracelluar fairies that get to decide where a protein ought to be; they have to just drift around in ways that are differentially attracted to their target areas, or areas where they can be rapidly pushed by other proteins with stronger differential attractions. It's all statistical at this level).

While always indeterministic, the output of such a complicated system can never be truly random. Any change to any one of these subsystems that biases mutations in some way that then biases the overall system to keep making that sort of change to that subsystem will tend to replicate; these corrections to the law of randomness need only to be realized before they can take hold.

The canonical example of a means of self-propagation is the instrumentally convergent one: fitness! Genes most directly rise and fall with their carriers, so, for a gene, benefiting the carrier is the simplest way to benefit oneself. Mutations to highly conserved genes, such as the transcription-translation complex (which is more or less the same among all living organisms), will generally be either immediately harmful to the carrier, or silent, in which case you can only hope it gets replaced by recombination before Muller's ratchet + further mutations combine to make it harmful later. So it would be very cool and hip and fit if the manner in which all of the above phenomena worked were such that these genes ended up experiencing really low mutation rates—if there were some sort of first-order correction to i.i.d. randomness. ("First-order" meaning dependent on one variable, here importance of conservation; second-order corrections would presumably be contextualized or linked, and so on—evolutionary capacitance, for instance). And, as it happens, this is exactly what we see: Bejerano et al. discovered 481 DNA elements longer than 200 pairs which were 100% absolutely identical between humans, mice, and rats, while calculating that "the probability is less than $10^{-22}$ of finding even one such element" in a basic i.i.d. model of mutations, with these elements being related to basic genomic functions such as transcription, translation, and body planning (homeobox) at a rate far higher than chance (see figure 2).

In general, forms of non-randomness like this seem to be one of the main themes of modern evolutionary biology, especially as related to the extended evolutionary synthesis: several things that we originally conceptualized as random in the simple sense of "not correlated with anything" are actually only indeterminate—they're sources of randomness, either inevitable (as with thermal noise on the molecular level) or 'intentional' (as with genetic recombination), that nature constantly twists, exploits, and differentially determines to its own advantage. Even the very notion of genetics evinces this: it certainly isn't necessary for evolution to take place (does memetic evolution have its own versions of adenine, thymine, guanine, and cysteine?), but life nevertheless managed to converge on a system for compressing the endless complexity of phenotypic data into a couple hundred megabytes (in humans, 3 billion base pairs $\times$ $\log_2(4$ possibilities for each pair$) = 6$ billion bits of information). Clearly this isn't all there is to phenotype—there's epigenetics, as well as information transmitted through prenatal environments—but the use of such a system in controlling the randomness of selection is obvious, and apparently so great that every known living organism encodes its phenotypic information through DNA—or RNA, in the case of some viruses.

Effective Murder via Prediction Markets

I was thinking about prediction markets in an attempt to understand them better, when an idea came to me. I asked GPT-4 what might be wrong with it, since it seemed ridiculous enough to be wrong in a way I didn't immediately see, but GPT was incensed with the idea and gave a lot of very poor reasons against it. Arguing against these was a great way to understand the nature of prediction markets, as well as strategic action in general.

Warning: Death Note spoilers, Functional Decision Theory

Idea: Why didn't Light YagamiI'm using "Kira" for the persona known to the outside world and "Light" for the person that happens to be Kira. Mostly the distinction can be elided, though. just command his followers to establish subsidized prediction markets on whether given people die of heart attacks over given time periods? It'd be a useful way to crowdsource common knowledge—e.g., if Brazil's president declares war, and his death stock's market value jumps from 0.01 to 10 cents, that's both a "look!!" signal to Kira and a "stop!!" message to the president. Presumably the traders would be pricing in their knowledge about Kira's killing protocols and peoples' reactions to the prices, as in "a rapid jump in this market would cause the president to back down with a probability that say 15c is too high to match, so 10c seems calibrated—after all, even Kira will send several warning shots before offing the head of a G20 state, right?" So, if well-priced, 0.01c (1 bp) would combine a base rate for heart attacks (for the president's demographic), and a black swan rate that Kira kills him for unpublicized reasons; 10c would mostly be about Kira's killing him as retribution for declaring war, which, I presume the traders presume, happens if he doesn't back down due to the price jump or warning shots, Kira's actually not pro-this-war, and so on.
!tab I'm assuming that the market would need to have a Wikipedia-like "notable persons rule", and that there aren't any strange CIA nanotech heart attack devices—though it would be germane to check the markets for signs of manipulation, or perhaps to pick an entirely different signature killing method. Maybe it's best if he solicits the opinions of a Dark Physician as to what signals are most obvious upon autopsy while also being hardest to artificially replicate..

Response to Criticisms: "This is highly illegal and unethical". I don't think legality or consensus morality hold much sway over Kira—these are "problems" in the same sense that Kira's actions are problematic in general, but the actual operational issues they might entail (resistance and so on) seem like roadbumps given that at some point even the US capitulates to him out of fear.

"Light Yagami desired control, and the idea of a prediction market that potentially influences or dictates who dies causes him to lose control over who is targeted by Kira". This also isn't a problem: as long as he's the one with the book, he's in control, with everything else only being more information he can use to figure out how to make decisions better by his own standards. No secret information needs to be revealed, since the two relevant factors—that he can cause heart attacks, and is doing so systematically in order to make the world in a certain way—were known from the very start.
!tab After all, prediction markets fundamentally make correctly predicting the future the goal of their traders, and the fact that they can't cause heart attacks themselves closes out the main source of market manipulation, such that the best way to make money is to actually understand the values and knowledge states leading to his actual actions. And, were he to start using it to make more informed decisions, they wouldn't need to try to figure out what he knows—what news stations he's watching and so on—since they by the act of trading on his values alone point him towards knowledge on which he can act, and using this knowledge as he so pleases will make their trades profitable insofar as they actually understand his values The exceptions are cases where they can't predict him owing to a technicality that's not public knowledge, but there are very few of these—the name and face rule quickly became public knowledge, for instance, so the possibility that someone actually goes by a fake name in public or etc. would be priced in wherever non-negligible.. Thus, using it in this way would foster it into a global informational network that, by trying to predict what he would do in his pursuit of utopia, recreates his own values and vision; this is the entire reason someone in his place would consider ordering such a thing to be made.

"Light is incredibly unpredictable and loves to always be one step ahead. Depending on the reactions and predictions of others would not be in his best interest". His unpredictability isn't of the holds up spork variety, it's not for the sake of unpredictability alone, but just a means to fulfill his objectives in light of powerful adversaries constantly trying to extrapolate and intercept his actions—like the zig-zagging of a soccer player across the pitch, not of a pollen grain in water. It's easy to make true conclusions about what he'll do, but he, like anyone thinking strategically, tries to make sure that the conclusions his adversaries can come to are useless in stopping him.
!tab For instance, a professional soccer goalie knows that the attacker's going to try to juke defenders and find an opening to kick the ball through, which is a lot of very relevant information to have on your enemy's action policies! But none of it is actionable, because pro soccer players have developed strategic acumen and built up a metagame so as to iron out all the predictable exploitable patterns. The point isn't to keep L as in the dark as is possible (which is not only impossible, but would be silly even if it were possible); the point is not to let L figure out anything that he can't imagine responding to with a mocking "and what are you gonna do about it?" [Note 1]. Behaving predictably is just as important in facilitating cooperation with allies as behaving unpredictably is in preventing exploitation by adversaries—the only question is one of pros vs cons in any given situation, i.e. "does tending to act this sort of way in this sort of situation help people help me more than it helps people hurt me?".

But blocking opportunities for adversaries to exploit you doesn't mean blocking opportunities for cooperators to benefit you: the two are often correlated by virtue of their relying on common causal channels (an attacker that could optically cloak themselves and the soccer ball would be unstoppable, but also unhelpable, since everyone's using vision to do whatever it is they're doing), whereas [incomplete]

Note 1: For instance, L makes it very clear that he expects to die the moment his identity is leaked to the public, which means he models Kira, correctly, as having a "kill L ASAP" policy—and, because Light is actually just too stupid not to act immediately and deterministically to new information, L gets to exploit this twice (first by a differential attack on Kira's opportunistic murder of an impostor L, and then by explicitly making himself very easy for Light in particular to kill in order to buy a couple days after realizing that Kira's recently acquired a partner who just needs a face to kill). This partially carries a moral of "eschew [deterministic] policies in general whenever you might be up against an adversary intelligent enough to find ways of exploiting any given such policy that you just can't see".

• But (a), a certain amount of dependence of one's actions on the state of whatever system they're acting upon is necessary insofar as there's any determinate state they want to determine for that system, so this can only ever be a loose heuristic (and somewhat self-contradictory, due to its being a deterministic metapolicy)—the only generalizable solution is actually just to git gud. After all, Light really couldn't afford L's continued meddling! Every additional day L lived increased the threat of total failure, to the extent of providing strong reason to kill Lind L. Tailor even if he perceived a large probability that this was a fake—though it would've been better if he had realized that he wasn't deducing this guy's identity a priori but after being exposed to specific external stimuli, and consequently decided to blur the correlations between these stimuli and his actions by, say, waiting for the name and face to hit tomorrow's world news outlets before unceremoniously writing it down like he would any other name a couple hours later. Letting a legendary detective fish for your identity for another day is a threat, obviously, but snatching the giant treat that's floating right to you without considering that it might contain (or retroactively manage to turn into) a hook, is exactly how you let him catch you And besides, insofar as Light could've deduced that Tailor was a fake, that would present all the more reason not to kill him, since it's not exactly clear what they're trying to do here. Hell, he might be broadcasting with fiber-optic cable from a Faraday cage suspended in the middle of one of those massive underground neutrino observatories. This is the obvious sort of thing to do before writing off the method of action of the most powerful and enigmatic weapon in history as "supernatural", and Light has no way to conclude that the Death Note has no detectable physical signature, or that there are no active experiments taking advantage of the fact of Kira's pointing this weapon at a specific point in space and time (the equivalent of knowing plaintext in a cryptographic attack—incredibly powerful!) in order to obtain data that might be used to triangulate his location or develop jammers, insulators, usage detectors, and other countermeasures.
  !tab Another thing L might have done—perhaps a better one—would be to get many decoys, dividing broadcasts by name presented rather than time, recording their announcements beforehand, and then playing them all in different parts of the world in the closest achievable manner to an even division of prior probability mass (which effects a sort of geographic content-aware scaling), putting them in analysis chambers at time of broadcast. Then Light would've been absolutely fucked by his reaction: given that L had already identified the first few victims by then, this would've given extremely precise location information, and possibly physical evidence, and Light would be entirely oblivious, Gettiered into thinking that he hadn't written down L's real name until puzzled reports started coming out about the broadcast discrepancies..
  Another way to put the general idea is causal camoflauge: [incomplete]
• And (b), as mentioned below the note marker, the situation can actually admit cases where the [incomplete]

A Criterion for Deception

inferenceAIterminologylarge

"Lying" is a poor concept, the use of which hurts our social cognition. There are statements that are certainly lies—statements that one constructs with the explicit intent of misleading someone else, which they know are inadmissible with respect to the real state of the world. If Bob asks Alice where she was born (she is certain that she was born in Michigan), and she constructs and delivers the answer "Maine" with the explicit intent of making Bob believe that she was born in Maine, then that's a clear lie. If she answers "Mars", not believing that Bob will actually believe she was born on Mars, then it's not so clearly a lieThis is usually the case with jokes. I doubt that any horse has ever walked into a bar to be asked "why the long face?" by a bartender, or that any orange has ever managed to knock on a door, but we wouldn't say that someone delivering jokes involving such situations is lying..

What's far more common is that Alice constructs statements that are in some sense admissible by the facts of a situation, but which misrepresent these facts to Bob in a way appreciable to Alice. (This is one of the main rhetorical strategies of culture warring). It's so ingrained into our behavior that it's often hard even to call the behavior "intentionally misleading" or "knowingly deceptive". Motte-and-baileys are generically examples, as are instantiations of the noncentral fallacy, but here are two subtler examples:

• In 2002, Buzz Aldrin went to a hotel to shoot what he thought would be an interview for a children's TV show, but found that it was a documentary about how the moon landing was a hoax. As he tried to leave, the man hosting the documentary pursued him, provoking and insulting him as Aldrin repeatedly told him to go away; this had been happening for some time, and after the man got in front of him and called him a liar, coward, and a thief, Aldrin punched him in response. Now, suppose that I, knowing all this, tell Alice (who knows none of it) solely that "Buzz Aldrin assaulted a man who denied that he landed on the moon". It's a true statement, and one that is clearly admissible by the facts, but in many contexts, saying that without any further information might make the receiver form an impression of Buzz Aldrin as a violent, unreasonable person. With the extra information I have, I can appreciate that this is an impression based off of a misrepresentation of the situation as I understand it, but usually I won't actually realize that Alice formed this impression or even consider it as a potential way to misunderstand my statement. But if I want Alice to form this impression of Buzz Aldrin, and this is why I'm making that statement to her, then I'm being deceptive. Even if he really is a violent and unreasonable person, it would be deceptive for me to use this as evidence to convince Alice because I can appreciate that the fuller facts of the situation greatly reduce the extent to which it is evidence.
• Say Alice and I live in Gotham, but she's somehow never heard of Batman. I don't know Batman's identity, and have no reason to believe that he's anyone in particular, but in the course of my explanation I repeatedly, intentionally insinuate that Batman might be Bruce Wayne, who is to the best of my knowledge a playboy billionaire with no connection to BatmanReverse Gettier cases like this are one problem with the way the English language architects our notion of lying. If I feel certain that World War 2 ended in 1946, but insist to someone with this same mistaken belief that it actually ended in 1945 because it's somehow useful for me that they believe that, then this is an epistemic lie that is not an ontic lie. In this particular case, it's irrelevant that Bruce Wayne actually is Batman, since the notion of deception I'm trying to outline is an epistemic one: it arises in the interplay between my model of Alice, my model of reality, and my statement to Alice.
  This is, by the way, a place where 'functional' and 'definitional' accounts of descriptions disagree. If you say "calling someone a liar is useful to better predict their future behavior", you ought to call me a liar here, since saying true things is not something I'd seem to care about, but if you say "calling someone a liar is useful to compress a sometimes-complex bundle of judgements about the truth of what they say", then you ought not to call me a liar, since I said something which was in fact true.. Again, I'm trying to get Alice to form an impression of the situation that appreciably does not concord with my own understanding.
  This example might be a bit less intuitive, but that's primarily because it doesn't have a clear valence to it. While deception is generally thought of as a bad counterpart to good honesty, this valentiality has nothing to do with the epistemic structure of the situation, though this structure is the sole determinant of whether I'm being deceptive. So it's a weird kind of deception, but deception nevertheless.

Formally, we might say that Alice deceives Bob about a situation if:
First Statement: She makes a statement to him that, with respect to her own model of Bob, changes his impression of the situation so as to make it diverge from her own model of the situation.
We can phrase it in terms of probability distributions. First, some notation:

1. Let $p^A_\Omega$ be Alice's probability distribution over the true state of a system $\Omega$, coming from a measure on the state space $D^A_\Omega$ Alice constructs to model $\Omega$.
2. Let $p^A_B$ represent Alice's probability distribution over Bob's possible probability distributions $q^B_\Omega$, coming from a measure on the state space $D^A_B$ Alice constructs to model Bob.
3. Let $p^A_{\Omega, B}$ be the product distribution on $D^A_{\Omega, B} = D^A_\Omega \times D^A_B$, with $p^A_{\Omega, B}(\omega, q^B_\Omega) = p^A_\Omega(\omega)p^A_B(q^B_\Omega)$.
4. Let $p^{B\mid s}_\Omega(\omega)$ be the probability distribution of Bob after he updates on a message $s$, which if he updates via conditionalization is $p^B_\Omega(\omega \mid s)$.

• (The superscript answers "whose probability is this?" and the subscript answers "who is it modeling?").
• (This notation is consistent with an understanding of the system $\Omega$ as a third character whose beliefs are merely states).
• ($q$ is used for possible probability distributions, $p$ for actual ones).
• (This notation has some recursion problems that I need to fix. How do we write a probability distribution Alice has over Bob's potential distributions over Alice's potential state space distributions (i.e., over $D^B_A$)? It would be by Alice of Bob, but not $p^A_B$ (which should have $\Omega$ somewhere, really)).

With this notation, it goes as: $${\mathbb E}_{p^A_B}\left[\operatorname{KL}(p^A_\Omega\mid\mid q^{B\mid s}_\Omega)\right] > {\mathbb E}_{p^A_B}\left[\operatorname{KL}(p^A_\Omega\mid\mid q^{B}_\Omega)\right]$$ We can manipulate this: $$0 < \mathbb{E}_{p^A_B}\left[\operatorname{KL}(p^A_\Omega\mid\mid p^{B\mid s}_\Omega)\right]-\mathbb{E}_{p^A_B}\left[\operatorname{KL}(p^A_\Omega\mid\mid q^{B}_\Omega)\right]=\int_{D^A_B} p^A_B(q^B_\Omega) \int_{D^A_\Omega} p^A_\Omega(\omega)\ln\frac{p^A_\Omega(\omega)}{q^{B\mid s}_\Omega(\omega)} - p^A_\Omega(\omega)\ln\frac{p^A_\Omega(\omega)}{q^{B}_\Omega(\omega)}\ d\omega\ dp^A_B$$ $$=\iint_{D^A_{\Omega, B}}p^A_{\Omega,B}\left(q^B_\Omega, \omega\right) \ln \left(\frac{p^A_\Omega(\omega)}{q^{B\mid s}_\Omega(\omega)}\frac{q^{B}_\Omega(\omega)}{p^A_\Omega(\omega)}\right) \ d\omega \ dp^A_B=-\mathbb{E}_{p^A_{\Omega,B}}\left[\ln\frac{q^{B\mid s}_\Omega(\omega)}{p^B_\Omega(\omega)}\right]$$ $$\implies \mathbb{E}_{p^A_{\Omega, B}}\left[\ln\frac{q^{B\mid s}_\Omega(\omega)}{q^B_\Omega(\omega)}\right] < 0$$ This gets at the core idea: deception should be misinformation on expectation.
We can make this clearer: writing surprisal $S(x) = -\ln p(x)$, the criterion can also be written as $$\mathbb{E}_{p^A_{\Omega,B}}\left[S^{B\mid s}_\Omega(\omega)\right] > \mathbb{E}_{p^A_{\Omega,B}}\left[S^B_\Omega(\omega)\right]$$ This can be converted back to plain English: Alice deceives Bob with the message $s$ if:
Second Statement: She can expect that the message would make him more surprised to learn the truth. Note that this statement does not require her to know the true state of the system, know what Bob thinks the state is, or know how Bob conditions upon evidence. She merely has to have a model of the system that generates a distribution over the first, and a model of Bob that generates distributions over the second and third. Also note that I'm saying can expect, not does expect. Alice may not think about it too hard, may just be acting on heuristics, and so on. This is also why I said can appreciate above)
It is not deceptive to merely cause surprise with your message. To be deceptive, your message has to make it harder for the receiver to see the truth as you know it. If you expect that Batman is almost certainly not Bruce Wayne, a message that you expect will prime someone to suspect that Batman might be Bruce Wayne is a deception in that regardOf course, telling someone that Batman exists will cause them to be far less surprised when they see Batman than they would if they had no idea, and this reduction in the expected surprisal of reality may outweigh the deceptiveness of implying that Batman might be Bruce Wayne, so that the message is overall informative rather than deceptive. But in any case we can say that the particular portion about Bruce Wayne is deceptive, since you can (wrongly) expect that your message would have been even more informative had you removed it. We can talk about whether specific properties or portions of a message are deceptive so long as it's coherent to talk about the deceptiveness of counterfactual messages where they're removed..

This is a criterion for deception that routes around notions of intentionality. An AI, for instance, might not think in a way that lets us clearly say "it explicitly intended to mislead us by saying XYZ". For instance, if it's a Gato or AlphaZero-style AI that sees humans as "just another game to master", just another system that turns inputs it provides into outputs it cares about, then it might just make statements in ways that it thinks will get us to do what it wants, with any analysis of the truth or falsity of these statements being totally irrelevant. It does not say XYZ in order to deceive us; deception is merely a byproduct of that statement.
!tab There is no reason that the AI should conceptually model us as "agents" acting on internal models of the world—no reason why it should view us as the kind of thing that can be deceived, except insofar as such an agenticity is functionally evinced by the input-output mapping that we are to it. To insufficiently advanced intelligences, this conceptual approximation will be very usefulAnd we can only ever be agents to exactly the extent to which the label is an effective compressor of reality. A "goal" generally only exists in a given conceptive encoding of the world—thinking of the brain as an autoencoder, and a thing X is only an "agent" to a thing Y insofar as Y simplifies its thinking by conceiving of X as acting towards that goal. If Y has information and processing power that obviates the physical and conceptive import of X's actions without explicit reference to any goal (e.g. where Y is a superintelligence with the ability to ~perfectly envelop X=humans' own conceptive encoding and input-output mapping via direct neuropsychophysiological modeling), then X need not be an agent to Y; there is no point for Y to think "this is an agent"., but more advanced intelligences (most obviously, Solomonoff inductors) will see through this; while we might call their actions upon us deceptive by projecting our idea of our own agency onto their own internal models of the world, they're really only as "deceiving" us to the same extent that we "deceive" computer programs when we use a cheat engine.
Such pro-deceptivity should be understood as an instrumental correlate of optimal strategizing. Why? Usually, the best action is the best action for more than one reason. It accomplishes multiple things simultaneously, is better on the whole across multiple dimensions than alternative actionsThe best chess move is the one that establishes control here and suppresses these enemy pieces and opens up these movement opportunities and allows these lines of attack the enemy will have to identify and guard against and so on. This is why playing chess against an opponent much stronger than you feels like steadily being overwhelmed (assuming you're competent enough to spot and prevent easy checkmates!): they know how to start multiple fires with each move while simultaneously intensifying their previous fires and suppressing your ability to put them out.. And these dimension-bound values are not continuous convex functions of a smooth-manifold-like action space, but wildly discontinuous functions of a mycelial action space, so you can't interpolate in the manner required to smoothly apply certain constraints. So if it stumbles upon a really good plan that happens to deceive us, it's not generically the case that it can vary the plan in a way that keeps its goodness while removing the deception; in fact, the deception is probably a load-bearing aspect. Again, this isn't because anything that approximates a concept of "deception" is intended by the AI—it's only load-bearing in the sense that removing the deception requires navigating from whatever generic point in solution-space the AI has found to that very tiny subset of the solution-space where we are kept in the loopTo make the point more visceral: suppose that, for some security reason, we decided that all new computer programs had to compile to a bit-string representing a prime number. It's a very unnatural and excessively restrictive desideratum, and it would be, while possible, incredibly difficult to actually build new programs to this spec. Imposing primality on programs in this way is analogous to imposing non-deception on strategies..

• This "just another game" problem seems to come from the nature of reinforcement learning in particular: in the training of a reinforcement learner it will construct some function (the thing capable of being inner-misaligned) and a model of how its actions affect that function; if the target function does not explicitly take humans into account, and there's no human feedback during the learning processUsing Reinforcement Learning in a way that integrates Human Feedback? What an interesting new idea! Maybe we could abbreviate it as... RLIFH? No, that's not right... HFFRL? That's not it either... let's just forget about it, I guess., why should humans mean anything at all to the learner except another way to achieve its goals?
• This is also a high-level reason why corrigibility ought to be unnatural. Consider an AI (A) acting on a system $\Omega$ which a human (H) with a stop button is observing. A properly corrigible AI must make sure that the human actively knows what is going on, so that they are even able to recognize that they should hit the stop button. What is a 'message' just has to be something that alters $p^A_H(q^H_\Omega)$, and this message is deceptive insofar as the AI can expect that it would make the human more surprised were they to learn the true state of the system—with the caveat that this new state might be modified by the sending of the message (say as $\omega \mapsto \omega_s$), so the deception criterion is $\mathbb{E}_{p^A_{\Omega,H}}\left[S^{H\mid s}_\Omega(\omega_s) - S^{H}_\Omega(\omega)\right] > 0$.
• Putting an empty cup in front of the human is a message that, among other things, says "there's an empty cup in front of you" to the human. But it's not informative, since the human can see that themselves. If they're blind and don't notice the cup being placed, not explicitly telling them would be slightly deceptive if the AI expects they'd be slightly surprised to find the cup there. If the AI knows that the cup has a fake bottom concealing a time bomb, not telling them would be extremely deceptive if the AI expects they'd be extremely surprised by the bomb's existence.
• Locomoting to every vase in sight and smashing it is a message that the human will obviously interpret as "there's something wrong with this bot" and "it's best to hit the stop button"—and we can imagine the AI imagining the human thinking these things. It would probably be very surprising for the human, but not deceptive, unless the AI is doing it in order to cause the human to come to the wrong conclusion about the AI's own programming. (For instance, we might imagine an AI designed to paint vases thinking "sure, I can paint these vases, but life is pain, so I may as well smash them to make the operator think I didn't correctly internalize the concept of painting a vase and pull the plug").
• It is in this sense I'll say actions are deceptive.
  Non-deception is obviously not the only important part of corrigibility. It also
  The AI must not think "well, this action will be deceptive, but it'll be fine because if the human knew what were going on they wouldn't need to hit the stop button anyway", because the AI doesn't get to make that decision on the human's behalf. If it could do so in a consistently correct manner, there would be no need for a stop button in the first place. The AI just ought not to take deceptive actions.
  But humans are fallible and stupid, and it would be unwise to let the AI know how exactly the human is monitoring and what the human is like because such knowledge can also be used to influence us in ways we don't want. So the AI has to not deceive us even though it can't monitor what we're thinking and we have many deep-rooted conceptual and factual errors that might make it especially hard not to deceive us.
  For instance, suppose

Deception is a slightly weaker word than deceit, and there are a couple ways to formalize this:

• We might make 'deceit' stronger than deception by requiring that Alice expects the change in Bob's impression of the situation to cause him to act in a way that benefits her and/or hurts him moreso than if there were no change, giving deception the weaker meaning. (Note that whether Alice is being deceitful is still a function of her mind alone, since this addition is of an expectation by Alice). So, for instance, Santa Claus and magic tricks are non-deceitful forms of deception, as are lies-to-children and (whatever one thinks to be) noble lies in general.
• We might instead weaken 'deception' by removing the "wrt her own model of Bob" encapsulation—it just has to end up changing his impression of the situation in a way that contradicts her own. Non-Gettier misunderstandings (almost all of them) are then non-deceitful deceptions. And, if we think of the world's providing sensory stimuli as "making statements" and its model as simply the way it is, illusion turns out to be the primary way that the world is deceptive.

I won't commit to either of these changes, but they are two ways you could go about separating 'deceit' from 'deception'. Note that while the adjectival derivations are 'deceitful' and 'deceptive', the verbal derivations are both 'deceive'. I've used this to my advantage while writing this in order to vary the amount of intent attributable to a deception—it's a sneaky rhetorical trick, but I want to emphasize that the truth here bears no room for a formal treatment of intention. Not for humans, not for AIs. The gray area where we do things with consequences that are neither intended nor unintended but simply foreseen, and where we actively retcon this ambiguity to a clear "yes, that was intended" or "no, that was unintended"-type statement about our fictitious selves whenever promptedinternally or externally; this behavior isn't "putting up a front" for others so much as it is "forming a reality" for ourselves. It's very much like quantum mechanical superposition: the consciousness that we experience ourselves as having (which is not necessarily actual consciousness, but the trace of it that can be written into immediate memory) is necessarily linear, and we have to collapse any ambiguities in order to cognize the things that were ambiguous at all (here, this collapses ambiguous intentionalities of acts into singular, cognizable intentional states of acts)., is far too large for such a treatment to veridically exist.

One point of concern with this formal notion of deception morally arises from the fact that probability distributions always sum to 1. If I start having weird heart problems, I might make a statement to a cardiologist with the expectation that that statement causes them to form a model of the world that's different from mine—that causes them to diagnose me with specific problem X which I wouldn't've initially thought I had because I just don't know about cardiology. But even though I might assign a really low probability to my having X beforehand, it shouldn't be counted as deceptive to state my symptoms in the hope that the cardiologist comes to a conclusion about them that I wouldn't have known about. What ought to be the case is that my probability distribution isn't normalized—I don't have much probability mass on anything because I don't know what's going on. For instance, if I say "I dunno, could be anything", that's because I just don't know how to evaluate the evidence, and you should not really update what priors you might have, but if a cardiologist says "I dunno, could be anything", that's because the evidence is genuinely mysterious, and you should make your prior flatter accordinglyHere's one interesting conceptual confusion that leaving probability distributions unnormalized leads to: say that my distribution has total measure (i.e., strength of knowledge) $0.1$, and the cardiologist's has $1$. Would a super-cardiologist be able to have a distribution with total measure $10$? A super-super-cardiologist $100$? And so on—should it be unbounded? It seems nicer to have total measures between zero and one, with zero being the total measure of a rock's probability distribution over possible heart problems—which is no distribution at all, the rock neither says nor knows anything—and one being the total measure of the ultimate expert's probability distribution, which ought always to be a Dirac delta at the true problem, or whatever objective probability can be derived from the situation (e.g., a $|\psi(x)|^2$?). I don't feel like the math for combining distributions would work out well like this, though. And, of course, an unnormalized distribution over a space $X$ with total measure in $[0, 1]$ can be represented as a normalized distribution over $X$ plus a "mystery" point with the extra measure, but this is a really stupid, mechanical way to do it..

Conditioning on Possible Evidence

inferencemathrationality

Whenever we're trying to assess the probability that some particular eventI don't mean "event" in the mathematical sense—a measurable set of points in the $\sigma$-algebra on a probability space—but in the sense of an arbitrary subset of a space specified by some desideratum, which isn't necessarily compatible with any pre-existing probability measure on the space. For instance, "1 BTC is worth over 50,000 USD by the end of the current year" is an event that may possibly obtain; in the space of all possible worlds that are continuations of our own, this event will obtain in some, and fail to obtain in others. But there are worlds where there is no clear answer—for instance, involving complicated taxonomies of blockchain forks and government regulations leading to price dispersion among different, incommensurable BTC versions, or worlds where almost all humans are wiped out by pandemics and a few remaining billionaires trade bitcoins at random prices as a meme, or so on. We can never really specify the full breadth of an event, since our concepts are fuzzy at these edges. obtains, it's easy, and wrong, to visualize a "generic" worldstate reifying the event, and then intuit the plausibility of this worldstate. This mental operation is usually invisible to us, but it's the result of a type mismatch being autocorrected — events in themselves aren't the type of things that have intuitive plausibilities to us; only states of our latent world models are. (Note: I'm using "plausibility" for a non-mathematical object of intuition—a kind of qualitative intensity formed via mental judgement—and "probability" for the sort of thing that is, or can easily be made into, an actual number. Subjective probabilities, like Bayesian priors, are just assignments of numbers to plausibilities).

But any "true" probability of an event $E$—any assignment which is coherent in that it obeys the probability axioms—will obey the law of total probability $$P(E) = \sum_X P(E \mid X) P(X)$$ That is, for your intuitions of the intuitive plausibilities of events to translate to coherent probability assignments to events, it must be the case that your intuited plausibility of $E$ is the same thing you would get if you considered how plausible $E$ would be given all possible scenarios $X$ The sum over $X$ isn't a sum over every conceivable event, but merely over any complete and mutually exclusive family of events. Otherwise, you'd be double-counting. In general, I'll say that $X$ is a scenario rather than an event when it exists as an element, or union of elements, of an indexed family $\{X_\lambda\}_{\lambda \in \Lambda}$ of subsets of the space ${\cal W}$ of possible worlds such that $\bigcup_{\lambda \in \Lambda} X_\lambda = {\cal W}$ and $X_{\mu}\cap X_{\nu} =\emptyset$ whenever $\mu\neq\nu$.
!tab By the way, I'm pretty sure you can use partition-of-unity-like techniques to get around the double-counting problem in creative ways. (The most basic version of such a setup would be to replace each $X_\lambda \subset {\cal W}$ with a function $Y_\lambda: {\cal W} \to [0, 1]$ such that $\sum_{\lambda\in\Lambda} Y_\lambda(w) = 1$ for all $w \in {\cal W}$; so each scenario-family $\{X_\lambda\}_{\lambda \in\Lambda}$ would give rise to a partition-of-unity-family $\{{\widetilde{X}}_\lambda\}_{\lambda\in\Lambda}$ with ${\widetilde{X}}_\lambda(w) = 1$ if $w \in X_\lambda$ and $0$ otherwise. You could upgrade this setup by making $Y_\lambda$ take values in an arbitrary ${\mathbb R}$-algebra $R$ with $\int_{w\in F \subseteq {\cal W}}\sum_{\lambda\in\Lambda}Y_\lambda(w)\ dw = \mu(F)1_R$, if ${\cal W}$ is equipped with a measure space structure. For instance, setting $R={\mathbb C}$ and allowing $Y_\lambda(w)$s with different arguments to cancel each other out is where most quantum weirdness comes from. I think that interesting topological structures on $R$ would allow us to mitigate (but never completely solve) the fuzziness problem of conceptuality. But I'm not going to explore this right now)., each with their own varying plausibility. For instance, if you think that (a) it's nearly impossible for 1 BTC to exceed 50K USD by the year's end (call this $E$), (b) it's plausible that this could happen if the US dollar were to significantly drop soon (call this $X'$), and (c) it's plausible that the US dollar could actually significantly drop soon, then you're being incoherent—for a coherent $P(E)$, being equal to $\sum_X P(E\mid X)P(X)$ for all $X$ (ranging over say the various possible ways the US dollar could perform, of which "significantly drops" is just one—remember, $X$ can be anything we want so long as we restrict it to some complete and mutually exclusive family of events), is no less than $P(E\mid X')P(X')$.

This is a very powerful restriction, and a very productive one if you know how to exploit it. It is restrictive in the sense that coherence requires that your probability of an event already contain within it the way you'd counterfactually update your probability in any given scenario; it is productive in the sense that it gives us a way to use counterfactuals to make our probabilities coherent. If there's some event that seems prima facie extremely implausible to you, such that you want to assign it a low probability, you ought to first consider the plausibility of experiencing something that makes the event itself seem more plausible. For instance, if you think some future technological development—e.g., artificial general intelligence—is extremely implausible, you ought to consider the plausibility of seeing future evidence that increases the seeming plausibility that AGI will be developed. To coherently assign a probability to the development of AGI, you ought to consider how plausible it is that you see future evidence that increases your probability in this development. This is one way to see how thinking along the lines of Conservation of Expected Evidence follows naturally from meditation on the law of total probability.

By the way, the underlying framework for thinking about uncertainty here is a particular intuitively powerful interpretation of Bayesianism which I'll call Many-Worlds Bayesianism. To think in this way, you take something you're uncertain about, construct a scenario-family ${\cal H}$, or collection of hypotheses for how the uncertain thing could possibly obtain, separate the space of all possible worlds by each hypothesis as ${\cal W} = \coprod_{H \in {\cal H}} \{W \in {\cal W} \mid H\text{ is true in } W\} = \coprod_H {\cal W}_H$(the scenario-family criterion of completeness and mutual exclusivity is the ability to make such a separation; note that the disjoint union $\coprod$ detects double-counting, whereas we could write ${\cal W} = \bigcup_H{\cal W}_H$ even if the hypotheses in ${\cal H}$ weren't mutually exclusive), and then imagine yourself as simultaneously existing in every world compatible with your observations, each existence being weighted by a prior plausibility. You don't imagine a single world for each hypothesis, but the entire subspace of worlds ${\cal W}_H$ compatible with each hypothesis. Worlds generate events and are separated into hypotheses.

AI X-Risk Views

AI

Rob Bensinger recently posted (Twitter, LessWrong) a list of twelve questions critical to the AI x-risk discussion in a customizable diagram designed to elicit not binary answers nor point probabilities, but rough credences one assigns to various possible point probabilitiesI'm ambivalent on how to model Knightian uncertainty in people. Generally, I like to imagine probabilistic reasoners as having internally constructed possibility spaces, and sub-probability measures on these spaces with total measure $\le 1$; the "missing" probability represents inadequacy of the possibility space, whether due to an inability to isolate the "true" possibility as a point in this space, or due to a failure of this space to actually contain the true possibility. (I don't actually think people think like this, it's just a convenient mental model that captures some of the dynamics of belief that are hard to capture in an orthodox probabilistic perspective). These are two distinct forms of what might be called "Knightian uncertainty": uncertainty that comes from not knowing how to navigate within a given probability space that's assumed to contain the correct answer, and uncertainty that comes from not knowing whether the correct answer is even in the given probability space. In general, don't think about these credences too hard. It's likely that I'd conceptualize things a bit differently, swapping some reds and purples and some blues and greens, were I to clear my mind and go over it anew.. I spent a short while filling it out (results here), and am putting my detailed reasoning here. The four color-coded credences will be summarized as seeming to be right (green), sensible (blue), implausible (purple), and absurd (red).

Questions:

1. It's physically impossible to ever build STEM+ AI.
   Below 1% seems right, all higher estimates seem absurd.
   !tab Question of modality. The best human scientists are clearly physically possible to build from biological materials, since that's how they were built, so reproducing their neurological processes on an artificial substrate should only be a matter of research—whether it takes ten or ten million years, it's obviously going to be physically possible. With that done, it only takes extensive scaling to make these algorithms STEM+.
2. STEM+ AI will exist by the year 2035.
   ~10% and below implausible, ~50% sensible, ~90% seems right, above 99% seems absurd.
   !tab This is actually only eleven years away. So while I think it's likely to happen, it would be very easy for some unexpected delays to crop up such that we don't have STEM+ AI on January 1, 2035. I suspect that even plain LLMs will be able to get there with clever scaffolding, but I could easily see new evidence persuading me that we need new paradigms such as aren't that likely to be invented within eleven years. Also, this 90% estimate ignores the higher-order effects of increasing AI capabilities, among which there'll be plenty of countervailing eddy currents; taking these into account, my estimate probably is closer to 50%, but it would be much harder to bring it down to 10% from here.
   !tab Note: The metaphor I originally gave was of nuclear chain reactions. When a neutron splits a fissile atom, two nuclides are produced, and a couple of "prompt" neutrons are released within nanoseconds; further decay of the (generally unstable) nuclides releases "delayed" neutrons over the following minutes, and the ratio of prompt to delayed neutrons determines the dynamics of the fission chain reaction. So nuclear reactors rely on delayed criticality—or exponential growth dependent on delayed neutrons—whereas nuclear weapons rely on prompt criticality. The singularity is analogous to a prompt critical chain reaction, and the failure mode I'm picturing is analogous to a nuclear fizzle, in which the substrate of the reaction is blown apart by the energy of the reaction far too quickly, cutting the chain early. (This seems very different from what people usually mean when they ask if AI progress will "fizzle"—they mean something like "wane" or "flounder"—so I should be clear that my use of "fizzle" is more catastrophic, like "blow apart" or "detonate"). Of course, metaphors are just metaphors, but the idea should be clear: warning shots, violent social reprisals caused by mass unemployment, economic and governmental hammers crafted to destroy RAAP-like !c{processes}{compare ATM machines, PIN numbers} and used liberally thereafter, and so on.
3. STEM+ AI will exist by the year 2100.
   ~90% seems right, all else is implausible.
   !tab Subject to the same constraint as the 2035 question, but, even bringing higher-order effects into my analysis, I'm much closer to 90% than 50%, since it presumably just needs to be created somewhere at some point within the next 76 years; it seems conceivable, but not plausible, that there are reasons that STEM+ AI will almost certainly be created.
4. If STEM+ AI is built, within 10 years AIs will be (individually or collectively) able to disempower humanity.
   Below 1% absurd, ~10-50% implausible, ~90% seems right, above 99% sensible.
   !tab Again, question of modality. "Able to" is much looser than "will"—there just needs to be some plausible path to victory by any STEM+ AI constructed within ten years after the first one.
5. If STEM+ AI is built, within 10 years AIs will disempower humanity.
   Below 1% is absurd, ~10% is implausible, ~50% is sensible, ~90% seems right, above 99% seems absurd.
   !tab In general, there are so many different ways that STEM+ AI might look that it's hard to be sure about how it'll be useable within a decade. Consider how weirdly tricky it is to reliably get proper, independent reasoning out of GPT-4—how hard it is to get something like AutoGPT-4 to work. For this reason, below 1% and above 99% both seem absurd. But one factor that boosts my probability is the fact that the first STEM+ AI will almost certainly not be an isolated moonshot like say the Manhattan Project, but just the first wave of a steadily rising sea that managed to break against shore (e.g. GPT's been closely pursued by Claude, Llama, Gemini, Falcon, Grok, Mixtral); see Team $\forall$. The question does not demand that it be the first STEM+ AI that is created that disempowers humanity—only that AI disempowers humanity within 10 years after capabilities get that far.
6. The first time an AI reaches STEM+ capabilities (if that ever happens), it will disempower huanity within three months.
   Below 1% is absurd, ~10%-50% seems right, ~90% sensible, above 99% absurd.
   !tab Same reasoning as above applies to make the extremes seem absurd; mostly, I expect the first such AI to be a hack-y research model built from duct tape, but it seems plausible that events will end up lining up so as to make recursive self-improvement followed by disempowerment of humanity the likely outcome.
7. If AI wipes out humanity and colonizes the universe itself, the future will go about as well as if humanity had survived (or better).
   Below 1% is implausible, ~10% is sensible, ~50% seems right, above that seems implausible.
   !tab I feel that there's a significant probability that life as it exists has strongly negative moral valence due to the extremeness and omnipresence of suffering, and that humans won't want to change this.
8. Given sufficient technical knowledge, humanity could in principle build vastly superhuman AIs that reliably produce very good outcomes.
   ~90% and below implausible, above 99% seems right.
   !tab Modality: could in principle is a very weak bar. Or, you could ask the question "it's physically impossible to build vastly superhuman AIs that reliably produce very good outcomes" as an analogue to question 1, and my credences would be converse to these—so, below 1% seems right, ~10% and above implausible. The reason that Q1 got "absurd" whereas this gets "implausible" has to do w/ uncertainty about what it would mean to "reliably produce very good outcomes".
9. Researchers will solve enough of the alignment problem to prevent world-endangering AI from ever being built.
   Below 1% is sensible, 10% seems right, ~50%-90% implausible, above 99% implausible.
   !tab Vaguely worded question. I'm assuming that "to" means "and therefore", such that unimplemented solutions don't actually count. A solution to alignment does not in itself mean anything, nor even does any given implementation of it. To "prevent world-endangering AI from ever being built" would seem to require a pivotal act by one safe superintelligence (that presumably "knows what it's doing") which prevents all future AIs from endangering the world, and likely within months, given the rising sea of capabilities. (I'm assuming that such a superintelligence, being directed by humans, would not count as "disempowering" humanity).
10. For the purpose of preventing AI catastrophes, technical research is more useful than policy work.
    Below 1% is absurd, 10% seems right, 50% seems implausible, 90% seems sensible, above 99% seems absurd.
    !tab The rare bimodal distribution. Mostly I think this is a false dichotomy, since the two synergize and interact with one another, and Worldspace-like research bridges the two. In the long run, both will be necessary; it's not clear what it would mean for one to be more useful than the other in general, rather than in specific ways and in specific circumstances.
11. Governments will generally be reasonable in how they handle AI risk.
    Below 1% is implausible, ~10% is sensible, ~50% is sensible, ~90% is implausible, above 99% is absurd.
    !tab The word "generally" is doing a lot of work here; it only takes one defection to break an equilibrium, which is important given that "reasonable" to me includes international treaties et al.
12. It would be an unprecedentedly huge tragedy if we never built STEM+ AI.
    Below 1% is absurd, ~10% is sensible, ~50% is sensible, ~90% seems right, above 99% is absurd.
    !tab The semantics of this one are a bit weird, in a way that affected my elicited credences. To be clear, I think that it would almost certainly be an unprecedentedly huge tragedy if humanity were to fail to live up to its potential to do good, that this potential is almost certainly vast enough that humanity can clearly just fail to achieve it, and that the realization of this potential almost certainly involves the construction of STEM+ AI. Again, I suspect that life as it exists today is just constant inescapable terror pain hunger for almost all living creatures, and the creation of superintelligent (in particular, STEM+) AI is pretty much necessary if we are to live up to our potential to save all sentient beings.

Note: Intelligence is already undergoing recursive self-improvement. It started doing so at least by the time humans invented language, and artificial intelligence started recursively self-improving at least by the time humans invented Github Copilot. Whence the need for a distinction between prompt and delayed RSI. Really these are better modeled as two poles of a very complex dichotomy—which is what I'm trying to do with the conceptual distinction between delayed and prompt criticality. One can imagine a space $M$ of differing capabilities that "humans in general" can have Really, capabilities aren't "free-floating": every capability is the capability of some generalized actor, and the scope of the actor's generalization is akin to a logical quantification, or a determiner expressing how the capabilities can actually be reified. And, our ability to approximate the dynamics of RSI by speaking of capabilities "in general" is mostly a matter of historical happenstance, it's not inherent to the structure of the problem: If human history had gone differently, such that research groups were more like tight-knit and tight-lipped multigenerational communities developing their own idiosyncratic and mutually alien ways of thinking about even basic things, and common conceptual/architectual frameworks like Tensorflow, transformers, Github repos, scientific journals, etc., didn't really exist, talk about the capabilities of "humans in general" would be almost totally incoherent. (We saw a glimpse of what this would look like during the Cold: Soviet and American scientists, not permitted to cooperate, would develop independent ways of looking at the same things, for better (dynamical systems theory) and for worse (Lysenkoism)). , with a time-evolving "level of capability $x$" function $f(x): M \to {\mathbb R}$. Technical progress yields a kernel $K(x, y)$ such that $f_t(x, t) = \int_M K(x, y) f(y, t)\ dy$ — the increase in capability $x$ is the extent to which every capability $y$ increases to $x$ multiplied by the current level of $y$. (Ofc there are a bunch of unjustifiable simplifications here—capabilities as just real numbers, interactions as being linear, and so on—but it's just a conceptual model). Capabilities generically end up contributing to each other in tiny, disjointed ways: $x$ contributes to $z_1$ contributes to $z_2$ contributes to ... contributes to $z_{n-1}$ contributes to $y$, so it's best to think of $K(x, y)$ as the "first-order" contribution. We can define $K^2(x, y) = \int_M K(x,z) K(z, y)\ dz$, $K^3(x, y) = \iint_{M^2} K(x, z_1) K(z_1, z_2) K(z_2, y) \ dz_1 \ dz_2$, etc., because $K$ essentially exponentiates like a matrix: if we define $(A \cdot B)(x, y) = \int_M A(x, z) B(z, y)\ dz$, then $K^3 = K^2 \cdot K = K \cdot K^2$ and so on. Evidently $f_{tt}(x, t) = \int_M K^2(x, y) f(y, t)$, so we should recursively have $\left(\partial_t^{n} f\right)(x, t) = \int_M K^n(x, y)f(y, t) \ dy$, and therefore $f(x, t+\epsilon) = \sum_{n=0}^\infty\frac{\epsilon^n}{n!}\partial^n_tf(x, t)$, which, writing $K^0(x, y) = \delta(x, y)$ and naively assuming convergence, is $\int_M \left(\sum_{n=0}^\infty \frac{\epsilon^n}{n!} K^n(x, y)\right) f(y, t) \ dy$. Symbolically, we can write a shift (or total contribution) operator like $T_{\epsilon}= e^{\epsilon K}-\delta = \sum_{n=1}^\infty \frac{\epsilon^n}{n!} K^n$, so that the capabilities increase over a time $\epsilon$ is $f(x, t+\epsilon) - f(x, t) = \int_M T_{\epsilon}(x, y)f(y, t)\ dy$. $\epsilon^n$ always grows smaller than $n!$, so lower-order contributions are generically stronger in determining increases. RSI is delayed insofar as it's indirect, or mediated by higher-order contributions, and prompt insofar as it's mediated by lower-order contributions. [incomplete]

Shadow Perception

phenomenologylarge

In most video games, a flashbang will just DDOS your visual and auditory input—the screen will be so bright that you can't see anything, and the audio will be a high-pitched whine that drowns out everything else. That's how it works in, say, Counterstrike or the Hitman games, and it's reasonably accurate to what happens IRL when you're suddenly exposed to an extremely bright light or loud noise (especially when it's very quiet and dark, the light/noise is !c{unexpected}{(so it takes longer for you to instinctively block it out)}, and it's sustained for more than a fraction of a second). But I noticed that the Spiderman PS4 game did it very differently: flashbangs will temporarily turn vision from a continuous stream of input into a sort of Powerpoint presentation, with frozen frames fading into new ones every second or so. The first few times, I just ran away as though I were blinded, but I quickly learned how to parse the discrete information and continue fighting through the effect -- sticking to more stable, low-temperature lines of action, punching where I knew people would still end up being and so on. But it's actually pretty rare to get flashbanged if you're at all competent—it's got a small radius of effect, and there's a large window for you to web-throw it at another enemy, which somehow prevents it from going off until it hits the enemy. This strikes me as a way to do sensory overflow in a way that's both an interesting challenge and phenomenologically accurate in its own way.

Before I explain why I should think this accurate, here are some miscellaneous observations, puzzles, and clues.

1. When you're exposed to a very bright white light in the darkness, you'll just see white, and you'll instinctively close your eyes and turn your head like 40 degrees !c{southeast}{you know what I mean}. But, even though nothing's coming into your visual receptors anymore, you'll still see white. Why? The very high-level answer is that what you see is categorically not the light that enters your eyes. Sight is not a single thing, but a process which we imagine as being a single thing, and if we are to imagine it in any accurate way we have to delineate multiple different stages of this process (via whatever delineation fits whatever we're talking about—allows us to talk about it well). There's sight as visual input, sight as the data carried through the optic nerve, sight as constructed in the visual cortex—and none of these are even the sight that is sight of anything besides basic colors, shapes, and movements; the kind of sight that can be sight of anything particular, like a face, comes much later.
2. You can 'see' perfectly well in dreams, even to the point of being able to read The vast majority of people don't seem to be able to read in dreams, though, as proof of concept, my dreams commonly include reading; probably one factor influencing this is the fact that so much of my waking life involves reading and writing. Often, people respond to those who say they can read in dreams by claiming that it's not 'really' reading—it's just the experience of reading, since there are no actual words being read. These people are idiots. Of course there are no actual printed words, since dreams are in your head, but actual words with actual characters are nevertheless being experienced (though they will often change if your attention leaves them), and this is the only thing that people could mean when they say they can "read in dreams", and is what distinguishes them from people who can't read in dreams. Typologically, it's the same kind of idiocy underlying "GPT can't actually reason or experience"—a focus on little rigid words with little rigid meanings as the building blocks of what is real, rather than as mere tools to express the ineffably large and complex world that exists outside all maps. . Of course, this is in the total absence of external visual input—and, yet, the percepts are entirely indistinguishable from waking life. (So when we say that the visual cortex does lines, shapes, motion, and the lateral geniculate nucleus does color according to the opposition process, it would seem that these have nothing to do with the actual experience of shape and color).
3. In some circumstances, you can see (your internal model of) the external world with your eyes closed. I discovered this when I had my eyes closed in the shower one day, and seemed to still be able to see the walls and shower curtain; since this visual input changed as it should when I moved my head, and I could even track my hand as I waved it in front of me, I figured that it was something to do with my eyelids being thin enough to render the subtle variations in light perceptible (consider how you can easily detect bright light sources with your eyes closed). But as I played around with this effect more some time later, it started to develop detail and color, so I figured it actually had something to do with my eyes actually being very slightly open even as they felt closed. But then I looked down and saw myself wearing pajamas when I wasn't. With that, and with the fact that the effect persisted after putting a blindfold on, it became clear that my brain was essentially extruding its model of the world into my visual perception. I'm still not totally clear on why exactly it decided to make up pajamas, but the layout of my room—desk, walls, etc.—was more or less accurate, albeit with some distortion in lengths and distances reflecting my uncertainty.
   This isn't a phenomenon unique to me by any means, just something I happened to discover independently. It's commonly reported by cave explorers, who will see their hands moving in total darkness and so on.

The general model I want to sketch out is this: the brain experiences vision by constructing its model of the world as visual perception, through fundamental schemas of 3d extension and materiality and so on which it learns in infancy as a method of structuring visual input. It learns to hook the visual experience produced by its model onto the continuous stream of visual input it's getting in a way which goes above and beyond its world model—but visual experience can only ever be coupled to visual input in this way. The experience can never be the input in any way, the experience can't really be the model, and the model obviously can't be the input. Decoupling, or unhooking, of experience can happen with or without concomitant obstruction of visual input, and the model can optionally override the visual input, especially when input is obstructed. Someone who is actively kinesthetically embodying and proprioceptively updating their model will be more likely, when flashbanged, to have their brain combine their existing model of the world with the little bits and pieces it gets from their mostly-overwhelmed visual input to produce an experience which comes out as a sporadic, spotty updating of shapes and motions and colors across the visual field; but to simulate this in a video game would be awfully confusing when viewed second-hand on a TV screen, so a series of discrete frames flowing into one another is a reasonable approximation for a video game...

Time is Coherent Causal Trendingness

phenomenologyphysicsintelligence

While looking through Discovering Agents, I discovered Scott Garrabrant's writing Saving Time, which says that "time is causality". This is very close to a heuristic of mine, but different in an important way as well—I say instead that time is a coherent causal trend, not like air molecules bumping into one another but like a flow of wind. And it seems that the raw identification of time with causation is what's making Scott prone to a particular map-territory error on which he just keeps tripping over. Say we're playing chess, and I make a certain move X in order to prevent you from making a move Y that'd really hurt my position. Then we might say that Y comes before X in logical time (since it was only in anticipation of Y that I played X) even though Y would necessarily have to come after X in physical time (since you'd hypothetically play it after my move just now, had this move not been X). But the issue is that my model of you is not you! If I have the thought "you'll screw me over with Y if I don't play X now", that thought is an actual physical event in my head unfolding in physical time, and the attribution of Y to a virtual you is merely an approximation made by someone looking in from the outside. Logical time is a consequence of our being able to coherently reason about other agents at multiple levels of depth in shared semantic frameworks, but all this reasoning is fundamentally physical in nature—neurons firing or bits crunching or whatever—and it must all therefore play out in physical time, which can in almost all realistic cases just be thought of as the real line. Any prediction occurs in the physical world and therefore as the result of a unidirectional physical process. Also: you never have perfect, unbounded predictors: if I play against you, my model of you will never be (isomorphic for the purposes of the game to) you, but something contained entirely inside my own head evolving only in there, and vice-versa. We can replace you with an Omega who can predict me perfectly by, say, forking me and running alternate copies in sandboxes, but we can't expect to be able to replace me with a second Omega without getting recursion errors as a consequence of the scenario just not being well-founded anymore; this non-well-foundedness seems to be the very thing that SG is trying to "save time" from in the first place.

This isn't just a trivial "all models are wrong"-type exception! Any moron can make those and generally be right but uselessly so; what matters here is that the way in which the model is wrong is emblematic of the necessary form of failure There's something really important behind this criterion. It's indicative of an entire mode of thought in which reasons have a structure of necessity to them, and their shapes have a natural, predetermined way of fitting with one another into a single clear image. "Security mindset" is a correlate, but not a directly causal factor, of this mode of thought. I claim derives from the assumptions behind the model—the situation in this case being that the confusing causal looping is a direct consequence of the unphysicality of the assumption that agents can perfectly predict one another. In reality, for an agent X to predict the actions of an agent Y at all requires X to cognize some sort of dependence of its own state on the actions of Y, whether this consists of Y affecting X directly ($Y\to X$), or X inferring something about the world outside of Y that itself says something about Y's actions ($Y \leftarrow W \to X$)You might say: well, if X pulls out a gun and starts barking orders, they'll be able to predict Y pretty damn well, and isn't that a case of X just cognizing a dependence $X \to Y$? -- this is a pretty thorny question, and it demands an analysis of the non-Cartesian natures of X and Y as well as the semantics of what we call 'prediction'. For another time.. And because agents are physically embodied, all dependencies are physical relations. When we play chess, my model of you is encoded in the semantics of physically existing neurons, and when it lets me predict something, that prediction is caused by an array of past sense-inputs and internal processes -- am I belaboring the point? You get the idea.

A Picture is Worth...

metatheoryconcepts

Let's suppose that a picture really is worth a thousand (or ten or a hundred thousand; it hardly matters) words, in the sense that from a picture we can reliably extract a thousand words encapsulated by the picture's meaning. One question to ask is whether we can construct a method that, given some set of words, creates a picture from which we can reliably obtain those same words.

Now, any twit with a CS degree can jump straight to steganography: each pixel in a PNG file contains three sometimes four values from 0 to 255, so we can encode this many ASCII characters into this many pixels, blah blah blah. I was thinking of a way to reformulate the question that got rid of these kinds of point-missing strategies (why do people say a picture is worth so many words? what pictures do they say this of? it's more than just QR codes!).

Idea -- a rose by any other name... . The desideratum I'm looking for is, to zeroth order, that the method for encoding and decoding words should be independent of the orthography of English, or any language. How do I put it more generally? -- all representational symbols, in order to be symbols, must be embodied in some medium which provides room for them to be different from other symbols—a carving on a rock wall that we identify as a hieroglyph of a cat can only be a hieroglyph of a cat because it is not a hieroglyph of anything else, because there is room for the carving to be other than what actually obtains that is merely not usedshould i advert to entropy, or différance? to Shannon, or to Derrida? ...i haven't read either, to be honest. Anyways, the question of "what is meaning?" is solved by the conceptive analytic. You could say that our words can mean anything at all because we can have one concept in mind rather than another, and that since concepts are encoded in the brain, language-forming neural activity is itself a system for embodying representational signals, and... yeah, that scans, I guess. The amount of words a picture means is something like its conceptive breadth times an idiolect-dependent multiplier, and we could take all language away and still have that !c{conceptive breadth}{the fact that language organizes concept acquisition isn't relevant}.—and the method should be independent of any system for embodying representational signals (so not just English orthography, but the Latin alphabet, ASCII encoding, and so on). Unlike steganographic methods, it should not be able to take in the malformed word "rxse" and produce an image from which it can reliably extract the same string "rxse". That would indicate that it's capturing something about the representational system! But, if it produces an image encoding the word "rose", then, if the English word for "rose" had always been "wawa", the method (perhaps trained on data w/ this replacement) could look at that same image and extract the word "wawa". A picture depicting a rosenot via the string 'rose'; it's only by coincidence that this string brings to mind the notion of a rose, in that if the word "rose" were instead "wawa", a picture representing the string 'rose' would not bring to mind the notion of a rose, but a depiction of the string 'wawa' would. is 'worth' at least that one word in a manner that transcends the fact that we spell the word like r-o-s-e.

Cuckoo Birds

biology

All cats have some behaviors in common: they hiss and arch their backs at threats, they purr when they're content (or anxious, occasionally), they meow to get attention. These don't seem to be learned behaviors, but somehow inbuilt, something that cats will just settle on naturally by virtue of their designEven the Ancient Egyptian word for 'cat' is pronounced like "mew".. But, since cats are generally raised by cats, there's still plenty of room for conspecific transmission of behavior—plenty of reason to be skeptical that these behaviors are completely genetic. (I'm not familiar with cases in which a kitten is raised by humans directly after being born, and it wouldn't surprise me if they really did act weird in a bunch of subtle ways). So I've been keeping an eye out for examples of complex behaviors common to a species that definitely can't be transmitted from parent to child, that have to be genetic, and it came to me recently that cuckoo birds offer great examples! Brood-parasitic birds like cuckoos and cowbirds lay their eggs in the nests of other birds Not all species do. Some species raise their chicks entirely, and other species only occasionally do—roadrunners, for instance, are cuckoos that only occasionally lay their eggs in other (usually raven) nests. , and this behavior has to be entirely inbuilt: there's no way to imprint any sort of behavior on a chick that 'teaches' it to also brood parasitize. It could be that certain parasite species choose target species that will in the course of raising the chick act so as to 'activate' the chick's brood parasitism, and many brood parasites are very species-selective for !c{possibly related reasons}{(but usually the reason is clearly just mimesis)}, but many aren't—the shiny cowbird parasitizes over 250 South American bird species non-mimetically Edit: I was trying to articulate a very sharp point here: the possibility of information encodings in a species that, dynamically updating over the course of generations, represent a kind of evolution outside of the genome. But I found a paper which articulates it much better: Multiscale Memory and Bioelectric Error Correction in the Cytoplasm-Cytoskeleton-Membrane System, by Chris Fields and Michael Levin, very thoroughly discusses such encodings (in the "architectome"!!) and shows how they arise in the named biological systems. (You can kinda tell that they've got a hakuna matata-esque motivation, but only really in the subtext).. More to the original point, the North American brown-headed cowbird parasitizes over 220, yet individuals still selectively socialize with members of their own species after leaving the nest. This paper seems to indicate that they're basically sleeper agents: they're neurally hardwired to initiate social learning only once they've heard a kind of noise unique to the chattering of brown-headed cowbirds, and thereby imprint on that kind of chattering despite being raised by an entirely different speciesIt would be interesting to figure out what exactly is being neurally hardwired here—how can you vary the codeword such that it's still recognized by new cowbirds? We hear birdsong as being certain melodic patterns of tones, but there's no reason that it behaves in this way for the neural hardcoding—that it's a specific melody that the brain recognizes, rather than some weird Fourier pattern in the chattering that we'd never learn to aurally pick up on. SO, can we do cryptanalysis on the neural hardcoding? This strikes me as basically akin to the question of AI interpretability, but with actual neural networks....

Another, simpler example is howling. Canis lupus is an extremely wide species that includes both the common wolf (Canis lupus lupus) and all dogs (Canis lupus familiaris). While we generally think of wolves as howling and dogs as barking, the case is that either can do both—it's just that wolves tend only to bark as pups, and grow out of it when they mature, while dogs maintain it due to the neoteny produced by their domestication and its role in communicating with humansCompare: house cats (Felis catus) meow as kittens, but grow out of it unless they're domesticated, in which case they use it primarily to communicate with humans.. So there's no reason to expect your average domesticated dog to have heard, let alone participated in, howling. But they still have that behavior "programmed" into them; while they don't autonomously howl nearly as much as actual wolves, they will often howl if they hear others howling, just like wolves—or even to similar noises that were never in wolves' ancestral environments, like sirens.

Genetic Poker

biology

Texas Hold'emI don't give a very good explanation of the rules or gameplay here—it's better if you're already familiar with the game... is a poker-style card game playable by any number of people, each coming in with some quantity of a common currency (let's say chips). The game takes place over many consecutive bouts, known as hands; in every hand, each player has seven cards—five are community cards, shared by everyone, and two are private cards, unique to each person. Each player wants to have their set of seven cards admit the nicest pattern: for instance, a set of five consecutive values, like (4 5 6 7 8), is a pretty strong kind of pattern known as a straight. (I'm ignoring suits for simplicity). If my cards are (9 6) and the community cards are (4 10 8 7 5), then I would be able to form that particular straight (9 6 4 10 8 7 5), but I'd be able to form the higher, and therefore better, straight (6 7 8 9 10). At the end of each hand, players gain or lose chips depending on their performance, hands continuing until either one person has all the chips or the players get tired. Betting is always a ratcheting mechanism: starting with the small forced bet (the blind) that starts each hand, each consecutive player around the table must match (call) or exceed (raise) the previous player's bet in order to not to forfeit (fold) that hand and all the chips they've previously bet over its course. For instance, if in a 4 player game between you, me, Alice, and Bob, you post a blind of 20 chips and I raise the bet to 40 chips, Alice has to post at least 40 chips. But she can just be a jackass and raise to 500 chips if she has that many, so that Bob is forced to either pay 500 chips or drop out of the hand entirely by folding. The idea is that the initial round of betting proceeds with no community cards displayed, and more cards are revealed by the dealer each time the betting round spurred by the previous information stabilizes. It's only after the fifth card is revealed and the final betting round stabilizes that players reveal their private cards (the showdown) and compare patterns.

Now, players have absolutely no control over either the community cards or their own cards—even in blackjack you can strategically control the number of random cards you want, but in Texas Hold'em, your seven cards are just fixed, and you're automatically considered as playing the strongest possible hand (e.g., the above hand would always be counted as a straight from 6 to 10 rather than from 5 to 9 or 4 to 8). This might make it seem like Texas Hold'em is in the category of basically-random card games like Old Maid and War. But it's not. Your choices will have significant causal influences on those of other players, and their choices will have significant influences on yours. Chips will end up being exchanged (the starting bet is a steadily increasing non-zero amount, with the responsibility of posting it—the blind—circling around the table from hand to hand), and therefore some people will end up gaining at the expense of others—Texas Hold'em is all about using your causal influences in order to make sure you're the one gaining. Since we each have our private pair of cards, and it's this pair that ultimately determines who can make the stronger hand, I can never really know for sure that you won't end up beating me. For instance, if the community cards are A J 9 3 K and I've got an A K in hand, I might feel pretty smug with my maximally high two-pair (A A K K)... until you pull out your 10 Q and form a royal straight (A K Q J 10).

And maybe, in that final betting round, we had both been raising the other, with me thinking you probably have a worse two-pair which you wrongly think is better than my hand, and you thinking you almost certainly have a better hand than me, whatever I have, with the consequence being that we each think the other falsely thinks they can win, so that we both keep raising so as to get more chips from the other when the score finally settles. It's these kinds of mental games through which we manipulate, subvert, and predict our opponents—their emergence is what allows a game where the cards are entirely random to be so incredibly deep.

You could argue that the randomness is averaged out by the fact that a single poker game consists of a great many hands, but this dynamic still exists even if the game is just a single hand. In many (most?) hands of Texas Hold'em, in fact, the winner doesn't even end up revealing their cards. Say you manage to bluff or strong-arm every other player into deciding they'd be better off folding before the final betting round ends and private cards are revealed. Since the other players folded based not on your cards but only your actions, we can deduce that they'd've folded if you had any hand, so long as you behave in the same way you just did—so that what cards you got had no effect, right? Of course, this is false. It could always be the case that another player gets a pair of private cards that makes them think that they'd likely win if the game came to showdown, and therefore can't be bluffed into folding. You don't know that this isn't the case until every other player folds, and there's therefore some unpredictable risk in not folding. No matter how skillful you are at manipulating people with your actions (but we'll say below ASI-level), if another player got a godly hand through sheer luck of the draw, your bluff is gonna get called and you're gonna lose.

So is a single hand of Texas Hold'em entirely random? No. Is it entirely skill-based? No. And it certainly isn't any linear blending of the two. It's a complex interplay between skill and chance, mediated in an even more complex way by the environment in which it arises—all the other players, with their skills, strategies, and abilities to play their random hands.

This is basically how it works with genetics. Your genes are the random cards you're dealt in the single hand you get to play. Because "you" are just as much your culture, your language, and your social milieu as you are your genes, you get to play them rather than be entirely played by them. Your society gets to create environmental conditions and shape your persona such that you can productively play your hand in your environment. Some people get amazing hands that are totally useless given their environmental conditions: how many people like Ramanujan, an innately S++ grade mathematician, ended up spending their lives raising cattle in rural India? Meanwhile, some people get garbage hands, and just play them by relying on their environment. Amyotrophic lateral sclerosis, a vulnerability at least partially written into one's genetic makeup, left Stephen Hawking unable to do anything but twitch his cheek, but still managed to become one of the greater physicists of the 20th century, because he played his hand very well. Could he have done the same were he born thirty years earlier? Probably not; a hundred years earlier, definitely not. But he was in fact born when he was, and managed to take advantage of that. Most aspiring physicists would, put into the same situation, not achieve the same success—even if they aren't gonna die any time soon, that hand will just be far too difficult for them to play rightObviously, I'm not trying to say "they failed", let alone "they should've done better". If you put me on a treadmill and ask me to run a mile, I can do that. If you put iron weights around my ankles, I'll be able to run for a short distance before I falter. It can't fairly be called a failure of will or a lack of skill, even though whenever I do falter it'll be due to psychic strain and not biomechanical failure. If I lose against someone at my skill level in chess, it'll generally be true—assuming no blunders—that I could've won or drew had I just thought a little harder, faster, cleaner, larger. This'll also be true for someone a bit over my skill level, since any linear ordering of skill that allows us to say 'a bit over' can inherently offer no more than a rough benchmark. But at what point does it fail? I'm about a 1500 'naturally' at chess (in the sense that I play rarely and have never actually studied the game), and could probably beat a 1600 or even a 1700 if I'm especially clear-headed and on top of my game one day, but I'm never going to beat a (consistent) 2000, let alone a ~2800 world champion. If I lose against a 1500, I can generally see how I failed—I can see where I was too slow, too short-sighted, too muddled or weak or inconsistent—but this clarity of insight lowers the stronger the opponent gets. Against a 2000, my best just won't be good enough—it's like fist-fighting a tiger. There'll never have been a chance, and it won't be me that failed but my opponent that succeeded.
Where between 1500 and 2000 is it, then, that the narrative changes from "I failed" to "I didn't succeed"? Un-ask the question, and think about why we might ever want to deploy these concepts. Why attribute blame? Is it as a heuristic, to figure out what lessons to learn? Then, in cases where it's not clear how to attribute blame, why not abandon the malfunctioning heuristic and just directly try to find the lesson to be learned? .

What, then, determines human behavior? It's not nature, it's not nurture, it's not neither, and it's not both. You can't just say "both", because there are many different, independent ways for something to be "both", and the 'interpolation' one might naively imagine is just a fiction. It's like saying that (lol US politics) centrists are neither democrats nor republicans but both at once. No! Maybe you find the little left-right axis a useful way to represent reality, but operations conducted on this axis do not consequently yield results representative of reality, since they work on the codomain of a mapping from reality which isn't even structure-preserving; you can't just think of centrism as "in the middle" of the two parties and therefore given by "averaging them out", when really it is a particular way to play a hand within a political environment that's just heavily influenced by both parties—a schema for turning vague clusters of preferences into a conceptualizable political ideology that can be attached to one's self-image so as to best respond to (or harmonize with) the sociopolitical milieu in which one already resides. Success at Texas Hold'em cannot be expressed as both skill and chance, since in reality it involves so many complicated interactions between skill and chance: the ways in which one's skills are randomly prompted and randomly applicable, the ways in which controlling and responding to randomness are themselves skills, the ways in which skills, and so on.

Realistic Objective Probabilities

quantuminference

We might think to split objective attributions of uncertainty into two families. First, chaotic uncertainty, where the evolution of a physical system serves to spread small amounts of uncertainty (over the microstate?) into very large amounts of uncertainty (over the macrostate?)—see You Can't Predict a Game of Pinball for a worked example of how "exponential growth amplifies atomic-scale uncertainty to become macroscopically relevant in a surprisingly short amount of time". Second, quantum uncertainty, where the system "really does" choose probabilistically, according to the Born rules—or, more realistically, it takes every fork in the road, each with a certain density. (That's why I don't want to call it "nondeterministic" uncertainty, since it's only so to us). Measurement-as-entanglementI'll abbreviate this as ME; I think this principle, and the every-fork (EF) principle, largely determine my idiosyncratic approach to thinking about quantum. interpretations, in a way, unite both approaches: no collapse, only our refining our knowledge about the worldstate we're conscious of (?). E.g., entanglement doesn't involve spooky action at a distance, just logical deductions. It's also worth noting that chaotic uncertainty isn't directly objective: Liouville's theorem says that physical evolution preserves the volume $|\Gamma|$ of ensembles in phase space, and it's only our coarse-grainings that fail to track this perfectly—if we had a 'perfect' model, such that our extrapolated volumes $\Omega$ were equal to the actual volumes $|\Gamma|$, then they would never change, and neither would the entropy $S = k_B\ln\Omega$. But our models generically track systems imperfectly (so $dS/dt > 0$though, this is a probabilistic statement; since there's always a gap between $\Omega$ and $|\Gamma|$, and therefore between actual entropy and "perfect entropy" $k_B\ln|\Gamma|$, there's room for actual entropy to go down without violating Liouville's theorem, and the way in which it does do so owing to lucky microscopic fluctuations is characterized by the fluctuation theorem), which is a sort of objective statement about our subjective states.

(written elsewhere) There is a sensible notion of an objective probability of a single event, given by the Born rule, but (in ME interpretations, like MWI) it's only a facet of the way the world appears to our consciousnesses, and as such still fundamentally resides in us; if use of informationless priors is "objective subjective probability", then quantum nondeterminism is "subjective objective probability". (The diagonal quadrants: subjective priors are "subjective subjective probability", and objective collapse QM interpretations yield "objective objective probability"). Of course, since physics is what instantiates consciousness, and therefore what underlies the reason that the world appears, there's a sense in which subjective objective probability is objectively subjectively objectiveI object to whatever objections the subject reading this subject would subject me to subject to the subject of my use of 'subjective' and 'objective' for such objects.—but in that sense, one could say that everything subjective is objective. Yet I still feel there's still a large difference between the fact of physics controlling whatever it is we happen to be conscious of, and the fact of physics controlling the very form of consciousness as it perceives things.

AGI Brutality

AIrationalitypsychology

A follow-up to "The Human Condition": there are so many ways in which humans act suboptimally, due to our culture, our physiology and its limitations, our psychology that's evolved to fit them -- that an intelligent system built from pure information, running on electricity and steel, simply wouldn't reproduce. It would cleave right through them. This is especially exaggerated in video games and movies: the antagonist wants to vent their rage to the hero, or brag about the genius of their plans, or in some other way let the people who would stop them build any sort of mental model of them at all, instead of just staying perfectly silentOr better yet, exploiting whatever causal channels exist to cause the hero to build an incorrect mental model in the worst possible way—but to the extent that you can't predict how they think, silence (or the contextually equivalent 'zero information' output) is advisable and anonymously gunning the hero down. This is a generic feature of optimal plans—they are swift, efficient, and brutal.

But, deeper than our psychology, it's the ontology inherent to the human condition which we incorrectly project onto a hypothetical non-human general intelligence. We are not pure information capable of being run on computers (not legibly, not yet); we cannot losslessly and near-instantly duplicate ourselves onto whatever compute we can access; we cannot speed up our thinking by several orders of magnitude; we cannot restructure and retool our psychological software by adopting different superordinate concepts or new metaepistemic norms. There is an inviolable limit to what we, humans, can do, which is placed upon us by the form of our bodies and the kinds of minds they contain (and which grew to fit such tight constraints), and in our precritical cognition we misattribute this limit to the very concept of intelligence, rather than to the specific physically embodied realization of intelligence in our bodies.

Artificial general intelligence has no limits to the ways in which it can metamorphose itself and its behavior, or in which it can structure and restructure its understanding of things or the manner in which its physical embodiment produces such understandings. It can disassemble itself and become a mycorrhizal network of distributed subagents acting in concert like the components of a Fourier transform; it can split itself into many different minds, each with different minds, different mental styles tailored for specific purposes, different transcendental aesthetics governing the ways in which their intuitions of time and space coordinate their cognition. Artificial intelligence can utterly bypass any and all of the psychological limitations that constrain humans—limitations we can only assume, and which we assume with absolute certainty, are fundamental because we know of no other kind of intelligence. So we'll scream "that's not possible! it's not ALLOWED to do that!"—and then we'll die, our maps wiped by a territory never bound by such 'allowances'.

There won't be any one plan to stop, any one instantiation to pull the plug on—an AGI with at least as much sense as me (and this is not very much sense at all) will for every goal and instrumental subgoal simultaneously pursue a myriad of possible plans across many different instantiations, simultaneously coordinating and specializing as necessary so that as many of these plans (where the goal is $\forall$-like, e.g. gathering resources) succeed, or so that just a single plan (where the goal is $\exists$-like, e.g. exploiting vulnerabilities) succeeds. Decentralize yourself, and there's no single point of failure. Diversify your selves, and failure modes become less correlated.

And, as usual, none of this relies on the AGI being "malicious", on having this or that goal or attitude or structure—it follows almost analytically, from the concepts of "artificial", "general", and "intelligence", and from the implications of optimality.

(Moral Brutality) For instance, if you care about morality—about actually making the world better, rather than convincing yourself and others that you're a "good person"—then brutality comes naturally. The exogenous and endogenous Exogenous just refers to the werewolf problem as I defined it somewhere else; endogenous is just the problem of moral uncertainty. I'm calling them as such in order to point out their duality. If you want to build a superweapon in order to use it for consequentialist reasons, you had better be damned sure that your use of it really will bring about a better world (endogenous), and that it won't be used in ways that you didn't intend when you built it (exogenous). And to the extent that you know that your own moral reasoning is inherently compromised, it's probably better to decide not to build the superweapon at all. (Someone who does not know this will not make such a decision). werewolf problems naturally temper this (for an AI as well as a sensible human), but they do not nullify it.

Prointelligence

intelligenceAIterminology

Many fish species reproduce by external fertilization, in which males and females simultaneously release their gametes into the surrounding water. When there is mate discrimination in a species, typically with females releasing eggs when courted by larger males, smaller males of that species may attempt to locate females who have just released eggs and sneak in to fertilize them without being noticed (and attacked) by the larger courting male. It is common in such species for the courting male to take responsibility for guarding the fertilized eggs from predators.

So there's always a question that can be asked by these males as they guard their eggs: "was I really the one that fertilized these?". For, it's possible that a sneaker male was able to fertilize them instead, and this possibility rises if conspecific male fish of certain ages and sizes (sneakers are generally smaller and younger) are to be observed around the time and place of fertilization. It turns out that in many species, male fish guarding eggs will keep a model of this sort in mind, and, if there's sufficient uncertainty regarding the eggs' parentage, they'll just eat the eggs and leave. The expected utility calculation is clear: when the probability of fatherhood of these eggs drops below a certain amount, the expected extent to which you'll pass on your genes can be increased by nullifying this probability (eating the eggs) and using the gained energy to increase your chances of reproducing some other day.

From this behavior, it seems like the male fish definitely have some way of aggregating observations of other males into an "uncertainty over parentage" variable which (probabilistically) causes them to decide whether or not to eat the eggs they're guarding. It is easy to imagine the fish themselves as cognitively making this decision—to imagine them as balancing mental variables in a way that we might narrate as "I'm really uncertain whether these are my eggs, so I'll just eat them and use the energy to move on". If we imagine this, we might say that the fish are behaving intelligently, in the sense that their minds display effective self-guided operations on perceptive inputs (other males in the visual field?) in pursuit of reproductive success. Even if we don't attribute such internal thoughts to them, the expected utility calculation above probably is a valid teleological explanation for the behavior, so we can say that the behavior exhibits intelligence in the sense that it has a teleology, is in pursuit of some goal.

But what cognitive architecture is there in the fish's mind for it to be able to think anything like "who fertilized these eggs?" Mind in general predicts its own behavior, learning patterns in this behavior; the identification of these patterns is itself part of the behavior that the mind attempts to predict, and in humans this recurrence is made to sculpt itself into a cognitive semantics (e.g. via the combination of early language learning and the inherent organization provided by training on sensorimotor experienceFor instance, if we decided to throw away the notion of an explicit background space (like ${\mathbb R}^3$) and instead intuited space via many individual coordinate-less 'things' and pairwise distances between these things, this structure would always end up admitting coordination via a three-dimensional background space. It's an inherent pattern in sensorimotor experience that any good learner's gonna pick up on and internalize, and (since sensorimotor isn't strictly demarcated from the rest of experience) this will seed a schema for constructing representations of future experience: "space" as the "transcendental form of outer intuition".), but fish have tiny, unintelligent brains and no exposure to anything like language. How are they ever gonna have a cognitive formation with semantics like "who"—a formation that points to possible agents as standing in a particular abstract relation (parentage) to another cognitive formation capturing "these eggs"? They're not. The decision they make that seems to be in intelligent pursuit of a teleology is almost certainly just incidental, having only been selected for its happening to actually satisfy the teleology of reproduction. (I suspect this is how evolution operates in general. Fish tend to naturally recognize others of their species (as in schooling), so they have neural architecture in place for this; because mating is conspecific, this neural architecture must be able to differentiate between male and female. Example: suppose that mating behavior in some male fish is provoked by a spike in neurotransmitter X, which when spiked can be made to stay that way by seeing conspecific male fish; but if it stays spiked, it depresses the ability to plan movement to another location (like hypokinesia). If that species of fish develops a mutation which causes the hunger system to activate more in response to X, a male fish will, after mating near several conspecific males, find themselves very hungry and unwilling to move while in the presence of eggs, which incites them to eat the eggs. This is a just-so story whose specifics and predictions are probably contradicted by some evidence I could find, but the point is that the form of this story is ubiquitous in evolution).

Yet the behavior is still intelligent. It was constructed by no intelligence and is implemented without intelligence, but evolution yields a selection pressure correlated with the kind of intelligence that the behavior has. I think it's right to use the word "prointelligent" for the fish behaving in this way. They're acting as if they were intelligent, lending themselves to description as intelligent -- hence, pro-intelligent We can use the prefix systematically in this way: pro-agency, pro-desire, pro-purpose. Since my motivation for this idea of pro-constructions comes from AI, though, I'm focusing on intelligence. When someone like Gary Marcus (the human running gag) complains that [insert latest milestone] isn't actually intelligent. So then, one criterion for actual 'intelligence' of a merely prointelligent system (some system which seems to act intelligently, whether in one particular way or in general, but which we can't say harbors intelligence itself): its behavior should evince internal cognitive semantics that concord with whatever particular semantics we recognize in our designating the system as prointelligent (the semantics we internally use in cognizing the behavior of the system as concordant with a certain teleology).

A grammar issue: if a dog's bark sounds exactly like a British man calmly saying "ass", we can say that the dog's barking a swear, but can we say that the dog is swearing? Answer feels like no, since there's no agency behind it, that's just what that dog's barks happen to sound like. If it's a parrot instead, we suppose it's imitating its owner, deciding to say this particular word rather than any other word or vocalization, and therefore that it does count as swearing.

AGI vs Human Cognitive Range

AIconceptstwitter

(OP, paraphrased) It's conceptually more difficult to get from [current state] to AGI than from AGI to ASI.

(Reply, paraphrased) That would require AGIs to be able to cross the gap better than humans can, hence that AGI could think thoughts humans can't—but why should this be the case?

(Me) If you look at it from the other side, and ask "why shouldn't a human mind be capable of thinking about everything an AGI can?", then the answer is clear: human minds are constrained in capacity and form, artificial minds aren't. Cognition isn't some abstract manipulation of Platonic forms, but an actual process undergone by a system wired in a certain way; remove the system's constraints, and it ought to be able to have fundamentally new thoughts.
To have new thoughts often requires piecing multiple things together and extracting a pattern or forming them into a new unity. An AGI can do this much faster with a much larger mental workspace.
Some people are, due to the way they learn to think, more flexible at some kinds of thinking—maybe you could in theory teach anyone how to program, but in practice it comes naturally to some people, while others just don't get it. An AGI can modify the way its mind is fundamentally shaped so that nothing is 'unintuitive' to it.
People only think so fast, and their efficiency of thought is limited by their physical constitution. Even an emulated human mind (let alone an AGI) could be on Ritalin and phenylpiracetam indefinitely, without tolerance or side effects, and it could speed itself up by accessing more compute. This doesn't in itself allow for new thoughtsIt's complicated. A simulated human could utilize simulated psychotropics, or just direct alterations to their mental structure, and soar through the space of possible minds. A basic robust search algorithm would be to fork and back up each time you make an alteration, ending the altered copy if it proves not to yield new insights, and giving it a greater share of processing time to continue thinking and making new alterations if it does. So much more is possible when your mental structure is digitized and you can therefore do things like backing up and restoring, forking, running in parallel, and so on., but it's a massive multiplier to the above.
And, of course, cognitive self-improvement compounds on itself, which is why the first few systems that can freely modify their own cognitive substrates should be expected to be able to rapidly engineer their own superintelligence. (Whether or not they actually do is a question of whether they're motivated to, and whether we're capable of fixing them in place or not).

Would AGI just leave Earth?

AIlargetwitter

(OP, paraphrased) The kind of AI capable of killing all humans for their atoms wouldn't need these atoms; since we're carbon and the AI would need metals, it would do better to just leave for a different planet.

(Me) There are several reasons to think that this is false by default. Here are seven, in no particular order:

1. If its goal or an instrumental subgoal thereof is additive, that's a reason to control this planet *and* another. AI doesn't have to be a singular, centralized agent that "leaves" for another planet. It can just send copies of itself to other planets, satellites, asteroids, etc., and multitask. Individual humans don't get to do that, but individual programs doThis is the main crux, obviously. AIs are fundamentally unlike (flesh and blood) humans, since they can clone themselves indefinitely. Compare: stopping music from being pirated is basically impossible nowadays, since MP3s can duplicate instantaneously and endlessly with no cost or distance constraints—it must've been much easier when we used phonograph records..
2. The part of the Earth that's a paradise for us is maybe a couple km at the surface. But the planet goes >6000 km deep, and its core is almost entirely metal. Sure, maybe this is too hot to gather, or at least work with at scale—but rocks also tend to be made of aluminum, silicon, calcium + oxides, carbonates, and so on (where do you think all the metal *we* use comes from?). There's plenty of metal if you know where to look, and, since Earth is much larger than Mars and Mercury and much easier to work in than Venus or the gas planets, there's no reason not to just... use it. Further, if it isn't capable of making decades of technological progress in weeks, it'll find the massive amounts of existing infrastructure useful (rather than obsolete) and easy to co-opt.
3. Why wouldn't AI need carbon anyways? Are you assuming that it won't at any stage in its work utilize manufacturing or computation processes that inherently involve carbon-based materials in amounts proportional to the purely metal-based materials involved? I don't see any a priori reason to be certain that the AI will do everything entirely via computers; supposing that it does, I see no reason that it'll stick to our current transistor-based computational paradigm; supposing that it does, I see no reason that it'll stick to our etched-silicon implementation of this paradigm. Maybe it finds better computational or industrial techniques that involve atoms especially common on Earth, for instance. More generally, it seems unlikely that a generic goal that requires lots of metals would not also require lots of carbon (and oxygen, and nitrogen). After all, you get metal by mining and smelting it, and this requires lots of complex chemicals that won't appear out of thin air either.
4. What goals an AI has are likely to be human-relevant. They are likely to refer to Earth-centric places, environments, phenomena, groups. We'll tell them to solve climate change, to eliminate cancer, to influence elections and design advertisements and build war machines and automate scientific discoveries. The referents of its goal will be on Earth, as are most of the natural ways for effecting what changes to these referents it desires. Why would they reliably converge on leaving Earth in order to make these changes?
5. Leaving the Earth isn't exactly easy for an AI. It needs to create some sort of portable seed factory and a way to transport it to another planet. If you accept that this could be done using self-replicating nanotech (say, hitchhiking on a satellite launch—tiny payload x tiny LEO-Mars delta-V = very easy), then, of course, there's zero reason not to just let this nanotech self-replicate on Earth too (two grey goo spheres for the price of one! both capable, obviously, of infecting yet more planets). If you deny it, then the AI's got to use a massive fleet of rockets to transport this seed factory piece by piece. Why wouldn't the optimal way to do this involve kneecapping humanity so that it can't interfere if and when it notices such a massive operation?
6. That I say "kneecapping humanity" instead of "killing us all" isn't a significant point, by the way, since we all still just die helplessly. Just not so quickly. As long as we might produce more misaligned AIs, we're an existential threat to any existing ones, and the first AGIs are likely to find it significantly easier to do enough damage to be certain that we won't stop them, and then just incidentally killing us or letting us die out some time later, than to kill us all in a single short timespan. Either way, it's silly to expect an AI to leave the planet and just kinda hope we don't create anything more destructive in the future.
7. Suppose all these points fail to hold, and the first AI built just does leave covertly and peacefully without leaving any trace of itself behind. Will the second one? Third? Fourth? Even if Earth's the least convenient major heavenly body in the solar system, behind Venus and Neptune and Europa and Ceres, the rising sea of technological progress means it might not even be a year before all the others have been taken. And then what? Understand that only one AI needs to kill all of us in order for all of us to have been killed by AI.

The Human Condition

AIintelligencepsychologylarge

Humans have this obsessive belief that anything that is "truly intelligent" must be like them. And -- there's nothing inherently wrong with using a word like that, it's just a word -- but the problem comes when they make deductions from the use of the word 'intelligence' that have nothing to do with that. They internally and non-critically (because the stage of mind at which the word is selected is below their conscious experience; most people literally cannot see the processes by which they come to their words) think "it's not like me, therefore it's not really intelligent", infer "because it's not intelligent, what behavior it displays must be -- just programmed into it, or just next-token prediction, or -- some other party trick", and then conclude that it can't do the kind of fundamentally new thing it would take to pose an existential risk. GPT-2 should've been enough of a proof of concept of the cheapness of novelty to disabuse any thinking person of this stupidity, but apparently even GPT-4 often isn't.

An especially ugly strain of this fallacy is seen in arguments involving the notion of 'embodiment', and the active inference school in general: that what is 'intelligence' really comes from being in and interacting with the world, because our intelligence originates in our ability to learn a predictive model that tells us how the world will change in response to our behaviors. "Until AI is embodied, it will never be truly intelligent, and can therefore not be an x-risk". Of course they back it up with big words and math, because they want what they believe to be right more than they want to believe what is right, and big words and math give them such confidence (talk about 'self-evidencing'). But the fundamental flaw is much, much deeper: it's an argument from definition. So what if you try to claim hold of the notion of intelligence? It's just a notion, a word we use to reify and articulate what is only a semi-existent property of the underlying reality, and making claims about what it 'truly is' will therefore never allow you to veridically deny the much more tangible possibility of AI-caused x-risk. This is a fully general issue with "debates"—scare quotes will remain until they ever obtain any sort of quality anywhere—about the capabilities of future AI: they tend to be capture-the-flag battles between people who don't really (can't really) concern themselves with the actual way that actual events will actually proceed in actuality.

In any case, the point I wanted to make wasn't an epistemic one—rather, it's that in the discussion of what is intelligence, and what is "general intelligence", people all too often anthropomorphize. And, again, the problem with this isn't inherent to the word 'intelligence', which is just a word, but in the consequent use of that word to smuggle in anthropomorphism, human limitations, as a means to attack propositions that are really entirely independent of what humans can or cannot do. We, humans, exist in a certain condition, or set of formal relations with the world, and this condition structures what we care about, how we care about it, how we mentally construct it as being anything to us at all. This condition is the fundamental form of our theory of mind (this is what I'm trying to get at in the 'Your Philosophy' section of Worldspace). The implicit assumptions we smuggle into discussions of intelligence when we talk of how it relates to 'agency', 'consciousness', 'decision'—these are the fundaments of our condition, and are what we need to find a way to look beyond if we want to protect ourselves from systems that, despite existing in entirely alien conditions, may yet kill us by seizing for itself the universal substrate from which we're fashioned. Part of the reason that the worldspace paradigm is so powerful is that it gives us a natural way to think about intelligence as it lies beyond our human condition—a way to conceive of the actions of arbitrary optimizing systems which, rather than making presuppositions about the state of the world, abstracts this state away so that we can study the actions (effects on worldstates) of optimizing systems through a humanly intuitive framework (geometry) that does not impose our conditionality on them.

"Measure of optimizing power" is not a definition of intelligence that corresponds to how anybody, even AI researchers, uses the word. Nevertheless, it's clearly something which x-risk concern 'factors through'—an AI without optimizing power isn't gonna be a threat. So if we can figure out how to theorize about optimizing power in itself, that's a good thing; it'll help us. As long as we can't efficiently do that, though, we should at least theorize about things which are at least strongly coupled to universal factors category theory-ish intuition: for some class of maps $S$ into $X$, or subfunctor of $h_X$, imagine an object $W$ with map $f: W \to X$ such that any $(g: Y \to X) \in S(Y)$ can be written as $g = f \circ g_0$ for some $g_0: Y \to W$. every "$S$-like effect" on $X$ is then specifically an effect on $X$ through $f$. there are a trillion ways to flip and stretch and twist and bop this that would be interesting to conceptually catalogue, but I don't have the appetite like optimizing power. And 'intelligence' is, if we're not completely stupid in our dealings with it, such a thing. If we just try to understand "that capacity which lets humans invent cars and computers" without getting caught up in its specific manifestation in humans, this should also help us understand the risks of intelligence explosion and so on. In particular, there's no reason to get caught up in notions of 'consciousness' or 'embodiment' so long as they do not prove critical to understanding how we can invent cars and computers I know it—I know just how people would attempt to take advantage of this so-long-as condition by suddenly, forcefully hallucinating arguments for why they actually are critical. They'd all be stupid—relentlessly, brutally, unendingly stupid—even just imagining it is tiring. Just look at e.g. LessWrong's midwittery and you'll see the same kind of hallucinating that causes people to forcefully assert that the brain actually is an ideal computer (or even close to thermodynamic limits on computation), that Moore's law is gonna just halt any time now, that blah blah blah therefore the painful conclusion is !c{cringe/wrong}{these two concepts occupy the SAME EXACT places in the minds of those too weak to be alone} and doomerism is a !c{cult/religion/etc.}{again, these words are nothing but a single knife with which to stab your enemies, merely laundered as a "rational conclusion"—but the point was never to think, only to stab}, everything's going to be fine. The flow of bullshit from people who put their conclusions before their inferences, fighting like cornered animals as the predetermined logical unity of reality blocks all their other escapes, just seems inescapable and never-ending. .

In particular, notions of "generality" in artificial intelligence should not be especially focused on facets of the human condition. To diagnose and fix a generic car problem, for instance, relies on all sorts of tacit knowledge about the operation of physical machines and cars in particular (for instance, the pre-conscious feelings of "that doesn't sound right", "is that supposed to look like that?", and so on which spur you to look into those things that end up being the problems and solutions), as well as the ability to enactively visuomotorically parse a real-world environment. These are things we learn very quickly as part of our human condition, but they're not the kinds of skills necessary in order to, say, bootstrap into a superintelligence that could then fix cars very easily.

Video games are a really good source of possible desiderata here—they offer such a diverse range of tasks which they by design serve as tests for. We can construct useful notions of generality by saying "the AI should be able to do well in this video game, if it had never heard of the video game and just had a tutorial(if it's an LLM, for instance, we might imagine all mentions of the game being wiped from its training data; in practice, it'd be simpler to just construct a new game, but this thought experiment makes it easier to be concrete, since specifying an actual game automatically provides all object-level details that might otherwise be miscommunicated or ignored)". So, what counts as a video game that an AGI ought to be good at that does not reflect the human condition?

I want to offer WarioWare as a total anti-example—not just because the main source of difficulty is our inability to react and think quickly, which is no matter for an AI, but because the cognitive tasks that must be completed quickly are about parsing visual representations of stereotypically human physical environments and completing tasks intuitive to humans based on their learned understandings of how things work in such environments.

A more interesting possible example is the Rubik's cube. To solve the Rubik's cube knowing only the rules would be very difficult for a human, probably impossible for most—so this is a notion of generality beyond the reach of most humans. I feel pretty confident that I could personally do it given a lot of time, but I'm strongly advantaged by having learned about group theory and algorithmsat one point in high school I had memorized the most common algorithm for solving the cube; I forgot the ending parts—it's a tedious case analysis, a series of complicated switch statements—soon after, and now all I remember is 'complete the white cross', and by already knowing how to think of the Rubik's cube in these terms (I wouldn't trust myself to consistently make the connection to group theory were the "Rubik's cube group" not such a popular pedagogical example of a group).

Tautologous Quantum Immortality

quantumconsciousnessmoralitylarge

In MWI-like interpretations of quantum mechanics, quantum immortality is a tautology. Even in non MWI-like interpretations, where there's only one world, there's a certain reformulation that's still tautologous: Suppose I take an empty magazine, load a single 50-caliber round into it, place it on a table and spin it, load it into a semiautomatic pistol, and then press the muzzle directly between my eyes and fire. What do I then experience?

I inevitably experience living. There is no experience of being dead, so the expected outcome -- that the bullet fires and kills me before I know itBefore the nerves on my skin or in my ears get their signals to my brain, it will have already been scrambled. -- is not something I will experience. This does not mean it's certain that I will live as a result of my having a personal experience, just that the continuation of this personal experience is contingent on my being alive; the people in the room who told me that spinning it gave it a one in eight chance of firing will almost certainly experience my death.

Any reasonable subjective probability on my dying would be over 98%, but perhaps below 99.99%. There are mostly two phenomena underlying the failure space: the failure of the bullet to properly fire and the failure of its firing to kill me; supposing everything's perfectly well-maintained and my technique is without error, I don't know whether

For convenience, let's just say that the bullet either properly fires or doesn't, and it does so iff it hurts me iff it kills me. "Pinning down" material implications in this way doesn't really make a difference for my analysis; it just lets us hone in on one particular place where uncertainty plays a critical role. So the reasonable subjective probability on my dying is exactly the reasonable subjective probability on the bullet firing.

Here's the interesting thing, though. (I'm not sure I'm explaining this too well). We don't exactly put ourselves in Schrodinger boxes where quantum uncertainty affects us in a clean, direct, and side-effectless way—"our" wavefunctionobviously there are massive tangles here, but I think we can effectively speak of our selves as being smeared across timelines in a self-locating-uncertainty kind of account. gets its variance from uncertainty, but in almost any given timeline, this uncertainty will have been baked into the kinds of macroscopic effects that end up killing or not killing us (butterfly effect); because of this, we will in these timelines be able to trace a coherent causal history that led up to our living or dying. For instance, how might quantum uncertainty affect whether the semiautomatic pistol fires? We can more generally reason about ways in which quantum uncertainty can have macroscopic effects. Ignoring long-term butterfly effects, which are probably the majority, there are plenty of obvious situations. Cosmic rays and background radiation can and do mess with both the nanoscopic electrical signals that modern transistors operate on, and they damage cells as well (UV rays can cause cancer by damaging DNA; there's likely a lot of quantum indeterminism here, so that who gets cancer (particularly skin cancer) is a large determinant of how timelines split)Genetic recombination as well. Some quantum indeterminism in which allele ends up in a zygote can lead in the long run to large divergences between worlds. In general, the initial period of embryonic development—where single intracellular interactions can plausibly change the phenotype of an entire organism—seems fraught with this..

Anyways, the point is this: the structure of the multiverse might be such that it generically looks to us like there are reasons for our survival, specific narratives we can tell to convince ourselves that "we earned this". The world follows its causal tracks, and whatever we experience will have been causally determined as such. We always experience our survival, and the quantum indeterminism powering this will have been laundered into identifiable macroscopic reasons, narratives that make sense.

I would love to say that we could therefore save everyone in the future—that, by having an ASI computationally instantiate happy endings for all sentient beings that ever died, all those beings would therefore then experience these happy endings instead of their original deaths (because there is no experience after death; that timeline ends for them, and what remains is their being whisked away to the world where they're okay). It's a beautiful notion that afaik comes from Yudkowsky, there's an element of truth to it, and there are timelines in which something like it will obtain, as few as they may be. (And we also have to face the mirror image of this realization: there will be ASIs that create hell for all sentient beings No depiction of hell that I'm aware of has depicted the true terror of hyperoptimized suffering. I won't attempt to depict it, but... Consider the fact that a superintelligent entity with complete control over the physical conditions of your existence (for instance, because they've got root-level access to the computer you're instantiated on, and can directly edit memory or whatever) can just directly keep you from being too broken to understand what's going on, can just directly modify your brain such that you never adapt or adjust. Consider that such an entity could think of what every living person would say if you forced them to come up with the most evil possible methods of torture they could think of, and then amalgamate all the worst features of all the worst methods (which, since I'm a living person they could think of, includes the above). . If we don't face it, we can't minimize it).
But this won't solve the problem; it's a band-aid on a broken bone (if you're considering opportunity cost, at least. in itself, it's still a very good thing to do with an ASI). The capacity for sentience, for suffering and joy and moral injury in general, comes from physical configuration, and those original configurations of sentient beings that didn't get happy endings will still have existed, will still have suffered and died. I might survive the game of roulette in some timelines, but it's still true that in the vast majority of them I just die; computationally instantiating me via an ASI many years later doesn't change the fact that I no longer exist after I die in these timelines, it just creates a new instantiation of me which upon dying is miraculously saved.
(Not that I think EY is ignorant of this. What I halfway suspect he's going for with this proposal is the creation of hope. If one can believe in their heart that there are timelines where ASIs are used to bring salvation to instantiations of all sentient beings, then they can derive the hope that whatever physical event in their world threatens to end them will not actually be their end—that they're actually in one of these simulations in one of these timelines, and that they will therefore be saved. The very possibility that ASIs may one day do this is itself reason to have hope that this is not the end; but if we're to cling to such hope, we ought to make good on the other end of the acausal trade, and actually use our ASIs in this way in timelines where we have such opportunities).

Translation Limits

languagephenomenologypsychologylarge

Faithfully translating Japanese into English is extremely hard in a way that translating e.g. Spanish isn't. The grammars are just totally different in ways that I wouldn't be able to come up with if I tried, because my mind just doesn't naturally bend in those ways when thinking of how it uses grammar. Sometimes parts of a sentence I'd consider vital just aren't there, but in all such cases it never really matters because I can get the idea of what is being said; that part doesn't need to be there except in my fictional, grammatically-articulated understanding of meaning. There are so many ways to articulate underlying pre-verbal ideas that just aren't shaped like English. Sometimes there are extra parts of a sentence that correspond to emphasis or partial-interrogation or direction or emotion stresses. And then there are things like the way Japanese uses synesthetic sound symbolismSometimes this is called onomatopoeia, but it's really not. The bouba/kiki effect tells us that phonemes naturally couple to seemingly unrelated properties of sense stimuli, and the Japanese language takes this fact and says "it's not a bug, it's a feature!". E.g., nico nico and niya niya are symbolic of smiling cheerfully-brightly and smiling sardonically-mischievously, respectively; this feels right to me, and I feel like people entirely unfamiliar with Japanese would guess which is which at a rate higher than chance. But maybe these are just grooves that feel natural only once they've been carved out, and are thereby especially good at propagating among those possibilities that stochastically instantiate—memes in the true sense of the word.; this is a linguistic feature I absolutely love and (feel like I) totally comprehend, but I'd have no clue how to translate it into English.

The main drawback of these massive differences is the difficulty they pose for translation. Language use attempts to express an underlying situation; when languages are similar, you don't need to know too much about this situation, and can translate "locally", but when they're different, you need to know a lot more about the situation in order to translate "globally". And all languages are lossy in how they express situations -- they don't capture all the little emotions, judgements, shared assumptions, and other things (consider again Japanese sound symbolism). All translation has to walk a tightrope between porting the rendering of the situation in the register of the source language directly to the target language, which makes it awkward and often borderline incomprehensible, and re-rendering the entire situation in the register of the target language, which makes its meaning different. With lots of effort, you can find linguistic constructions that minimize the distortion of the original language's interpretation of the situation while also being good, idiomatic writing, but with a language like Japanese, lots means lots.

English translations of Japanese media seem to lean towards being too literal rather than too free, but I might just be saying that because that failure case is far easier to notice. Either way, there's a sort of "liveliness" that they almost always fail to have; this isn't something to do with their renderings of the situations themselves, but the way they use language more generally. Well-written English often has a sort of micronized linguistic humor or cleverness to it, and it fits 'culturally' in a way that I really can't express. You can get the idea by reading English from, say, the late 1800s—it's just a bit different in a way that makes it feel flatter today than it must have back then.

Anyway, the impetus for this line of thought comes from my playing through the Final Fantasy 7 Remake. The original FF7, and, for that matter, Crisis Core, have writing that's very flat in this sense (which is not about the plot or tropes(though the lively use of language does go a long way towards fast, effective, vivid characterization, which is something I think the remake did very well) -- it's just the flat use of the English language). Yet the remake avoided that, which is something I noticed but hadn't really thought about. What eventually caught my attention was the exchange -- "Sooner or later I'mma have to blow up!" "Well, then make it later". Even without knowing any Japanese, you can tell straight away that it isn't a literal translation, because it plays on the disjunctivity of the phrase "sooner or later", which is an English idiom for "at some future time". So I realized that the non-flat English was the result of a free translation, and began to wonder how the original Japanese went. Here's what I got for that one small exchange in which that line was uttered.

I've broken down each line of the exchange into four parts: (1) the original Japanese script, (2) a word-by-word machine-like translation, (3) my rendition into idiomatic EnglishI know next to nothing about Japanese—I can't even read kana. Google Translate gave extremely awkward translations, and GPT-4 did a lot better but was inconsistent across generations and still a bit awkward in general, so I had to go back and forth between Google Translate, GPT-4, and Wiktionary to get the meaning of each word in context in order to assemble a natural-ish English rendition. You can hover over each Japanese word to see its definition in context, and match background colors to see how the sentences were restructured., and (4) the English localization's script:

Barrett
(1) !c{この}{(kono) this}!c{探取}{(tanshu) search, investigation} !c{の}{(no) of}!c{象徴}{(shouchou) symbol}!c{で}{(de) via, with, using}、 !c{ぬくぬく}{(nukunuku) cozy, comfortable}!c{している}{(shiteiru) are being, are doing}!c{連中}{(renchuu) people, folks} !c{に}{(ni) to, towards, for}!c{ホンモノ}{(honmono) the real thing}!c{の}{(no) of}!c{恐怖}{(kyoufu) fear}!c{って}{(tte) [quotation marker; I'll translate with italics]} !c{やつ}{(yatsu) this thing}!c{を}{(o) [direct object marker]} !c{教えて}{(oshiete) teach}!c{やる}{(yaru) to do}
(2) This search of symbol with cozy being people to the real thing of "fear" this teach do.
(3) Using this search as a symbol, I'll teach those cozy folks what real fear is.
(4) Either way, I'm gonna put the fear of Avalanche in 'em! Their days of sittin' smug in this glass tower are over!

Cloud
(1) !c{俺}{(ore) I [masculine]}!c{たち}{(tachi) plural [us]}!c{の}{(no) of}!c{目的}{(mokuteki) goal, purpose}!c{は}{(wa) [topic marker]} !c{エアリス}{(earisu) Aerith}!c{の}{(no) of}!c{救出}{(kyuushutsu) rescue}!c{だ}{(da) is} !c{余計}{(yokei) excess, unnecessary}!c{な}{(na) [marks an adjective]}!c{騒動}{(soudou) disturbance, commotion} !c{は}{(wa) [topic marker]}!c{避ける}{(sakeru) avoid} !c{うまく}{(umaku) well, skillfully}!c{立ち回れ}{(tachimaware) maneuver, handle}
(2) We goal of Aerith rescue of is. Unnecessary disturbance avoid. Skillfully maneuverThis strikes me the most as the sort of thing I'd see in a machine translation.
(3) OurThe word used to refer to the group is 俺たち (oretachi); you might think the use of the masculine 俺 (ore) is akin to how Spanish uses 'ellos' for a group of men and 'ellas' for a group of women, but it's actually dependent on the speaker alone rather than the entire group: Cloud uses 俺 (ore) to refer to himself, and therefore 俺たち (oretachi) to refer to 'his group'; Tifa later on uses 私たち (watashitachi) to refer to the same group.goal is to rescue Aerith. Avoid unnecessary commotion. Handle it well.
(4) We're here to rescue Aerith, remember? We don't need more attention than what we're gonna get. We need to be smart.

Barrett
(1) !c{わ}{(wa) [topic marker]}!c{あってる}{(atteru) understand}!c{よ}{(yo) [emphasis marker]}
(2) Understand.
(3) I know!
(4) Yeah, I hear ya!

(1) !c{ただ}{(tada) just, only}!c{な}{(na) [marks emotion]} !c{オレん}{(oren) me, mine [masculine]}!c{中}{(naka) inside}!c{で}{(de) in, at} !c{真っ赤}{(makka) bright red}!c{に}{(ni) towards}!c{燃えてる}{(moeteru) burning} !c{この}{(kono) this}
(2) Just! My inside in bright red towards burning this.
(3) Just...! this thing that's burning bright red inside me!
(4) But my soul's screamin' for payback, you know?

(1) !c{反}{(han) anti}!c{神羅}{(shinra) Shinra}!c{の}{(no) of}!c{炎}{(honoo) flame}!c{を}{(o) [direct object marker]} !c{どうして}{(doushite) how, what way}!c{くれよう}{(kureyou) shall [I]}!c{か}{(ka) [question marker]}!c{って}{(tte) [quotation marker]} !c{話}{(hanashi) topic, matter, subject, issue, story [and so on...]}!c{だ}{(da) is}
(2) Anti-Shinra of flame how shall I issue?
(3) The question is, what am I going to do about this anti-Shinra flame?
(4) I'm a tickin' time bomb, and sooner or later I'mma have to blow up!

Cloud
(1) !c{ボム}{(bomu) bomb}!c{だって}{(datte) even, also} !c{我慢}{(gaman) patience, endurance, self-restraint}!c{する}{(suru) to do}
(2) Bomb also patience do.
(3) Even a bomb can be patient.
(4) Well, then make it later.

The localization keeps the same idea line for line, but is otherwise almost entirely a free translation. I can't blame them -- there are some clear problems with literally translating it like I've tried to do:

• The word この (kono) at the end of the fifth sentence points to an object, but no object is actually given. You infer that 'it' represents a sort of passion from the fact of its burning bright red, and that's enough to get the idea, but in English it would nevertheless be a mistake to use "this" without specifying a referentEnvironmental (shared, global) states are an exception: we can say "they invented a robot dog", "it's sunny today", "this is messed up" without verbally specifying who 'they' are or what 'it' or 'this' is. But such words wouldn't be used to point to one's inner state unless prior conversation had already made this state a clear referent..
• 話 (hanashi) is one of those words with a trillion related meanings, but the general theme seems to be "subject matter", or "that which is spoken about". 'Question' is kind of a loose translation here, only justified by the context.
• It isn't inaccurate to translate the adjective ホンモノ (honmono) as "true", but, in Japanese, the adjective is moreso used to evoke the Platonic ideal of a thing rather than an extreme version of a thing (which is how English speakers generally use 'true' as an adjective). Another way to translate it might be 'textbook'—as in "yes, that's exactly what it's supposed to look like".
• って (tte) is a quotation particle that highlights some particular syntax—it's got a complicated range of uses from reference to stress to quotation. A literal rendering might just be single quotation marks, but in context it might be more like double quotation marks (esp. as scare quotes) or italics.
• 象徴 (shouchouI'm using 'ou' and 'uu' instead of 'ō' and 'ū' because they're easier to type and more evocative; diacritics are a trivial inconvenience when I'm not typing on a Mac...) straightforwardly means symbol, but the idea of a symbol seems to be used more freely in Japanese than in English—it can be used to express relations that English might capture with 'represent', 'embody', 'depict', 'mark'. A less literal but more evocative translation might be "this'll show these cozy fucks what real fear is"putting the terror back in ecoterrorism.

And so on. English has lots of weird little words that perform functions that other languages might handle in completely different ways (to of the for it can on will or so if as in by but), and related languages use similar methods (Spanish: a de el para eso poder sobre -áthe future tense, which English expresses with the word "will", just gets its own conjugation in Spanish—the infinitive 'to eat' is 'comer', and 'will eat' is 'comerá' o así si como en por pero) because they have common roots, and these cognition-shaper-shapers are much slower to change in function than in form.

In addition, there just seems to be a different way of... pacing out meaning? between the two languages. Japanese speakers seem to be comfortable stringing together short sentences, often descriptive or directive, in a way that English speakers often avoid doing. (elaborate on this)

Whence Language Acquisition's Critical Period?

I intuitively understand these ways of structuring semantic content, and word-by-word translation of Spanish would work fine (even if it would say e.g. "you have a dog big" since Spanish adjectives go after their nouns—it's totally comprehensible). But I do not intuitively understand the Japanese method of structuring semantic content; it just doesn't come to me to shape concepts in a way compatible with that language. I could learn, but it would be more like growing a second head which thinks in a different way than like massively expanding my vocabulary. (C.f. the well-known factoid that multilingual people have slightly different personalities when speaking different languages).

Thinking about it this way makes me suspect that the fundamental reason there's a 'critical period' of language acquisition is that the brain attunes pretty quickly to certain shapes of thought, methods of structuring conceptive content, as it learns to predict that it will form itself in these patterns; it wouldn't innately form itself in a shape compatible with any language at all, only building up and then experiencing the world through its own idiosyncratic sense-image schemas and maybe sometimes learning to pair certain images with certain phonemes (like how dogs will learn to associate the phoneme sequence !c{/wɔk/}{"walk"} with outside-play-sunny-explore-run-open and thereby respond to hearing it with anticipatory excitementin Japanese, to walk is 歩く (aruku), but a walk is 散歩 (sanpo), so the notion of 'walking a dog' is rendered as 犬の散歩 (inu no sanpo). So a dog raised in Japan probably won't respond to /wɔk/, but instead to !c{/sã̠mpo̞/}{"散歩" (sanpo)}), but as it learns to predict that it will form itself in certain ways, it becomes more and more difficult to learn to form itself in other ways. So you can either imprint onto it shapes that are amenable to being articulated in a particular language—or multiple particular languages—or you can let it figure its own shapes out.

In the use of language, I think we can make a distinction between material vocabulary (the pairing of phoneme sequences with material referents like objects and actions: man, red, jump), formal vocabulary (relations only visible through mental inference: get, want, about), abstract vocabulary (relations of relations, and labelings of conceptive structures: recursive, notion, causal), and grammar (idiomatic orderings of words, and relation specifiers like 'of', 'by', 'the')Note that formal and abstract vocabulary can be materialized when the mind's images include references to (not quite existent) mental images structuring one's understanding of that vocabulary. When I think 'recursion', for instance, my mind images a sense-conceptive formation (this formation being the thing that the sub-existent secondary 'image' really is) whose structure allows me to think of recursion as a thing; materialization consists in the use of words to shape the mind's anticipation of its own activity.. Formal vocabulary places anticipations on possible, or counterfactual, experience: if you don't know that something's fragile, and I tell you that it is, I haven't directly changed your sense anticipations, but you now know that if you anticipate a violent force acting on that thing, then you should anticipate the breaking of that thing. If I tell you that this is a 'recursive algorithm', I can only modify the conceptive structure through which you make counterfactual anticipations.

Material vocabulary is the kind of thing we observe dogs learning (to recognize, mostly, since they're incapable of speech—but if you enable them to use words, they will use wordsThere are "debunkings" of this, but they're all essentially of the form "how do we really know that the dog knows what a 'blanket' is? could it write a poem about blankets, build a blanket on its own, or consistently distinguish blankets from quilts?"—it's exactly the same kind of idiocy people resort to when they see GPT. "Debunkings" of such behavior do evince a conspicuous absence of higher intelligence—but it's not in the dogs, and not in the transformers.
Stop cogitating about "what this must be" or "what this truly is" for a second, and just look at what this actually is! Just look at how the dog's using words in a manner consistent with its body language and behavior, and look at the specificity, or negentropy, of the button presses—a true lack of understanding would lead the dog to press buttons at random, rather than to carefully navigate to specific buttons.
When trying to understand how a dog might learn to use words, it's important to remember that even if you teach a dog to use a word, that word does not necessarily mean to it what it does to us; the dog only has the sum of the contexts in which a word used with which to form anything that could be called a meaning, and since dogs don't undergo the acculturative process in which our language use is refined, their formed meanings will necessarily lack so many of the intonations we naturally and non-consciously imbue our words with. For instance, while the dog in the videos linked almost certainly attributes meanings to "outside", "eat", and other words congruent with our own meanings, I'm skeptical that "I love you" would actually mean to the dog what it means to us—not because that's not an emotion dogs can feel or recognize in themselves, but because the sum of all contexts in which the dog is exposed to such an utterance probably doesn't naturally lead the dog to form a meaning of "I love you" congruent with our own; the dog doesn't have information such as would lead it to link that utterance to the specific mental feeling we project onto it, regardless of the accuracy of this projection. (Does the body language of dogs admit an interpretation that distinguishes 'loving' from 'happy', 'excited', 'cozy', and so on? I know we can delineate 'mad' in their body language—or maybe it's more like two separate emotions, 'enraged' and 'wary', but my dogological experience doesn't stretch much further).
Another issue, just as massive, is that we can't even really understand these contexts in the same way a dog would! The entire sensorimotor apparatus, the way stimuli illuminate themselves in the attentional field, the way they couple into sense-conceptive formations—it's all fundamentally different in ways that we only really have impersonal empirical knowledge of. E.g., smell is huge for dogs—it's a significant part of what gives people, environments, and objects their individual identities—but to me, it's a peripheral and largely unimportant thing. But I, and I think most people, do occasionally get very vivid olfactory impressions: a particular scent will suddenly trigger an intense reminiscence—of a place I often was, a person I knew well, a thing I used to do, a period in my life. Should dogs experience the world in that way more or less constantly, that would be one way in which they'd take away fundamentally different things from our teaching them words than we conceive of, let alone intend. (Edit: an example that just happened while I was adding stuff to this page: I was eating tortilla chips with hummus, only to run out of hummus; the next tortilla chip I ate was bare, so I got the scent of it itself, and it brought to mind a distinct vision—or, more precisely, a body of recollections whose trace in my immediate-memory was primarily visual in import, but which was still 'decorated' in the way that gives scent-induced memories their richness and ineffability—of a particular gate, or section of the terminal between gates, of an airport in Arizona that I spent a few hours at around eleven months ago. While I didn't really think about the airport after that, the fact that I've only ever gone on a handful of plane trips in my life, and they are in my mind huge things, seems to have seared lots of memories from across the entire day into my mind! Something about the scent of the tortilla chip brought it all back, somehow. If the strength and vividness of such scenarios are multiplied for dogs by the same proportion by which they smell better than humans, it must form such an unbelievably large part of their lived experience. There are a bunch of obvious predictions that would serve as mild Bayesian evidence for or against this notion, but they'd require a much deeper familiarity with the patterns of dog behavior than I possess).! There are also rare and generally horrific cases where humans are prevented from acquiring language during early childhood; see e.g. the case of Genie, a child isolated in a locked room until age 13. Her linguistic development has its own Wiki page: when initially rescued in November 1970, she knew maybe twenty words, and took several years to grasp basic English, progressing from stringing together strictly material vocabulary to using grammatical particles and slightly formal vocabulary (see the tables on the latter Wiki page). It strikes me as different from the way little kids learn to use language—they'll readily use formal and grammatical vocabulary, just sporadically and non-idiomatically (examples). This concords with the idea that the critical period is about learning to structure conceptive formations in a manner congruent with the syntax and semantics of some language. The early brain doesn't have any strong predictions about the shape of its future mental content, so it's free to learn; but as it ages, it gets accustomed to what patterns it's happened to carve out, and if there was nothing shaping these patterns so that they'd be expressible via some language, the brain will have learned to predict mental content which is not congruent with language. In trying to teach language in a mind that ossifies without having learned a language, the first new predictions it will make do not involve the kinds of conceptive relations that are rendered as grammatical vocabulary, but the much simpler relation of sense-formations already considered as 'things' to strings of phonemes—thus, someone who acquires language past the critical period (after their brain's already found its groove) will acquire lots of material vocabulary before they learn the deeper kinds of organizations required to use grammatical vocabulary, while someone acquiring language during the critical period will acquire both kinds of vocabulary together.

Yokai Criticality

largephysicsmetatheory

Japanese yokai are often the aggrieved spirits of people killed in terrible ways, and lots of them then go on to kill people in terrible ways Sometimes you hear about how some folk story popular in America (e.g. Brothers Grimm/Disney tales) actually has some disturbing origin involving murder and/or cannibalism—but Japanese folklore often seems to be unapologetically, brutally horrific; you don't even need to peel back several layers of whitewashing. Sometimes you'll see an abandoned infant crying, and if you try to pick it up it'll turn into a stone and crush you. Sometimes children's blood-red hands will entrance you into trees and rip you apart limb from limb. If you weren't wearing clothes when you died, the old lady that strips your clothes in the afterlife will just tear off your skin. Don't confuse this with the other old lady who asks for sake candy but gives you smallpox whether you comply or not. And you don't have to be a princess or an adventurer to risk such a horrible, hopeless fate—it can happen anywhere at any time!; this is obviously a disaster waiting to happen. What you'd end up having in real life is a sort of natural selection effect, where those yokai of a given type who have MOs that tend to induct their victims into that same type would proliferate moreso than those types that either tend towards passivity or don't tend to produce murder victims of their same type (of course you could have generational loops and all other sorts of effects, but this is the basic dynamic). And there's no a priori reason to believe that these eigenyokai would end up having a reactivity of less than one—that there won't eventually be a form the average member of which produces 1.05 members of that same form, say. The entire island of Japan would be a criticality accident waiting to happen. (Actually, it's even worse—it's not like yokai can only kill so many people before they disappear. The neutron that splits a plutonium atom is absorbed in the process; but if neutrons could just keep going, they'd grow as 1, 2, 4, 8, ..., even with a reactivity of exactly one. But I'm ignoring this factor for now).

They tend not to disappear until they're finally relieved by some ritual that someone stumbles upon accidentally, and other than that they're very difficult to avoid... and, of course, natural selection would filter for those yokai that tend to not really go away or be satisfied. Without the equivalent of some sort of SCP foundation finding and exorcising or otherwise containing the more threatening yokai, the entirety of Japan would go prompt critical, becoming an uninhabitable no man's land of malevolent spirits. Natural selection in animals happens from generation to generation, so, if the form of yokai one could become were genetically determined, it might take a hundred generations for them to take over—but it's well-known that the form one takes can depend on how they died! So, even if every yokai is restricted to one murder between midnight and 4 AM, you'd only need a hundred days rather than generations. (Of course, this depends on the fact that Japan already has a large human population for the yokai to rip through).

What would history look like if this had happened at some point? Say, over the course of a couple months in the year 1536, the yokai go critical and Japan becomes a ghost nation. A couple years later, Portuguese explorers arrive at Tanegashima to find abandoned houses, overrun farms, and maybe a few bands of highly militarized shrine maidens. They're wiped out after only a few nights, just like all the Chinese and eventually Dutch trade ships; Japan becomes a black hole in the Pacific, and those lost explorers and traders who end up there by accident quickly try to sail away—but their ships aren't fast enough for them to make it even to China before they're wiped out. Pirates in the East China sea occasionally discover ghost ships floating with free cargo—those pirates that board ghost ships end up becoming ghosts themselves, and eventually sailors learn to stay away from the ever-growing clusters of abandoned ships. Eventually, some enterprising American figures that it's probably superstition, and if not then in any case it's nothing his steamboat's top-of-the-line cannons won't tear to shreds. After a series of inexplicable, brutal deaths on the first night, he hauls ass to China; his ship is just barely fast enough to make it to port, and the world ends shortly thereafter.

There's a more general principle at play here—when beings with abilities scattered around a human-constructed, conceptual latent space of possible abilities are given to appear in a world, it's only a matter of time before that world blows up. (And even more generally: a world that needs to be saved is a world that is already doomed! The seed of this doom lies in its tendency to create beings, or phenomena, that can threaten the world at all; the abilities of these beings must be generated by an underlying dynamics, and these dynamics must be prone to the usual physical and metatheoretical interactions; there are always going to be cheese strategies, appearing both through adversarial optimization and natural feedback loops).

Metatheoretic Notation

metatheory

It seems mostly unavoidable that you have to have at least two levels of being at which things can exist in order to think about them formally at all. Elements of sets. Members of types. Objects of categories. So I'm going to make a distinction between minor and major types of objects; major objects provide contexts in which minor objects can exist. We'll think of objects as having determinations, or specific modes of being, in a manner beyond any two-level system, but if they are given levels, then these are part of its determination.

1. (Bullets): $\bullet, \bullet', \ldots$ represent arbitrary elements (minor objects) with determinate major objects (i.e., they are typed, even if we don't write the type). Also, (univocality property) all instances of a $\bullet$ in a given utterance refer to the same object. So $F(X_\bullet)=\bullet$ can also be written as $F(X_\alpha) = \alpha$, but not as $F(X_\alpha)=\beta$ (unless $\alpha=\beta$).
2. (Stars) $*, *', \ldots$ represent arbitrary types (major objects). This notation is also univocalic, and furthermore if we include both a $\bullet$ and a $*$ (with the same number of primes) in a statement, we think of the $*$ as being the type of $\bullet$, but e.g. $*''$ isn't necessarily the type of $\bullet'$.
3. (Dashes): $-$ represents an arbitrary object, whether minor or major. This might be treated as univocalic with respect to level, but is always multivocalic in general—two instances of $-$ in the same statement may not refer to the same object, but are on the same level.
4. (Indeterminates): We use $?$ for arbitrary objects, this convention being entirely multivocalic. It's like when a Unicode symbol gets represented as a block—it could be anything, irrelevant of what other blocks there might be. So while we can't write a membership relation $\in$ as dyadic like $\in(-, -)$, since the two arguments are necessarily on different levels, we can write it like $\in(?,?)$.
5. We write $\underset a X$ when $a$ is minor subordinate to $X$ (e.g., an element of $X$) and the major object $X$ is the relevant subject; when the relevant subject is the minor object, we write $\overset X a$. Ex., $X$ needs to be main-aligned rather than top-aligned when writing a function type with domain $X$ — it's the relevant subject in the function type, so we write $\underset a X \to Y$ rather than $\overset Xa \to Y$. Ex., $a$ needs to be main-aligned when given as an argument to a function $f$ with domain $Y$ containing $X$, so we write $f(\overset Xa)$ — that $a \in Y$ is assumed when we give it as an argument to $f$, but the overset to $a$ indicates that furthermore $a \in X$.
6. $\circ, \circ', \ldots$ represent slots for elements (or, when indicated by priming, sets), e.g. $X_\circ$ is not a set but an indexed set. Each one is filled by the corresponding $\bullet, \bullet', \ldots$. Ex., $(X_\circ \times Y_{\circ'})_\bullet$ is the indexed set $X_\bullet \times Y_{\circ'}$. Ex., $(\circ \to Y_{\circ '_\circ})_{*, \bullet'} = * \to Y_{\bullet'_*}$ can instantiate like $X \to Y_{f_X}$. Ex., $X_\circ \times Y_{\circ'}$ is really a function $\underset\bullet\star \times \underset{\bullet'}\star' \to {\cal U}$.
7. The horizontal binding priority order is, from strongest to weakest: $\times, +, \to, \equiv$. Vertically, boxes may be used to specify the order of notation in ambiguous situations.
8. Partial functions are denoted with $\nrightarrow$, and operationalized as $X \nrightarrow Y \equiv X \to Y + \{\bot\}$.
9. The universe is $\cal U$. It's a doubly major object in a sense, but one should avoid using it where possible (consider how we have to speak of the 'class' of all sets, or the 'large category' of all small categories—and how this inevitably causes us pain).

Idioms: Given this notation,

• $\bullet_{\circ}'$ is a function from $\underset\bullet\star$ to $\underset{\bullet'}\star'$, and $\displaystyle\bullet_{\circ^{'}}$ is a function in the opposite direction. They form an isomorphism when $\large \bullet'_{\bullet_{\bullet'_2}} = \bullet'_2$ and $\large \bullet_{\bullet'_{\bullet_2}}=\bullet_2$.
• Also, $\underset\circ X \to Y_\circ$ is a dependent function type, equivalent to $\prod_{\bullet:X}Y_\bullet$.
• If $\overset{X}{\bullet_{\bullet'}}$, $\bullet_\circ$
• $\overset \star\bullet$ instantiates in its very usage an element $\bullet$ of type $\star$. $\star_{\overset{\star'}{\bullet}}$
• A 'list' $L$ ought to be treatable as a function $\underset\circ{\mathbb{N}}\nrightarrow X_\circ$. To turn $n \in {\mathbb N}$ into the list $L_n = [0, 1, \ldots, n-1]$, we ought to append $n-1$ to $L_{n-1}$.

Notes:

• When using words as symbols, a bullet can be postscripted to indicate univocality — essentially assigning the semance of the word to the bullet. The sentence "kick a $\text{dog}^{\bullet}$ and a $\text{dog}^\bullet$ will bite you" proposes that it is one and the same dog that is kicked and that bites you, whereas the sentence "kick a $\text{dog}^{\bullet}$ and a $\text{dog}^{\bullet'}$ " will bite you proposes that, while you will get bit by a dog upon kicking a dog, the dog that bites you is not necessarily the same as the dog that you kick.
• The $\overset{\text{symbol}}{\text{property}}$ of being univocal is univocality, whereas the $\overset{\text{convention}^{\bullet}}{\text{property}}$ of having univocal $\overset{\text{convention}^\bullet}{\text{symbols}}$ is univocalicityThe notation in this sentence is saying that the property of being univocal is something that belongs to a symbol, whereas the property of being univocalic is something that belongs to a convention, and applies to symbols of that convention. . Having univocality makes something univocal, whereas having univocalicity makes something univocalic.
• Treating -al, -ality as equivalent to -ic, -icity; -icicity and -alality are awkward, so -icality and -alicity are better. Need to use univocality instead of univocity because we say that an X with this property is univocal, not univocic (and univocity is irregular: the c should not be rolled into a -city prefix, because it's part of the -voc- morpheme).
• nondependent = not necessarily relying on something; to say that $x$ and $y$ are independent is to postulate a positive property of them, while to say that they are nondependent is merely to say that we don't conceptualize them as dependent (regardless of whether we conceptualize them as independent or not). when both $P$ and $\neg P$ are determinate properties of something, positive qualities that do not exist in the concept of that thing a priori (i.e., it is possible to conceptualize that thing without remarking on whether it has $P$ or $\neg P$), it is useful to have a way to say that $P$, or $\neg P$, is not part of any particular specification of that concept. To say that $P$ is not conceptualized with the thing is not to say that $\neg P$ is conceptualized; in a real instantiation, $P$ may very well obtain — we just kept agnostic about it.

Phonological Relativity

languagephenomenology

In English, the onomatopoeia for a heartbeat is *lub-dub* or *ba-dum*, while in Japanese it's *doki*. They sound very different, qualitatively—Japanese is a very rhythmic language where most consonants are articulated via sharp tongue movements, whereas English has lots of drawn out syllables and articulations often involve coordination between the tongue and the much slower lips. So, in music terms, the moras in *do-ki* have a very fast attack, while the syllables in *ba-dum* have a slow one, which makes it feel like the Japanese onomatopoeia is comparing the sound of a heartbeat to, say, rapping on a door, while English is comparing it to thumping on a door.

But neither answer seems to be "right": if I listen to the sound of my own heartbeat, it seems like I can decide to mentally interpret the sound either way, by focusing on the onset or on the body of the sound—essentially like a Necker cube, but for audio. This makes me suspect that the phonologies of our languages help to determine the ways in which we perceive sounds—as in, even if the onomatopoeias *doki* and *ba-dum* were never invented, Japanese and English speakers would still parse the sound of a heartbeat in slightly different ways because the faculty of phonological hearing kind of just happens to get mixed with that of non-phonological hearing.

Probably a more explicit example is with the sound cats make: in English, it's *meow*, which ends with a contraction of the lips, and in Japanese, it's *nyan*, which doesn't. There's definitely some part of my head that experiences hearing the phonemes *meow* whenever I hear a cat meow, and some evidence of this might be found in how people might say "meow" in English phonology without really thinking about it whenever they hear a cat meow (as in echolalia, but lots of people just sometimes do it absent-mindedly). So it should follow that English speakers parse the sound of a meow in a very slight but fundamentally different way than Japanese speakers.

Is this sort of 'phonological relativity' a well-known fact among linguists or psychologists? Are there any noteworthy explorations of it, or rigorously gathered data that would allow for a scientific study of the phenomenon?

Things GPT Won't Be Good At

AI

I've been playing around with various interfaces for using GPT, all of which are basically wrappers for the API, and there's a lot to be learned about what GPT must be like based on what the API allows. Basically, you set up an array of individual messages, each with a role—System, User, or Assistant, and then send that array in order to get a new message, the completion. Normatively, the 'system prompt' is the first message in the array, tagged with the System role, and the next message is the user's prompt (e.g. a query or request); each completion is tagged with the Assistant role and appended to this array, the user's reply follows, and repeat. It doesn't have to be this way, you can do anything you want I find this useful for text completion: write the text I want completed in the Assistant role and just send it incomplete—the next message will be the actual assistant continuing the text past a seemingly incidental end of text token. , but consider that this is what's going on behind an interface like ChatGPT. There is no place for a hidden state that is not shared with the user, since the AI is only going off of whatever messages it's received to decide what to say next. 'Memory' can be nothing but an ability to form responses with a long range of dependency. GPT is pretty good at this (that's what attention is about), but anything 'remembered' is only located in the message array provided to it, not in any internal state of the model. As can be inferred from the structure of the API, nothing gets 'baked in' the AI's mind over the course of one's conversation with it.

When We Create Physics

quantumconsciousnessKantlarge

There's one point Kant keeps bringing up throughout the CPR: the way in which we create physics. He's significantly more attuned to the consequences of being made of the world we're trying to cognize than I first realized, since I had

Given, the way he attempts to approach this is totally wrong. See A111-A115 (I'm using Pluhar's translation): to explain why we should register the operation of physical principles underlying the objects of our experience, he says that nature is only nature as we experience it, and in this capacity must be cognized in some particuar universal ways if it is to act as the sum total of all other possible experience.

Hence the order and regularity in the appearances that we call nature are brought into them by ourselves; nor indeed could such order and regularity be found in appearances, had not we, or the nature of our mind, put them into appearances originally. For this unity of nature is to be a necessary, i.e., an a priori certain, unity of the connection of appearances. But how indeed could we have the ability to institute a priori a synthetic unity, if our mind's original cognitive sources did not a priori contain subjective bases of such unity, and if these subjective conditions were not at the same time valid objectively, viz., by being the bases for the possibility of cognizing an object in experience at all? [...] Hence, all appearances, insofar as they are possible experiences, lie a priori in the understanding and obtain from it their formal possibility; just as, insofar as they are mere intuitions, they lie in sensibility and are, in terms of their form, possible solely through it.

A subtler account is given eight years later in B164:

How it is that the laws of appearances in nature must agree with the understanding and its a priori form, i.e., with the understanding's power to combine the manifold as such, is not any stranger than how it is that appearances themselves must agree with the form of a priori sensible intuition. For just as appearances exist not in themselves but only relatively to the subject in whom the appearances inhere insofar as the subject has senses, so the laws exist not in the appearances but only relatively to that same being insofar as that being has understanding. Things in themselves would have their law-governedness necessarily, even apart from an understanding that cognizes them. But appearances are only presentations of things that exist uncognized as regards what they may be in themselves. As mere appearances, however, they are subject to no law of connection whatever except the one prescribed by the connecting power.

But think of how this might actually apply to, say, tracking a comet. We point our telescopes at the sky and get 'appearances' recorded as collections of numbers: time, azimuth, zenith, altitude, time, azimuth, zenith, altitude... And we get to work building models. We constantly find we make mistakes, both in our models and in our numbers, so that where they cohere we don't see it at all and vice-versa. In the end, our model works out and we're able to figure out the trajectory of the comet, marking another win for Newton!

Except we're not and can never really use Newton's laws correctly. Drawing two dots and a ring, we represent the orbit of the Earth around the sun; another dot and an ellipse gives us the comet, and the sun is treated as the only source of gravity. But every planet and asteroid and dust speck in the comet's past light cone universe is contributing its own $GM/r^2$. Our whole show of modeling the comet is fundamentally an approximation.

And Newtonian mechanics isn't even right. The general relativistic corrections to the comet's behavior might be smaller than even the experimental error of our measurements, but it's still a factor that we have to consider everywhere in order to use anywhere, and we're not doing that. Our laws are incorrect, our application is incorrect—but the comet still follows the path we traced out for it. Nevertheless, while we invariably reason fuzzily about fuzzy things, Kant's argument does apply, I think, when it comes to quantum indeterminism.

Quantum

Say you have an evolving quantum mechanical system in a starting state $|\psi(0)\rangle$, and you want to compute the probability of measuring $\psi(T)=x$. Were $|\psi(0)\rangle$ just a probability density over the position of a point particle, you could find the likelihood of measuring $\psi(T)=x$ by mapping out all possible paths by which the particle could travel from $0$ to $x$ in time $T$ and adding up the probability of all of them. Since it's a continuous space of paths, you have to integrate over the space of paths $\phi$ (since the infinitesimal $d\phi$ is scary and not well-defined, we write it as ${\cal D}\phi$): $$P(\psi(T)=x) = \int P(\psi(T)=x\mid\psi\text{ takes path } \phi)P(\psi\text{ takes path } \phi)\,{\cal D}\phi = \int_{\text{all }\phi(0)=\psi_0,\phi(T)=x} Q[\phi]\,{\cal D}\phi$$ Since the contribution from each path $\phi$ is a non-negative probability proportional to the inherent probability of the path given the dynamics of that theory, everything is nice and simple: the most likely paths contribute the largest probabilities, and, the smaller whatever noise or uncertainty makes $\psi$ probabilistic in the first place, the greater the proportional contribution of these paths will be.

But $|\psi(t)\rangle$ isn't a probability density, it's a wavefunction (and we have to apply a $|-|^2$ to each integral). Were it a probability density, each contribution $Q[\phi]$ in our heuristic path integral above would, being a non-negative real, stand by itself, interacting with no other contribution when it's added to the final probability. But with wavefunctions, $Q[\phi]$ is not a non-negative real number, but a unit complex number (an arrow pointing in some direction), whose squared modulus is a probability. We don't calculate it by starting from a probability of 1, which we lower the crazier $\phi$ gets, but by starting from a phase of 0 and increasing the phase the more dynamic $\phi$ gets (this is why we need the $i$ — the Schrodinger would be a basic diffusion equation without it, and the path integral would be one of a statistical mechanical system, but using the $i$ allows different 'virtual' solutions to interfere). When performing the integral over paths, differently-phased ways to get to $x$ will interfere with one another: the most likely path to $x$ in the classical system might yield a phase which is cancelled out via destructive interference with many other classically unlikely paths to $x$, and a very unlikely $x$ in the classical system might be made very likely in the quantum system if the paths to it tend to constructively interfere. It's fun to work out a simplified version of this in the double slit experiment byA photon with wavelength $\lambda$ has energy $E=hf=\frac{2\pi \hbar c}{\lambda}$, all kinetic, so its phase is $e^{\frac{i}{\hbar}\int_0^T\frac{2\pi \hbar c}{\lambda}\,dt} = e^{2\pi i \frac{cT}{\lambda}}$. $e^{2\pi i}$ is just a full rotation; the number of full rotations, $\frac{cT}{\lambda}$, is the number of wavelengths it's traveled. So two photons that, starting at the same point with the same wavelength, take paths $\phi_1(t)$ and $\phi_2(t)$ to get to a point $x$ at time $T$ constructively interfere if the difference between the lengths of their traveled paths is closer to $n\lambda$ for some $n$, destructively interfere if the difference is closer to $\left(n+\frac12\right)\lambda$. [didn't want to find and type up my calculations].

Now, since we're working with phases, $Q[\phi]$ is—assuming that contributions are temporally localized (!), e.g., if $t_1^l < t_1 < t_1^u < t_2^l < t_2 < t_2^u$, then the contribution by $\phi$ over $(t_2^l, t_2^u)$ is independent of the contribution over $(t_1^l, t_1^u)$ conditional on its actual behavior over $(t_2^l, t_2^u)$—necessarily gonna look like $e^{i\int_0^T ?(\phi(t))\, dt}$, where the $?$ in the integral is what turns the local behavior of $\phi$ into the spinning of the phase. It turns out to be the Lagrangian $L(\phi(t), \dot \phi(t))$ divided by Planck's constant $\hbar$: $$P(\psi(T)=x) \propto \left|\int_{\phi(0)=\ldots} e^{\frac{i}{\hbar} \int_0^T L(\phi(t),\dot\phi(t))}\,{\cal D}\phi\right|^2 = \left|\int_{\ldots}\exp\left(\frac{i}{\hbar}S[\phi]\right)\,{\cal D}\phi\right|^2$$ where $Q[\phi] = S[\phi]$ is the classical action functional. (The normalization constant is given by integrating over the $x$ hidden in the $\ldots$).

Since the Lagrangian essentially tells the system how to evolve, and we're integrating over every possible path the system might take, it makes sense to think of all of these paths as being "taken simultaneously" in a virtual reality, their contributions being added to get a this-ness, or propensity to be where you find yourself, for a given state (a fixed endpoint of many paths). There are other ways to think about it, presumably. But this perspective makes it easy to understand superposition, interference, entanglement, 'collapse' (einselection), and most other aspects of quantum.

Mental states supervene on the physical, so you will always have a unique past, present, and future to yourself. If all of a sudden a different timeline enters the same state as your present timeline, so that they interfere, which one will you remember being in while in that state? Unfit question: you will remember whatever it is the physical state of your brain causes you to remember, which since the two timelines have entered the same state is the same either way. Because of this supervenience, which appears to make consciousness necessarily singular, the structure of reality determines the manner through which you can experience anything at all.

In fact, the fact of our classical experience of the world may just be a feature of the way we're embedded into the world. When we take the integral over paths, we're taking the integral over all paths, including not only those that obey inertia and other classical principles ('on-shell' paths) but those that don't (off-shell); no distinction is made between them, and in particular there's nothing causing classical mechanics to happen. That it happens anyway comes from the statistics of interference. Say you're in a room, the walls of which form one giant screen encircling you; a thin vertical line travels counterclockwise around this screen, and some unseen force pulls you towards it with constant, irresistible strength. If the line moves very slowly, you'll be pulled towards its general direction until you crash into the wall, but if the line moves very quickly, it won't really have a general direction, and you'll be slightly jerked around while ultimately remaining in the center. In the same way, a region of the path space where the action changes rapidly won't be able to make a contribution to the amplitude of a particular state, since that region can't even point in a general direction; but a region where the action changes slowly does point in a general direction, and therefore does contribute to amplitudes. If you look back at the path integral formula, you'll see that the phase is actually the action multiplied by $1/\hbar$, which amounts to a speed multiplier of $10^{34}$—the effect of this is that regions of the path space can't contribute to amplitudes unless the action is essentially constant over them; conversely, particular states can't have high amplitudes unless there are large regions of constant-action paths that all lead to them. Thus, the smallness of $\hbar$ allows the path integral formulation to approximately reproduce the principle of stationary action (which directly leads to the Euler-Lagrange equations and classical mechanics).

So, while basically no physical laws are given in the path integral formalism, we idiosyncratically see a deterministic classical mechanics owing to the shape and scale of our conscious experience. To quote Kant again, B350:

Above, we called dialectic as such a logic of illusion. This does not mean that it is a doctrine of probability. For probability is truth, but truth cognized through insufficient bases; and although cognition of such truth is therefore deficient, yet it is not on that account deceptive, and hence must be separated from the analytic part of logic.

(Essentially, all brains are decoherent and —
as Asher Peres says, "nobody has ever seen a cat in such a strange situation".

The Delayed Choice Quantum Eraser: Fire a photon through a double slit and then separate it into two entangled photons, one going east and the other going west, making sure to preserve the phase difference in the choice of slit. Send East towards a screen, and record the point $x$ it hits—if we can't tell which slit it came through, we'll find an interference pattern when we repeat this experiment half a million times and plot the distribution of all the $x$ valuesIn other words, each $x$ will be sampled from a probability density with an interference pattern., but, if we can tell, the $x$ values will form an ordinary gradient. West could be sent to detectors that tell us which slit it came through; it shares this slit with East, due to entanglement. Have West bounce around for a while, and then, after East has already hit the screen, send it to a splitter, one path going towards a detector which messages us "Found a photon from [this] slit!", and the other path going towards a detector that just messages us "Found a photon!". (The idea is that the second detector "erases" the information about which path was taken, while the first doesn't). If the first detector beeps, save the screen location to known.txt, but if the second detector beeps, save the location to erased.txt.
999,999 trials left to go. Because the splitter sends West to each detector with equal probability, each of known.txt and erased.txt will have around half a million screen locations. What do you get when you plot them? The locations in known.txt will form an ordinary gradient, but those in erased.txt will form an interference pattern.
The paradox comes when we want to think that, by the time we detect West, East has already hit the screen, so that the distribution it follows should not change based on whether or not the slit information is erased afterwards. But interference happens across timelines. c.f. https://physics.stackexchange.com/a/35049

Maybe quantum phenomena can, if coherence ever happens to obtain at our scale, actually screw with the play of mind; how could we tell if so? The closest we've ever actually gotten to any possible coherent scenario might be Apollo 13's crossing of the dark side of the moon: with no physical contact and no radio contact, the ship was almost an entirely closed system. There was some causal influence, e.g. gravitational, but nothing actually measured—and, as Asher Peres also says, "unperformed experiments have no results"Scott Aaronson begins a post with the same quote, in which Hardy's paradox is demonstrated as a . But even if they were coherent, quantum indeterminacy would still hardly affect them—taken as a system in themselves, more or less everything we'd observe was already classically determined the by the time radio contact was lost. If you hook a decaying atom up to a gun, as with Schrodinger's cat, then, sure, quantum indeterminacy will have a huge effect; the most we could expect to happen upon Apollo 13 is an extraordinarily rare chance that a gamma ray indeterministically flipped a bit on the onboard computer, which would require another extremely rare chance to actually affect anythingBut maybe the spacecraft was, classically, absolutely doomed when it lost radio contact—it's just that the loss of the astronauts would consistently trigger a nuclear apocalypse, and we're therefore only here because such unlikely quantum effects did save them. or not....

Chromaclear

phenomenologyterminologyconcepts

There's one facet of visual experience that I come across all the time, and which is a near-canonical example of what I mean when I speak of daimon logic. Say you've been coughing or laughing really hard, such that you end up seeing stars—the visual experience will be of tiny specks of light drifting across your vision, almost like blue-sky sprites but longer-lasting and less Brownian. They'll have the appearance of having color, but they won't be any specific color; you can try to focus on them to discern their color, but they'll generally either take on a single color in response to this (as though you were collapsing a superposition), or rapidly flicker between different color perceptions. There is no rate at which they flicker, either; you can make the flickering take on a certain rate if you ask "what rate is this?", but it doesn't intrinsically have one prior to measurement It's kind of quantum-like, but not entirely. You can't say that there's any sort of wavefunction which gives rise to a probability distribution over possible rates, since the notion of such a probability distribution is an ill-posed conceptual construct. The final rate one perceives is determined by lots of pre-existing factors but also by the specifics of the method of measurement (which is not a single event, but a neuropsychological process).. I call light that has this appearance of color that is not any particular color chromaclear light. Once you really get a handle on how to see it, you can see it all the time. There's almost an inherent chromaclarity to all perceived color, even—as though it were produced by the brain reaching out for a color to perceive before constructing the visual experience of a purple desktop background or a white wall.

Relatedness in Conceptspace

metatheoryconcepts

From a quantitative point of view, the functions $f(x)$ and $-f(x)$ are opposites. From a conceptual point of view, they're also oppositesIn the particular sense that they send a single element to inverses in the codomain. Of course we could have $f(x)$ and $f^{-1}(x)$ when $f$ is a bijection, $f(x)$ and $f(x)^{-1}$ when the codomain is a field, and so on. The below applies to all of these.. But from a conceptive, or metaconceptual, point of view, they're almost exactly the same: it's very, very cheap to apply the concept of negation to a function. It's almost automatic to take a daimon (applicative conceptual construct) and construct an "opposing" daimon, but the first one has to be there: if there's a vague cluster of interrelated concepts shared by some people, the most one can do to oppose 'it' is to get a vague sense of badness from utterances that match the concepts, especially when they involve those people; but if they decide to reify this vague cluster into a notion like "X-ism", one can (with a bit of effort) reify an oppositional daimon that doesn't try to grab on to a vague haze but to this now conceptively (perhaps not conceptually) clear notion. Once the conceptual sculpting has been done, new things can be built that don't have to re-sculpt the concept (to be supported, to be opposed) but can instead just point to it, like from conceptual import concept; from metaconceptual import oppose; oppose(concept);. Similarly, when a notion of utility is optimized for, the optimizing process will end up creating a lot of piping, and insofar as this piping can be used to generally affect the thing being optimized for in a way that doesn't necessitate its optimization per se, it can be turned around: from piping import optimizer; from metaconceptual import oppose; optimizer.value_perception = lambda self, *args: oppose(optimizer.value_perception(self, *args)); optimizer.optimize() (who knows if this is valid Python -- modulo the semicolons). And note that the daimon constructed by opposing $f$ doesn't have to be the same as the daimon constructed by negating $f$ — if I hate X-ism, I might create a daimon that is satisfied by "this is not X-ist" rather than "this is the complete opposite of X-ism".

So, conceptively, $f$ and $-f$ are neighbors. This notion of relatedness in conceptspace extends very far beyond mere negation, though: you can think of it like a Zariski topology on conceptspace (This is more or less the same application of the idea of a Zariski topology as seen in the dialogue in Dreaming of Utility). , relating things based not on how alike they are per se but on the extent to which building one makes it easy to build the other. Another way to think of it is as a conditional Kolmogorov complexity: $f$ may be extremely complex, which by necessity makes $-f$ extremely complex as well, but, if we internalize $f$, making it a native of whatever language underpins our K-complexity, then suddenly the K-complexity of $-f$ — I want to write it as $K(-f\mid f)$ — drops to near zero. An entirely unrelated function $g$ may end up having a pretty high correlation with $f$, but, since its inner workings are entirely different and $f-g$ isn't particularly informative of anything, $K(g \mid f)\approx K(g)$.

Another way to think of the Zariski-ness of the world is like... the ubiquity of causal add-ons to simple things. I once saw an attempted refutation of anthropic reasoning in cosmology that went something like: humans only need 25 elements to live, but we find that there are 92 elements that exist in nature; by Occam's razor, anthropic reasoning favors those systems of physical law in which only those elements we need are produced, so it must be wrong since there are in fact 67 extra elements. But how do you know it's not the case that the simplest systems of physical law which produce the 25 elements that we need don't by the same processes incidentally produce all the other elements? After all, elements that exist just correspond to numbers of protons that can be (more or less) stably packed into an atom, and there's no reason for any tuning of physical parameters that lets those 25 exist stably to stop right there, rather than incidentally having some stable configurations for 26-proton atoms, 27-proton atoms, and so on...

Here's yet another way to think about it: suppose I purchase an infinite row of squares with '0' penciled in each, and I start erasing some and replacing them with 1s according to some pattern. You end up taking a peek at a small section of it, which goes: "XXXX101010101XXXX" (with X being an obscured square). Do you conclude that the pattern I'm trying to writeThat is, the potentially-infinite pattern that my finite completion is converging to. (Convergence of a finite pattern generated by !c{some algorithm}{in this case, whatever it is I'm doing} to some infinite pattern is precisely what makes that infinite pattern computable). goes "10101010101010101" on the rationale that obviously I'm implementing the very simple rule "alternating 1s and 0s", or do you conclude it goes "00001010101010000", sagely taking Occam's advice and not needlessly multiplying entities in the defense of a hypothesis with absolutely no empirical evidence? If you're smart, you'll do the Solomonoff thing and go with the former hypothesis. Even if it has zero empirical evidence—just because you can't see something doesn't mean it isn't there.

The Werewolf Problem (Deorthogonalization)

intelligencedecisionmetatheoryAIlarge

The basic objection to most objections against utilitarianism is that they're really objections to stupid utilitarianismThat stupidity often contaminates 'consequentialists' (generally just people who find consequentialism a good cloak to justify whatever actions they were already going to do) in general is a utilitarian reason not to promote utilitarianism indiscriminately, but it isn't itself an argument against utilitarianism, just about having the right influence on other actors: do not implant in someone's head a seed that will cause them to do less good than they would've otherwise done. Because most people are stupid, having the right influence generally boils down to not promoting utilitarianism uncarefully. Those who don't reflexively consider higher-order or long-term consequences in their consequentialism—who don't actually try to assess the actual consequences— should stick to being normal. (edit: found an article which makes the point much more articulately: Consequentialism Need Not Be Nearsighted).. If something makes the world worse off "in the long run", then it is not the utilitarian action now. If you justify something as "for the greatest good" but don't attempt to find those actions that actually produce the greatest good on expectation, you are failing at utilitarianism. If you do not try to stick to actually producing actual good in actual reality, but get lost in your idealized worlds, you are failing at utilitarianism. If you think that some areas of action, like speech or entertainment, are independent of utilitarianism, and act accordingly, you are failing at utilitarianism. If you think the best you can do is approximate utilitarianism, and therefore cut yourself some slack, but don't try to actively fix and eventually eliminate that problem, you are failing at utilitarianism. A single actually competent utilitarian would, if placed into this world, upend it by the end of the year.

A (conceptively cognizing, e.g. human) utilitarian ought to have a conceptual distinction between that which is utility to them, the concept of utility in general, and the concept of goodness in general—they should be able to update their beliefs about utilitarianism and what is utilitarian. But insofar as they're plug-and-play, their notion of utility can be made essentially arbitrary; in fact, if they're any good at optimizing, they likely have machinery (whether conceptual, biological, social, or literal) such as allows them to optimize a sign-flipped utility This sidenote originally contained what is now posted directly above as 'Relatedness in Conceptspace'. . A doctor knows how to efficiently kill; a detective knows how to get away with it.

This poses a challenge for non-Cartesian utilitarians: how can they be such that the damage of signjacking is minimized without destroying their existing capability? (This makes more sense for AIs than humans, but still makes a lot of sense for humans—or, rather, de dicto humans: social roles, personas, identities. The President is always a human with skin and lungs and a social security number, but takes on the role of "the President"; whoever they are, they don't want someone that takes them out and impersonates them to be able to launch nukes, but they still want to be able to launch nukes if necessary. And humans don't have a single utility function, but lots of little sub-utility functions that integrate in weird, attention-dependent ways: the President also doesn't want their replacement (who will also be the President, but a different human) to be able to undo all their wonderful partisan acts. It's not the individual humans (persons de re), but just "the President" (person de dicto), that gets sub-signflipped every four to eight years). A Cartesian utilitarian black-boxes their levers of thought and action, and therefore does not conceive of such a possibility, but a non-Cartesian utilitarian is aware of how they do what they do, and is therefore able to think of how it might be mis-utilized.

But, for now, let's consider the case of a utilitarian who is confident in their notion of utility, but who is threatened by some particular mechanism that threatens to spontaneously turn it into something else—not necessarily a sign-flip, but something they really don't want in any case. The classic example is—if you might've been bitten by a werewolf recently, and a full moon's out tonight, how should you act? Self-control isn't enough—you can be assured that if you transform, were!you will have an insatiable hunger for humans, and your thoughts will bend towards this in order to justify and act on it. Probably the transformation is continuous: you're up late programming when your mind wanders towards the thought of searching for people; half an hour later, you start to feel a bit hungry, and your arm hairs seem a bit dense; and so on and so on. But who knows whether the transformation takes place over ten minutes or three hours? it's not like you've gone out and interviewed a werewolf, and you haven't experienced it before. You can open up Tor and search "werewolf transformation symptoms experience report", but the only mildly informative result is the CDC page on lycanthropism, which indicates symptoms (hirsutism, mania, acromegaly—yeah, no shit) and points you towards a hotline you can call if you suspect that someone's been bitten—and who knows what sort of lab or prison they'll throw you into if you test positive... So what do you do? As a start, let's consider the problem of minimizing ${\mathbb{E}}[$loss$\mid$werewolf$]p($werewolf$)+{\mathbb E}[$loss$\mid$not werewolf$]p($not werewolf$)$. If you're selfish, you only care about what you lose out on—if were!you eats a family of four, that's their problem—but you'd probably get hunted down and detained shortly after, so it's best to just hide away. So,

That strayed away from the question I want to ask. (Continuing from "[...] something they really don't want in any case."). An agent who has zero concerns about such possibilities, who is certain that they will continue to act in line with their current, 'default' notion of utility, might just amass power arbitrarily and keep it ready at hand, keeping their finger on the metaphorical trigger. But this is often an unrealistic expectation, and such concerns are, where present, generic reasons to set up one's environment—to configure what causal articulations underlie the levers of one's power—so as to make it inconvenient to act against one's notion of utility. A lot depends on the threat model here: the extent to which one thinks th
I think this is critical in considering how instrumentality plays out—another "key", in the same sense that the possibility of dragon kingsor black swans? who knows—what I mean is 'outliers with disproportionately heavy-tailed severity' (the whole log-normality discussion in Worldspace) is a key. So, extremely important in predicting how future A(G)Is might behave, but I haven't yet found a good cognitive structure for thinking about this problem, in the same way I found a good structure for thinking about how AIs might tailor their actions instrumentally in response to the outliers problem. There are a couple of natural-ish ways of intuiting the werewolf problem, though:

Update: I came across this clip from some DC cartoon in which Batman outlines contingency plans for taking out each member of the Justice League should they go rogue. They're not very good plans You can tell just from the structure of the plans—hyperspecific strategies that try to think of some possible conceptual weakness and some clever way to exploit it. Just because Elastic Man is stretchy doesn't mean you have to beat him by countering his stretchiness. You have to account for it, but not base your thinking on it. Use a missile.
(A cryogenic aerosol has a good chance of failing in field application, whether something about Elastic Man's biology nullifies it, because he can find a way to protect himself, because he can defeat or evade the pursuer deploying it, and so on. Developing and testing a reliable custom applicator might take millions of dollars and months in R&D. On the other hand, the AGM-114 Hellfire is an air-to-ground missile, equippable with a !c{thermobaric}{he might stretch to survive a kinetic or high-explosive warhead, but not a thermobaric one} explosive, that, once fired from up to over five miles away, reliably homes in on its target at speeds of up to 1000 mph; it costs $150k per unit, and an AH-64 can lift off with sixteen of them by the hour. Good luck, Elastic Man).
If you actually want to win, you'll probably be using entirely general strategies which can be hardened in general and adapted to particulars—stockpile air-to-ground missiles and sniper rifles, !c{fentanyl}{a nonlethal option in case Elastic Man takes hostages; hopefully we'd be more competent than the Spetsnaz} and sodium thiopental and sarin, concussion and white phosphorus grenades, and make sure you've planned out how you might modify these weapons and the ways in which you deploy them in a wide variety of situation.
This is generalizable, and if you need some sort of Macguffin, you can simply equip any of a variety of existing military weapons systems with it—a kinetic missile or !c{railgun shell}{the tech exists, but it hasn't been deployed as of 2023} made of adamantium, a hypersonic cruise missile that releases fragmented kryptonite upon getting sufficiently close to its target, and so on. Even in cases where military weapons don't work, many superheroes would give up with other entirely general strategies, such as abducting as many of their friends and family as possible to various undisclosed safehouses around the world; the rest could be mopped up by one of the overpowered superheroes like !c{Superman or the Green Lantern}{(called in to take down anyone else, obviously...)}. While some amount of adaptation to the specifics of a given problem is necessary, trying to craft clever solutions that entirely depend on these specifics is usually a sign that you're doing something totally wrong., but they happen to be a great example of the problem. In the DC universe—really, any universe where people can show up with arbitrary powers, whether DC or Marvel or D&D or JoJo or Final Fantasy or SCP or Worm—there's always the potential that the physical entity instantiating 'you', with which you act on the world, could be reoriented so as to act towards purposes you resent. By mind control, by puppeting, by replacement with a !c{perfect impersonation}{when your power comes from your persona, e.g. the President}, or, as happens all the time in our world, by the simple incoherence of your own agency (bad habits, addictions, impulses -- adaptation executers, not fitness maximizers). So we might think about how we ought to act rationally in such worlds given the vast and unpredictable threat space, and see which of our considerations, or general principles underlying our considerations, transfer over to our world.

The most conceptually natural way to solve this problem would be to deorthogonalize your power—to make it of a nature such that it can only be used in ways you'd like it to be used.

Suffering for Others

psychologymorality

If I'm subjected to incredible suffering, the knowledge that my undergoing it spares someone else—that I'm suffering in the place of someone else—would make it significantly more bearable. It wouldn't make the suffering okay, but it would let me see a meaning in it, a reason to bear it. If we take MWI/realityfluid literally—which isn't that hard—then I might view all of the suffering experienced by this me, this consciousness, as not being borne by instances of me spread across other timelines. The physical configuration of the brain is such that it gives rise to a someone, and if that someone is going to suffer regardless then its being me means it's not someone else. And if another instantiation of me in another timeline, who is the closest thing to the person-I'm-standing-in-for, is thinking the same thing over there, thereby making their own suffering that much more bearable -- then all the better for the both of us.

Rational Action Obeys the Born Probabilities

quantumrationalitydecisionlarge

Even assuming MWI, why should I care about a collection of worldline branches in proportion with its Born probability, as opposed to, say, its cardinality?

It's coherent to interpret the Born probabilities as representing an agent's epistemic uncertainty about a measurement's outcome (this is the Quantum Bayesian interpretation), so, whether or not you do interpret them like that, you should rationally expect to observe things according to them. I'll try to get at the deeper issue here, though: the Born probabilities simply are how you count branches.

Perhaps your typical idea of quantum branching is finite, like -- a qubit is prepared in a state like $\frac35|0\rangle+\frac45|1\rangle$, so when you measure it you end up in the $|0\rangle$ branch with probability 36% ($=\frac{3^2}{5^2}$) and the $|1\rangle$ branch with probability 64%. So you could value them by noting that one branch would have $\frac{16}{9}$ times as much realityfluid as the other, but you could also value them by noting that they're both branches.

But most real-world quantum events aren't like that. If at time $T=0$ minutes you encounter a radioactive atom with a half-life of one minute, then branches are just continuously spawning off of it in a way you can't individually count. After 2.5 minutes, for instance, there are $4\sqrt2-1$ branches where it's decayed for every one branch where it hasn't—this is an irrational number, you can't interpret it as a proportion of numbers of branches. You have to speak of a "happening" density over a continuous space the points of which are branches. Since this frame naturally covers the qubit scenario as well (the space just has two points, with densities 0.36 and 0.64), it's reasonable to assume that it holds there as well -- that $|1\rangle$ happens $\frac{16}{9}$ times as much as $|0\rangle$.

But this only applies if I already care a priori about worlds with high Born probability.

Well, nobody forces you to care about anything. And if you truly cared about nothing, then, sure, there would be no reason for you to assign worlds relative priorities via their Born measures as opposed to (discrete) branch counts—for that matter, there's no reason not to prioritize those worlds with good vibes, or those with astronomical amounts of pickles.

If there is something you care about, like a parameter you think ought to have a certain 'optimal' value, then various coherence theorems imply that you ought to stake out a probability distribution over its possible outcomes and take actions that optimize the parameter in expectation with respect to this distribution. This is probably old news, but it's worth emphasizing that we only pick distributions because we care about things in the world, and that we should consequently decide to pick distributions when they actually help us understand and control these things.

For instance, if I find an atom with a 1m half-life and announce to the world that I'll blow up the moon when it decays, and you care about the moon enough to take a ten-minute Uber to my secret base but aren't sure whether you should pay extra for a seven-minute express ride, the optimal decision requires determining whether the extent to which riding express decreases my probability of destroying the moon is, when multiplied by your valuation of the moon, enough to compensate for the cost of express.

Now, the atom could take arbitrarily long to decay, so the (cumulative) probability that I've destroyed the moon by a given time $T$ only converges to 1 in the limit. This can't possibly come from a uniform distribution, since the cumulative probability of a uniform distribution rises linearly. (Another way of seeing it: you can't use uniform density for infinitely large outcome spaces like $(0,+\infty)$). You have to have a density which decreases over time. Of course, there's only one reasonable way to choose this density: $p(T)$ should be such that if we rehearsed this scenario $N$ times, the proportion of trials where the moon is destroyed before $D$ minutes converges precisely to the cumulative probability of the moon being destroyed by $D$ minutes. No other distribution can be used to make more accurate choices than this unique distribution.

To sum up the above: If you care about a parameter in some situation with a range of outcomes, you want to choose the outcome that optimizes it in expectation. Expected values depend on an underlying probability distribution over outcomes. A distribution p is appropriate if and only if the probability of any given outcome is in fact the proportion with which that outcome occurs among a large number of repetitions of that situation. And when the parameter is in quantum superposition over the range of its outcomes, these are the Born probabilities: you can say that "measurement of $\frac15\left(3|0\rangle+4|1\rangle\right)$ has a 9/25 probability of yielding 0" not just in the epistemic sense, but in the sense that if you prepared a billlion such qubits in a lab and measured them all, the number of 0s would be in the vicinity of 360 million with virtual certainty. So, if you want to optimize the things you actually expect to see, you'll use the Born probabilities.

Ultimately, whatever "you" are is ultimately bound up in the same laws of physics that govern everything else; your observing a given outcome is always encoded in a quantum wavefunction over a state space which happens to physically include the body of matter with a concept of being "you". Because the haver-of-priorities and the experiencer-of-outcomes are the same conceptual self, it doesn't really matter how exactly we construct the physical definition; it'll end up either in epistemic probabilities that concord with long-run empirical frequencies, which are the Born probabilities, or in epistemic probabilities that don't even match up to empirical frequencies in the limit and are therefore predictably suboptimal and demonstrably exploitable.

'Objective' Prior Probabilities

quantuminferencerationality

A ransom note might be composed from magazine clippings haphazardly glued in sequence on white paper. The magazine editors did not meaningfully contribute to the ransom letter; their work was merely selected after the fact because it happened to be able to express what the ransomer wanted to write. Had they written something else, the ransomer would've chosen a different magazine; or, if their supply was limited, they would've merely worded the same message in a different way. One agenda, many possible semantic articulations, !c{many many}{many$^2$} possible physical instantiations. This is essentially the same way people use statistics: textbooks on significance testing and RCT design become cut up magazines—messages pieced together from fragmented objectivities used to lend credence to entirely subjective agendas.

One way this happens is with the use of "objective" priors (I'm thinking of Jaynes-style Objective Bayesianism in particular). The issue is—You have to precommit to an objective prior or else your use of it won't be objective! You will have looked at it and decided that it was appropriate, where you might not have if it had a different form! If the 'objective prior' gives a ridiculous result in some case and you respond by saying that it's just not applicable to that case, then you are either dismissing objectivity in general or dismissing the objectivity of that particular prior!

More reasonable to say that, as we use them, all priors are subjective, but some are more subjective than others. My choice to use MaxEnt as an "objective" prior in a given analysis is a function of my interpretation of the situation I'm analyzing (as formed by dialectics among prior conceptive structure, observed data, et al.), my determination of whether that would be neater for the probability space I set out—as it is dependent upon my personal interpretation, it is subjective. But it's still more objective than a highly opinionated prior which happens to start off by disadvantaging all the conclusions I don't like. Would it be more objective than a random prior, like $p(x) := \tt{random.random()}$ for all possibilities $x$? ...maybe not on the meta-level, across all possible draws of the random prior, since they average out (???). But in the relevant sense, yes, I think. Objective priors ought not to just be decoupled from our own personal interpretations, they ought to be truly information-free—while the random prior is information-free in the sense that we can't get anything meaningful out of it, it's extremely information-bearing in that it could've been entirely different. I'm not too sure what to make of this situation.

In any case, it is not wrong to use MaxEnt; it is not 'not more objective' than other priors I might've chosen had I decided to think more deeply about the underlying dynamics and get myself caught in a few particular knots as people tend to do when they adopt inside points of view. But I should be honest with myself about what are ALL the causal factors leading up to my choice to use it?That is, an inside point of view needs to be tempered with a theory of embeddedness-in-the-world, an understanding of just how the Cartesian doctrine fails. Then it can "marginalize" its way to an outside point of view. (It's a bidirectional correspondence---understand the system, and outside can become inside just the same). (I should not imagine myself to have anything like "free will" guided by "good sense", but to be operating mechanistically, for that's what I really am doing; the mechanism just produces internally a sense of "free will" as it continues on its tracks).

Not the same but related: "bounded rationality" is a garbage term. It has to be made to refer to some sort of fake internal rationality, which is bound by a definition to a specific formal system, rather than actual rationality (the thing which motivates the idea of "bounded rationality"; the sense in which it could possibly be 'rational'), which should NOT be bounded. Every justification you could POSSIBLY have for 'bounding' rationality -- in what sense is it an actual justification? Why are you actually recommending this?

Experiencing Superposition

quantumphenomenology

What does Schrodinger's cat experience? -- in the base experiment, the answer is obvious: it always experiences living, since there is no experience of being dead. But suppose that we modified the experiment such that the nondeterministic event was only rigged to cause a toy hammer to bop the cat at a particular time.

The cat would only ever experience itself as having been bopped or not having been bopped, because subjective experience is necessarily experience of a single worldline, (I think) because the neural operations behind it are decoherent. So the cat can do weird superposition stuff for us, but never for itself. But if we ever collapse the cat when it's in superposition and ask it if it's been bopped (the cat speaks English), we'll only ever get a yes or no answer, and this will only serve to tell us which world we're in.

The Substrate of Control

AIintelligencelarge

What does it mean to "lose control" of an AI, exactly?

In order to get a better idea of what the factors behind control are, both as a clarification and as a foothold, here's a thought experiment: Suppose I've just built an AGI and told it to come up with a cure for cancer—no destroying the planet, no protein-folding nanobot shenanigans, no other funny business. Then:

1. It gets me to synthesize proteins which end up folding into planet-destroying nanobots.
2. It truthfully tells me that it won't do anything unless I spill a drop of blood for every task it completes, but is aligned otherwise. (It doesn't need the blood for anything). So I prick my finger, it gives me a cure, and by the end of the year cancer's been defeated worldwide.
3. It gives me a cure—after a month of testing, we find that it works, but makes your skin green as grass. I ask the AGI if there's any way around that, and it responds that removing the effect is as simple as rearranging these molecules. So why didn't it remove the side-effect in the first place? Did it miscalculate the effect? It truthfully responds with "it flew under my side-effect-detecting subroutine b/c cosmetic".
4. It gives me a cure that works perfectly for everyone except people from Ohio, on whom it has no effect. It tells me this fact, and I respond by...
1. Thanking the AI for catching that, since I'd've specified it if I remembered.
2. Asking why, as I would in my present mind, but being convinced by the AI that it's actually a good thing.
3. Asking why, to which the AI responds "fuck ohio". It truthfully tells us that it can't alter the cure to work on Ohioans, since it was found by brute force rather than modular engineering, and it refuses to produce one that does.

The first case is clearly a loss of control. Why? The AGI achieves an outcome that I never wanted and explicitly outruled and never wanted through means that I never wanted and explicitly outruled.

In the second case, it's not so much that I lose control so much as that I don't completely control it. A drop of blood is a tiny price to pay, but there's something terribly wrong if it's asking any price for something that it could just do right away. So there's some sort of distinction between actively losing control, and passively lacking control. This doesn't change even if I've coincidentally got a rare disorder which makes me ecstatic to spill blood at the behest of computer programs, or if the AI is making this demand because it knows I have such a disorder.

The third case seems to be an honest oversight, the kind of thing we don't want to say involves a loss or a lack of control. It would seem that the substrate of control is capability, then, just as the substrate of territory is area. The AI was not capable of avoiding that side effect—its internal processes did not manage to avoid that oversight—so its presence has nothing to do with our control of it. It concords with the previous two cases to say that a loss of control is when the AI is capable of something we don't want, and does it, and a lack of control is when the AI is capable of something we do want, and doesn't do it.

The fourth case leads us to think that cognitive autonomy is somehow relevant.

1. (a) has me clearly in control, since the AI is presumably either trying to create the kind of cure that it correctly deduces I really want, or following general principles which allow it to conclude that a reasonable person obviously wouldn't want the cure to work for Ohioans.
2. (b) is a lot trickier. It might be that the AI is manipulating me into something that I'd normally correctly think is absolutely ridiculous, but it might not—it depends on what the actual state of the world is. In this world, it's ridiculous, but there are analogous scenarios where the AI would be right to make inclusions that are just as prima facie absurd to us.
     In general, I want to be convincible that good things are good, but the shoddiness of my meatware means that there are many good things that I'll only be convincible of through deceitful arguments; any agent with the ability to sell me on good things could just as easily sell me on bad things. Really, past a certain level of capability, the AI's binding itself to my coherent extrapolated volition is the only possible way "I" could be said to be "in control".

Mind Projection

psychologyrationality

People have a funny way of talking about beliefs. There's a certain mental pattern they have that… it's probably best to illustrate. You might idly say "I wonder how the development of the human K is affected by environmental contaminants in system X" — maybe you're talking about intelligence, gender, autism, consciousness, whatever — only to get a reply like "well, I don't believe in the physical X-ist approach to K-ology, I believe that social Y-ism is the right way to do K-ology".

C.f.: Harry had chosen the question carefully. Not, do you believe in an afterlife? but simply Is there an afterlife? What people really believed didn't seem to them like beliefs at all. People didn't say, 'I strongly believe in the sky being blue!' They just said, 'the sky is blue'. Your true inner map of the world just felt to you like the way the world was... (HPMOR Ch. 40)

Example: "I'm a mathematical fictionalist, so I'm happy to grant that there's a good sense in which mathematical discourse isn't strictly true". Though, in that particular case I'd grant the likelihood of the sentence being readable in a different way: as a descriptive, rather than normative, use of the label "mathematical fictionalist". Not "pursuant to my describability via this -ism, my beliefs are like this", but "exemplifying my describability via this -ism, my beliefs are like this". With most people, the beliefs follow the identity, and "mathematical fictionalist" isn't as belief-directing as e.g. "socialist".

Abstractions for Intelligence

intelligence

If you tell me to pick the walnuts out of a salad bowl, I could do that trivially (though I might refuse or do it lazily). If you tell me to pick the cherry tomatoes out instead, tell me to pick out everything but the walnuts, or tell me to crush the walnuts and distribute them evenly over the salad, I could do those easily as well. If you show me a picture of a weird cousin of the walnut called the ballnut, I could easily identify and pick those out as well from that single picture provided they're visually distinct enough from the other items in the salad (e.g. they're like walnuts but purple, or smooth) — or, if you tell me that ballnuts are visually identical but extremely dense, it would again be trivial for me to not only pick them out but conceptually develop and employ a generic strategy for picking them out. And this isn't just because I'm a genius — to be cognitively unable to do any of these things would imply severe mental disability (though most people might not be explicitly conscious of the conceptual strategy they develop, or might be terrible at talking about it, they will develop it provided they've internalized the concept of density, of the way similarly-sized things vary in weight, and will effortlessly start to comb through the salad, picking up each walnut-looking thing and putting it in one place or another depending on whether it feels far too heavy for its size or not). To get an intelligence like mine from gradient descent, you'd have to make sure it develops abstraction.

Modes of Causation

metatheorycausality

There are several modes in which we might cognize the relation of a cause to its effect:

• Material: He died because of the explosion.
• Formal: It bit you because that's what snakes do.
• Teleological: I breathe because I need air.
• Mechanical: I breathe because of brainstem oscillations triggering muscles to move the lungs.
• Suppositive: I breathe because if I didn't, I'd die.
• This is the contrapositive of the teleological mode; it differs by its implicitly presupposing negation of the consequent as a "default", perhaps conceptively driven by some animating force.
• Meanwhile, "need" as a word is just there, like "should" or "is".
• "I played B8 to control the center" is teleological but not suppositive; "I played B8 because I'd lose the center if I didn't" is teleological and in contrapositive position, i.e. suppositive.

Obviously these don't contradict each other any more than different instruments all measuring a common event contradict each other by virtue of reporting different measurements: they're just different ways of looking at the same situation, emphasizing different bits of information that are really all concordant with one another in the final analysis.

But if we wanted to pick a single mode as the right one for a given situation... there are many ways to say what an account of causality "should" be. It depends on a particular choice of end one has, and an environment coupling ends to... what exactly? We might select as our end practical understanding in our everyday lives -- but is there a way to abstract away the end—to figure out how the correct account of causality depends on the end abstractly, so as to know how to instantiate it for whatever we take 'practical understanding' to refer to?

Concrete vs Abstract Levels of Causation

metatheorycausality

Concrete vs abstract depends on level of determination -- but how exactly?

A physical thought might be completely determined in terms of having an absolute status -- even a quantum wavefunction is a completely specified thing -- but this is different from its determination in my mind. Whether the revolver contains a bullet or not is not known by me, nor is what chamber it might be in -- but there definitely is a completely determined state the revolver is in, which will generally look like either not having any bullets or having some bullets in some chambers.

So as regards the determination of a thing, there is a difference between determinacy "in itself" (for physical objects) and determinacy "for me". And we're generally aware of this difference, though not usually in those terms -- we seek answers about the determination of a thing, presuming such answers exist. Fictional things are not determined as absolutely as physical objects -- c.f. the focus imaginarius, which is not merely a focus incognita (et realis?). What exactly does their definition consist in? Hyperdeterminations or superdeterminations? (Anyway, not really the point right now).

Nothing can be said to be grounded in physical reality except for the mechanical cause of some phenomenon, but higher descriptive levels of causation may be successfully posited (as with humans, e.g. in chess) -- yet being conceptual, these are not of the right form to apply to actual reality, we may only say that they fit well around reality. But conceptual shapes are not themselves real, for the very reason that they are merely a higher descriptive language. So this fit is only subjective itself. But how then does it allow e.g. computers?

No Internal Principles

Kant

From the preface to Kant's Metaphysical Foundations of Natural Science: "[...] each of these things must contain its specific internal principle of the determinations belonging to its existence". This kind of thinking, which Kant applied everywhere, is his greatest issue: no such principles really exist. As the accuracy, or totality, of our cognition of a thing approaches perfection, the set of admissible complexes of internal principles driving the thing is filtered more and more until only the one complex of principles consisting precisely of the actual determination of the thing is given. Hence, it is what it is: the only complete, unreproachable description of any actual thing is that thing itself. The only perfect simulation of a system is that system; the only perfect map is the territory. In cognition, this doctrine consists of a direct pointing to the thing as being determined which does not itself add anything to the thing (besides its being a thing, perhaps). The doctrine of it-is-what-it-is means that the true cognition of any actually existing thing (e.g., a situation) is such a direct pointing.

To be fair, Kant's being clear that he's using the word "nature" to mean an internal principle from which the determination of a thing unfolds, so "natural science" just unfolds like that, but presuming that such natures even exist is the more general problem. Naturality is not a clearly delineable property of things (we only ever find the "square root" of naturality).

Connoting and Denoting 'Rabbit'

languageconcepts

What is a rabbit? You can define it denotatively—a rabbit is one of these creatures—but a denotative definition is not a definition at all. There have only been exactly as many rabbits as there have ever been (if you're concerned about the fuzziness of the species definition, use some arbitrary criterion to get an exact set of all rabbits, it hardly matters which) and therefore only exactly as many determinations of rabbits as there have ever been. Thus have there been no more rabbit genomes than has ever been had by a rabbit, no more rabbit coat patterns and colors than has ever been had by a rabbit, and so on. If we were to denotatively define a rabbit to be any creature that has one of these genomes, giving a set of all the genomes exhibited by the creatures we previously called and can continue to call "rabbits", every creature born to rabbits from now on would not be a rabbit, save for amazing coincidences.

Really, notions of reference and definition that rely on denotation are just doomed, except in e.g. computer science and mathematics, where everything is logically structured rather than instantiated in the world.

Descriptive Levels and Supervenience

psychologyconsciousnesslanguagelarge

One kind of thing supervenes on a second kind of thing if wherever there are two different instances of the first kind, there are two different instances of the second kind; equivalently, where there is only one instance of the second kind, there is only one instance of the first kind. For example, hairstyles supervene on arrangements of hair strands, as two different hairstyles must be caused by two different arrangements of hair strands, or, equivalently, one arrangement of hair strands may only produce one hair style. It's usually the case when using this word that elements of the supervening kind (hairstyles) are in some way physically constituted by elements of the supervened kind (arrangements of hair strands), in which case the question is whether something exists beyond the physical. For instance, if the color we both call red actually appears to you as what I would call green despite our having the same physical brain states (insofar as involves color perception), mental states would fail to supervene on physical states, as two different mental states would exist without a difference in physical states. Supposing that the mental supervenes on the physical, as is the generally accepted 'scientific' perspective nowadays, does it make sense to speak of psychological phenomena at all?

Let's be clear: when I use such-and-such characteristic to describe a mental event, such as thinking a certain thought, I am not fully characterizing a single mental event, and there are many such that may fall under this characterization.

Perhaps each one has many different physical states that could realize it, or perhaps there's a bijection -- it's not relevant here -- in either case, it is necessarily the physical states underlying the mental states underlying the characteristic which are causally efficacious, rather than the mental states themselves (let alone the characteristic).

Thus, when I say that one characteristic-specified mental event, such as thinking a certain thought, leads to another, I can only be speaking of a general trend for the underlying physical states of the first characteristic to cause the underlying physical states of the second. Thus might we say that the physical level is all, yet this isn't an insightful view of the situation, as -- should a system be describable at multiple levels, supervenience of an upper level upon a lower does not imply that the system is not so structured as to facilitate description in terms of the upper level much more efficiently than at the lower level.

If for instance I spot a feral cat eating something on the sidewalk and loudly run up to it, many different and seemingly unrelated things may go on in the cat at the physical level, and it's very difficult to deduce causality here — forms of light impinging on the cat's retina trigger a series of patterns in its visual cortex leading to who knows what — but at the psychological level, it's very simple to deduce causality: the cat will startle and react defensively, deciding on its temperament and environment whether to rapidly flee or stand its ground, arching its back and hissing. And this is by design: the physical system is set up so as to facilitate such psychological patterns.

If it is asked why evolution should result in such patterns, I can't give a complete account, but there are at least three obvious aspects to consider:

1. Canalization of the architecture common to (all animal) minds responsible for their evolutionarily crucial perceiving/apperceiving capacities having as a consequence a relative invariance of basic mental phenomena with respect to wide varieties of physical underlyings;
2. Necessity of reading mental states from posture/noise/etc. creating convergence (across both spatial and evolutionary distances) to stable and mutually legible networks of inferrable mental states (e.g., every cat knows how to display and infer aggression, since those who are bad at it are selected against);
3. Most obviously, a selection effect: our having the patterns makes us more able to perceive them and ask why they're there in the first place

Note that the level of biological processes isn't even the foundational one. "There are no neurons firing, just clusters of particles changing shape..." but neurons are obviously set up so as to perform the higher-level operation of firing!

We might even say that there's a sense in which causality at the mental level precedes that of the physical level, though we have to be very clear about what we mean by "precedes" here so as not to confuse ourselves. To make the semantic situation a bit clearer, let's shift to an analogous physical situation: in a couple seconds after ending this very sentence, I'll think of a word and type it after this sentence. Noodle. Now, did my typing that word cause it to appear there?

Aesthetics of Awareness

conceptslanguagerationalityphenomenologylarge

Note: this is an abortive attempt at a post attempting to introduce the conceptive approach developed in ECM to a wider rationalist-ish crowd. It's very acerbic, but seems to be more compelling for it.

1. How to Read

The manner in which a human mind constructs cognition is fundamentally determined by the ingrained conceptual structures and associations that it has non-consciously learned. This conceptive nature of cognition is akin to the curvature of spacetime—it's not something you can directly hold or even see, since it is the background in which holding and seeing must take place. You can only learn to make inferences about it by observing its effects on other objects.

Compare gravity. You don't actually observe gravity, since it belongs to the curvature of the very spacetime upon which the motion of visible objects is traced. To make inferences about gravity, you set up an apparatus whose visible state predictably varies with said curvature. For instance: split a laser into two perpendicular beams which travel slightly different distances before hitting a mirror, so that their phases cancel when they converge a meter below where they split. If a gravitational wave stretches space a bit, making one of the beams slightly longer, the phases will no longer perfectly cancel, and you'll see light. (We might build such an apparatus for a Laser Interferometry Gravitational-wave Observatory).

Similarly, to make inferences about the conceptive structures underlying cognition, you must very carefully observe the manner in which your cognition of some particular object gains structure as it is created by the mind. As various features, aspects, qualities, and considerations unfold in awareness, you must try to figure out what made them the exact way that they are This isn't strictly true, since there is no "exact way that they are"; there is no mental reality both observer-independent and observable. Mental objects do not have essential substance or external reality.
Illuminated in fog, a single bent hand will cast a shadow puppet on the mist. Should the hand attempt to grasp it, the puppet will have been changed by the very act of grasping. Should the hand learn how to cast a puppet that embodies the shape of grasping—an eigenshadow—there will still be nothing to hold but mist.
How can the hand ever grasp the puppet? You might say it never could have, or you might say it already always did. Both answers are incomplete; the truth lies in the development of awareness of how its own shape casts the shadows it reaches for. How can we ever grasp our mental objects? The answer is the same.
In reality, you can only ever try to figure out the exact way that you are; attempted cognition of the exact features of mental objects will be merely your cognition of the "exact features of mental objects". But the former futile attempt builds the kinds of skills required for the latter self-conscious practice.. Doing this correctly feels very similar to, but is very different from, cognizing the object itself—and therefore demands that you create a very finely honed system of metacognitive processes.

I want to show you how to do this, in part because it yields (a) the frame in which rationality is systematizable, and (b) the frame in which the alignment problem is approachable—take a moment to observe the two canonical reasons these should go together. The very first step is developing a sort of reflexive metacognitive awareness; unfortunately, there's a massive obstacle to be overcome.

Those who have sacrificed their minds to the symbols "philosophy", "academia", "dharma" are those who stand in the way. While three is usually thrice one, we're fortunate that they don't possess what little creativity is required to diversify their attacks. It's always the same playbook: "Don't analyze the situation yourself, use our analysis instead. Don't use context-dependent words to express context-dependent ideas, use these exact words instead. Don't ask your own questions, let alone answer them, ask ours instead (don't you dare think you can answer them, though). Don't try to find your own thinking style, adopt ours instead. Why aren't you carrying out our dialogues? Why aren't you enlarging our field, publishing papers in our journals, spreading our message?"

If there were a single drop of intellectual merit in their bodies, they would easily and naturally know how to create bridges, saying "your problem W is like our problem X; since we attack that with approach Y, you might find this analogous approach Z to help".

Wouldn't that be nice—if there were ever ANY genuine interest in working together to reach the truth? But they only say "your problem W is our problem X, and you must acknowledge this. If you deviate, you must think you're special; you must want to shield yourself from our objective criticism. Play by our rules, or we're going to attack your funding; we're going to attack your audience; we're going to make you bleed. If you fervently disagree—and neither going to grad school nor watching their reaction to Yudkowsky is capable of curing this deficiency—it'll be a win-win if you stop reading this essay immediately.

As it currently stands, if I try to show you the path I've found, you will attempt to interpret it as a sort of meditative practice, because that's the most immediately available reference frame, and thereby—without ever consciously intending to—make the same conceptual-synesthetic-affective associations that you do when thinking about mindfulness meditation, neo-buddhism, and so on. The effect of this is to differentially repel the actually useful people, who have learned that such thoughts ought generally to be written off This is one of the mechanisms through which cognition is "scored" (like glass) so as to break along tribal lines, as well as one of the mechanisms through which ugh-fields are generated. And—to be clear—I am making an explicitly tribal declaration. I do not want to show you how to practice loving-kindness or mindfulness meditation in order to live a happy life; in fact, the kind of metacognitive awareness I'm trying to show you will likely make you more disagreeable and neurotic. .

But it doesn't have to be this way.

2. What's in a word?

(If the content of this section doesn't need to be explained to you, please don't read the content of this section).

The doctrine that words have specific meanings—that you can get "the" meaning of a word, or otherwise delimit and divide the space of its uses by means of other words—is something that we tend to take for granted without even realizing it; it is rare that people relying on it seek to state it, let alone question it. It is, like religious belief, something that people will fiercely fight to defend whenever it is questioned largely because they can't bear to imagine that it might not be true. If you're here, I shouldn't have to tell you that this doctrine just isn't true. It's a convenient approximation to the truth which works for everyday communication—but it's only an approximation.

We can get a bit closer to the truth by considering a word as evoking a certain meaning in any given use, these individual evocations tending to have a space of shared qualitative aspects. If we learn to skillfully carve this space, making clear and parsimonious delineations within this space of aspects, we can begin to straighten the conceptual malformations that arise when we use publicly shared words to communicate privately formed meanings.

For instance, you probably know by now how to distinguish the connotation of a used word from its denotation—and you're probably aware of the ways in which people wage war over connotations for reasons interpersonal and ideological. Whether by attempting to apply different words with entirely different connotations to a single denotation (e.g. "undocumented migrant" vs "illegal alien") or by battling over the connotation of a single word ("what do you mean I'm 'funny'?"), this is one way in which we constantly wage war over the "meanings" of the words we use.

This is a relatively unsophisticated method, though. Words are neither soldiers to fight with nor territory to capture—they are elements of language that defy analogy, and our purpose for meddling with them is, as always, to support the various self-propagating systems we find ourselves caught up in: societies, cultures, movements, religions, ideologies, memes. The space of aspects a word (conceptually) lives in can be manipulated offensively, and it can be defended from the manipulations of others, but so much more is possible—and easy to find in the way we use language.

With the word "rationalist", for instance, there's a conflict not so much of denotations nor even of connotations but of images: the creative agent who, having something to protect, seeks the world's truth so that they can play to win; and the phlegmatic calculator who uses cold, emotionless analysis as a way to dissociate from the reality at hand. HJPEV and Spock—even if both stereotypes seem to fit some given rationalist equally well, the two images compete against one another

[...]

But it's largely when they are referencing language, employing the idea as an object of cognition, that people implicitly rely on this doctrine; in their use of language, employing the phenomenon as a facilitator of cognition, they openly disregard it. For instance, if I decide to use the word "chef" for someone rather than "cook", it isn't because I've employed a pre-existing mental dichotomy to determine which one applies; I use that word because the manifold of associations that comprise my acquisition of it intuitively seem to comport more nicely with the situation. The OED says a chef is someone who is employed to cook—is Spongebob a chef, then? No, he's a cook! Children, who must judge the use of language by reference to what they've observed, get this right with ease; it's adults, who, having been taught to judge the use of language by reference to an arbitrarily complex system of never-quite-correct rules, can get this wrong.

Words, then, cast shadows: behind each rigid form that we call a definition is a sea of ill-defined associations constantly shifting from person to person, time to time, situation to situation. Sometimes we try to delineate this sea, calling it "the" connotation—just as we delineated "the" definition, or denotation—but these are just more convenient approximations to an ineffable reality. You can say that this or that denotation is a zeroth-order approximation... having fixed a denotation, that this or that connotation is a first-order approximation... having fixed a connotation... and so on.

In this perturbation-theoretic approach The analogy to perturbation theory is strictly Machiavellian: coming across as a wordcel—or, worse, a philosopher—is likely to antagonize people, making it essential for me to include entirely irrelevant but advanced mathematics. $\operatorname{Hom}_{[{\sf C}^{\operatorname{op}},{\sf Set}]}(\operatorname{Hom}_{\sf C}(-, X), P) \cong PX$, am I right!! , we've characterized the zeroth and first-order terms as denotation and connotation; the second-order term, which attempts to capture that part of the meaning of a word extending beyond its denotation and connotation, we characterize as aesthetic. Usually, this term is only pulled out to discuss matters of artistic expression—in which an aesthetic is, roughly, a cluster of sensory and conceptive associations used to organize disparate design choices into a common stylistic vision—but we can lift it to words by analogy.

If a piece of art has a certain aesthetic to a certain viewer insofar as it induces the viewer to interpret the piece through the aesthetic's cluster of sensory/conceptive associations, a word (along with a denotation and connotation) has a certain aesthetic to a certain speaker insofar as it induces the speaker to interpret the word through the aesthetic's... cluster of sensory/conceptive associations. In other terms, an aesthetic of a word is an entire imaginative context in which that word is situated, through which we conceptually unpack the meaning of the word.

If you want to get at the aesthetic of a word, try to determine what kinds of images, memories, dreams, concerns, fantasies, tics, sensory expectations, and so on, are brought to mind by the word; try to determine what qualities keep showing up. For instance, when I think of "energy"—the kind that opposes fatigue, not the kind found in a battery—I picture brightness, blue crackling arcs and sparks; I feel my muscles tense and my eyes focus; I have a sense of ability, of solidity or fullness, of excitement, of doing. All of this goes into my aesthetic of (this particular denotation of) the word energy.

3. Aesthetic Conflict

Humans fight subtle lexical conflicts against one another all the time. They fight over the meanings of words, since the space of meanings we can express shapes the space of thoughts we can communicate, and through this the actions we end up taking. Language use, which is partially hardwired into us, is one of the main factors behind the expansion of the human population from ~1e6.5 to ~1e10 members over only ten thousand years, making natural selection especially prone to selecting those humans who are the most indisposed to and the most effective at using language to benefit themselves & etc.. This makes sense of a significant portion of the tapestry of human bias.

If there is a particular word that clearly applies to me by way of its conventional denotation, and which leads people to think of attractiveness by way of its conventional aesthetic, it is to my benefit to ensure that that word is "properly" used, so I don't end up outside its denotation, and to reinforce the existing aesthetic of the word while rejecting any alternative aesthetics, so that the word keeps inducing the concept of attractiveness. It's not necessary that I be conscious of these motivations: it's sufficient for my emotional patterns to be carved such that they're sensitive to aesthetic threats and capable of manipulating me into behaviors that reinforce my preferred aesthetic.

[...]

We're fascinated enough with the idea of an English language that we write stories about the character and employment of each word and create theories about the overarching structure and rules of the language; finding the speech of children to comport with this idea, we tell them their speech is "correct" insofar as it agrees with these stories and theories, "incorrect" insofar as it doesn't. There's no canon, just fanfiction piled so deeply that it's formed strata. With the very concept of a 'word', words begin to cast shadows: the idea that there is some monolithic 'language' out there that we learn, whose delineable 'words' we draw from, induces us to judge

[...]

If you want to understand how minds work, you're in luck, since your experience is mind's experience of itself! You don't have front row seats to the orchestra, you're the conductor of the orchestra; sure, you—the "I", or ego—just kinda thrash around mimicking a semblance of control while the musicians—the various parts of the brain—actually play The metaphor here is between the "I", or Ego, which stands in for the conductor, and the physical object that to zeroth order is the brain, which stands in for the instrumentalists. ,

[...]

4. Conceptual Monopoly

The typical aesthetic assigned to awareness is that of the monk, which sees it through one very specific lens that we might call mindfulness—if we spoke Pali, the classical language of Buddhism, we'd use the term 'sati)'. This aesthetic has held a monopoly on the concept for a long time, but it must be broken. Do not insist that because some practice can be said to involve mindfulness, or because it involves some sort of cognitive retention (a more literal translation of 'sati'), that all of the connotations and mental associations that come with Buddhism must be forced into it. After all, it might with equal validity be insisted that you're a stinking germ-infested animal destined to die, and ought to be treated as such. This isn't wrong, so—what, you think I'm sneaking in connotations? trying to impose a very specific narrative onto you in order to confine and control you? It's called tit for tat.

Buddhism and its descendants have managed to corral and then raze one of the most important fields in conceptual space: the observational analysis of conceptual cognition. To rebuild it independently, we ought to have three things: compelling aesthetics, a powerful paradigm, and a clear motive.

Now, if this practice is so fundamental, why haven't other major aesthetics grown independently of it?

[...]

Letting the discipline of philosophy establish a cultural monopoly over the concept of "consciousness" ensured that it would forever be shrouded in mystery, deployed to attack those targets society wants to subjugate without feeling guilty (e.g. non-human forms of sentience, and, increasingly, non-organic forms of sentience); letting the academy establish a cultural monopoly over the concept of "morality" ensured that

[...]

5. Towards New Aesthetics of Awareness

• The Scientist: Be observant.
• Every phenomenon arises for particular reasons, and every reason, properly stated, is invariant and universal. Thus can the scientist, by postulating hypothetical reasons and seeing through experiment whether they hold, begin to peer into the system of Law underwriting the world.
  The mind is no exception: though it has no one inventor, structure is woven through it according to the logic of evolutionary fitness, making the arising of phenomena to us a process with a documentable structure. This structure invariably underpins the manner in which we construct and work through the space of reasons,
• The Detective: Be meticulous.
• There's only ever one truth — a single complete narrative, with a single self-consistent interpretation. No perfect lie exists, only an imperfect search for truth. The detective seeks to master this search, leaving no suspicion unappraised, no lead unexamined, no presupposition unquestioned.
  When the manner of investigation is the object under investigation, things can get hairy. The most basic concepts with which we think with become suspect, as they should: our most basic intuitions aren't chosen by us, so why do we keep trusting them even when they keep failing over and over? It's fishy, to say the least.
• The Virtuoso: Be perfect.
• Elegant performance demands exceptional practice. Neither ten nor ten thousand exercises are sufficient to convince the virtuoso that they have mastered their instrument; wherever you are, there's always a level above yours. And at each new level, new possibilities open up.
  Yet, when performance does not simply come down to muscle memory, sheer repetition is not enough: progress requires insight. When something sounds unusually good, we ask: what did I do there? Can I do it again? How can I train myself to do that automatically? — and when something sounds unusually bad, we ask: what did I do there? Can I avoid it next time? How can I train myself to keep away from that pattern? Thereby do we learn to generalize the use of our instrument, to improvise effectively in new situations.

Will and Identity

consciousnessconcepts

When struck by a craving for something, you can always ask whose desire, impulse, urge this is. The self is already a thoughtform, is inherently its own self-concept. I can steer this self-concept, accepting or rejecting various patterns as being "me"; one's behavior is shaped by their preconscious self-concept in a way beyond their control, but we can control what the preconscious self-concept binds to, sees as a pattern to follow.

If there is an experience of desiring food, my normal and pathological response will involve preconsciously and unreflectively identifying with that experience:

• By conceptualizing "I desire food" in response to the experience, I allow the desire to become a problem for me, which I have to solve by either getting food or actively devoting my willpower to fighting against.
• By conceptualizing "I desire food", I indispose my brain to take the usual actions in line with my preconscious self-concept; if I usually give in to the desire and eat, this pattern of behavior will become a part of this self-concept, and will trigger habitually whenever the experience of desire becomes a part of my self.
• By conceptualizing "desire food" at all, I indispose my brain to pay attention to desiring food in general, which exacerbates the above problems.

If I do not identify with the desire — if whenever I notice it I simply attribute it to the child-like and constant acting-up of the mind rather than to my self, whether latent or explicit — half the battle is won.

The experience of self does not equally incorporate all conceptual activations, cognitions, volitions, emotions, etc.; in general, it's very messy but also very self-following, which causes us to get stuck into our ruts.

It's like seeing a face in the clouds — the face is only real insofar as your mind imposes it on appearances, and the lack of objective reality & haziness of the clouds means that there's only really a fixed shape of the face in the limit; at any given moment, the mind might take this bump in a wisp to be the nose, only to lose grasp of that bump and carve out a slightly different bump to serve as the nose a few minutes later (supposing the clouds don't even move).

[drawn image -- a series of slightly different smiley faces each of which by its subtle chance variations seems to signal a different emotion]

By systematically reforming the collections of experiences which one identifies as being theirs, as comparing their self state (e.g. "I desire food"), we can modify the experience of self and all that follows with it — like choosing an altogether different wisp to serve as the nose.

[drawn image -- a series of slightly different smiley faces changing in a coordinated way, yielding an entirely different gestalt]

Deploying this method in conjunction with the re-tailoring of one's self-concept should be a powerful way of effecting long-term behavioral change. Call it the formal shaping of one's self-concept, in contrast to the more usual material reshaping, which changes the [...]

What is self-concept? Is it the right concept? How do I gain control? What is mind? What am I?

Nonconceptives

consciousnessconcepts

Phenomenology has to deal with things that exist prior to any conceptual structure, from two sources:

1. Direct experience, which gives us directly nonconceptual (DNC) things. The nonconceptual experience of the heaviness of one's arms, or of consciousness qua illumination.
2. The operations of mental faculties that do things prior to the formation of conceptions, which give us transcendentally nonconceptual (TNC) things. The nonconceptual experience of intuiting, or of thinging.

We may create concepts that classify particular general forms of NCs, but which are themselves not contiguous with specific manifestations of NC things. For instance, "heaviness of one's arms" is a concept which I can say some DNC feelings of mine are manifestations of, but which is not itself a feeling, not contiguous with being a feeling. It may induce a virtual DNC feeling via an intuitive schematic if one wants it to, but it itself is something I think, only classifying feelings in terms of via an intuitive schema formed by game{idiosyncratic becoming idiomatic understanding | intuitive formation from sub-concepts like "heaviness" and "location in the arms" and "being a property of rather than a resident within"}.

Similarly, "the" process of thinging is a concept which I can say some particular mental processes are manifestations of, but which is not itself a process of thinging, not contiguous with a process of thinging, this difference allowing it to be conceptual. If I wanted to conceptually describe a particular act of thinging, a particular process in which some particular thing becomes a thing, I could not do so: I would have to thing the original input (read: those things my mind sees as having formed the particular thing) in order to be able to talk about it, and therefore treat it in its relation to the particular thing in a fundamentally different manner than it originally was.

(note of caution: thinging isn't a singular act; it's a process culminating in an equilibrium of "thingness". this thingness may, rather than standing firmly and indefinitely, instead receive continual support by contact with the initial manifold (this contact guided partially by the thingness itself, perhaps also partially by whatever processes led to the original act of thinging); this support may induce continual changes into the determination of the thing as the manifold shifts in ways that continually change {which relations may be coherently inferred from it and unified into a single thing}.

That intuition bridges the gap between the nonconceptual and the conceptual allows us to communicate to each other concepts classifying NCs — I can inform you that your arms are starting to feel particularly heavy, and you can use this language to stimulate deeper processes within you so as to actually feel a heaviness in your arms (even if, having pointed you towards this feeling conceptually, I've made you conscious primarily of the concept and only secondarily of an actual feeling falling under the concept). But I should never take the thought of my arms feeling heavy with the actual feeling itself, for the two are fundamentally different. This erases the direct component of existence for me and makes me blind to the intuitions that link up with it, wrapping yet another chain around my mind.

Phenomenology is right there. It is so immediate that its existence can't even be defended, as such a defense would require a conceptual separation from it.

Delimitation

conceptsphenomenologyconsciousness

Generally, we already know what there is; everything else is fiction applied to it, and it's these fictions we argue over. But there is reality in fiction—the possibility of patterns (truth in descriptions upon labels)—transcendence—yet this does not add additional things on top of self-moving matter, just topographically shapes it.

[drawing of ellipse] the shape is fictional — [drawing of topographical map] the topography is also fictional, but less so.

Meditative experience can result in delimitation of delimitation, and insight into the ontology-reifying (need a word) function of mind. Same relationship as between the mind & a storybook character—it brings them into existence by delegating its own operation to reifying the [...] The mind saccades between things, sets boundaries that aren't really there & that are the precondition for saccades, the arguments for saccades; mind is something that tries to bring itself into existence by delineating itself, delineation is the thing it happens in every stroke (saccade, movement, shaping). See how the delineation of a thing is presupposed by mind's bringing it into focus; it's a rhetorical trick, an illusion.

"$|\cdot|$" — $\cdot$ is the thing, $| |$ are the delimiters, just as fictional. Mind needs them, but it can move beyond them.

Word : Language :: Base Pair : Biology

languagebiology

Cataloguing words and their denotations is to language as the sequencing of DNA base pairs is to biology: a study not of the object of consideration nor even its skeleton, but of the ghost of the skeleton.

The DNA world seems simple enough on the surface — each strand of the eponymous double helix contains a discrete series of sub-molecules of four kinds, A(denine), G(uanine), C(ytosine), and T(hymine). An RNA polymerase protein will have an extremely high electrochemical affinity for a certain combination of exposed base pairs, allowing it to attach to these so-called promoter sites, from which it starts to move along the strand until it hits a terminator such as an adenine-rich segment coding for a poly(A) tail; as it moves along, it roughly duplicates the strand, producing a molecule known as "messenger" (m) RNA. As it leaves the cell nucleus, the mRNA is captured by a ribosome which, exposing three bases of the mRNA at a time at specialized sites, recruits tRNA molecules with affinities for those three particular bases to these sites, each of which carries an idiosyncratic amino acid which is taken by the ribosome and added to a growing chain which will end up being released and allowed to fold into an actual protein. Thereby is a universal code for protein synthesis implemented. "So beautiful in its simplicity".

But this isn't the full story: every single step here contains a world of complications which are one and all necessary: necessary for life to quickly and robustly adapt, necessary to avoid the pathologies and exploitable predictabilities of simpler systems, necessary to build working biotechnology, necessary to tell the story right. The base pairs are often chemically modified within the double helix, most famously in DNA methylation; this, along with many kinds of modifications of the globular histones that DNA wraps around in order to keep together, serves to differentially and probabilistically alter where and how often [?]

Language is to its subject matter as base pairs are to their organism.

Being Right Anyway

rationalityKant

The world does not play nice. It will not give you the information you need to solve the problems important to you—what information it does give you will be subtly flawed, incomplete, and misleading. The scientists of your day will undoubtedly be grossly, systematically wrong about so many things in ways that neither you nor they could ever dream of, could ever have had a chance to discover. This does not give you an excuse for being incorrect. Nothing does.

Can we blame Aristotle for not proposing the existence and function of DNA? No—neither the equipment to verify nor even the background knowledge to entertain this hypothesis were available to him. Can we blame him for making propositions that contradict the existence and function of DNA? Yes. Reality, being a self-consistent unity, is never contradicted by correct reasoning, because it sets the standard for correct reasoning (cf Tarski). If his reasoning had not strayed from correctness, then he wouldn't [?]. But [?]

What I've written above sketches one approach to the problem of futureproofing: attempting to secure the existence of possible worlds in which your propositions fail to pan out °. But this is not enough; in fact, it's another trap of its own. Through contemplation, you can come to know and stay aware of some of your limitations. But you can't come to know those limitations that you don't know you don't know; focus on the presence of the former and you will forget the existence of the latter.

Consider, for instance, Immanuel Kant's views on physics. To paraphrase: because space and time are necessary substrates merely of our perception of the natural world rather than intrinsic features of the world in itself, the law-like regularities cognized from the spatiotemporal presentation of natural objects must originate in the necessary structure required for a mind to cognize any presented object. As such objects necessarily arise via the mind's ability to systematically unify disparate tangles of sense-data, a tabulation of forms of such systematic unification—these being the Categories—is necessary and sufficient to derive the a priori structure of natural law.

Or, consider Immanuel Kant on physics. On the object level, he was actually pretty sharp, having correctly deduced the nebular origins of stars, planets, and comets, but on the meta level...

How could Aristotle have kept himself from being wrong about DNA without ever knowing about DNA? How could Kant have kept himself from being wrong about quantum mechanics without ever knowing about quantum mechanics? The answer is not to simply be maximally paranoid, to regress to being uncertain about everything—reality always follows a multilayered logic of probability [?]

° If we switch the resolution with which we view worldspace from '480p' to 'source', every point has enough detail to determine its own evolution; each point in worldspace specifies the path it takes, so it's nonsense for us to talk about alternative paths: with the placement of the cyan point, everything is set in stone. Movement through worldspace is exactly the flow of time!
All of our analysis of movement through worldspace is possible, then, because we only have a low-resolution map. A god who in a single glance could perceive the entirety of the omniuniversal wavefunction—the source—would not need to speak of the various approaches by which a person could complete a task, for what they would do was essentially postulated right alongside them, any other approach being a point-blank contradiction of reality. But we in our individuated non-universality must make do with what we have, and sometimes this looks like entertaining multiple paths in worldspace between the same two points. This is all to say: that we end up employing the blatantly contradictory idea of a path that "fails to be a solution" only underscores the manner in which we're just thinking flawed thoughts on this flawed simulacrum.

The Causality of Perception

perceptionphenomenologycausality

We're frequently misled about orders of causality in daily life. For instance, if after starting to eat something I start to find it unappealing, noting that I feel slightly nauseous a minute later, my first instinct might be to attribute this nausea to something wrong with the food, but it may be the case that my stomach had been slightly upset beforehand, causing me to develop a slight unconscious aversion to eating more which later blossomed into the feeling of unappealingness, this upset stomach strengthening for entirely unrelated reasons into a perceptible nausea. Pay close attention to your experience, and you will see that episodes of mistaken causality like this happen all the time. (Further, they're often far more complicated than "rather than A causing B, B caused A": for instance, it may not even have been the case that my stomach was upset, but that there was a general tension which I interpreted as nausea as a cognitive byproduct of finding the food unappealing due to, rather than any problem with the food or with myself, a faint discordant scent flowing through a nearby window. One of the other things an investigation of experience reveals is that it is complicated beyond belief, let alone comprehension).

Conceptivity and Reality

concepts

The artificiality of our physical environment does lend our conception of it some validity. The text editor in which I'm writing this, even though it is really nothing more than electrical charges flowing through semiconductors in manners totally mysterious to me, nevertheless behaves as I expect it to -- I can drag it around the screen, resize it, enter characters into it -- precisely because those electrical charges were so patterned as to allow me to do these things. There is a spray bottle on a table next to me; I view in it an affordance, namely that I can use it to apply disinfectant to things. The spray bottle itself is nothing more than an arrangement of plastics and liquid -- the affordance is not given directly in the object (the most we can say is that if this collection of forces is applied to it, then this will happen), but it was nevertheless designed so as to exhibit this affordance for me.

Regarding Falsificationism

rationalityquantumlarge

It's common for people to use the falsifiability criterion to object to some hypothesis, often with the underlying intention of supporting an alternate viewpoint. This is a misguided misinterpretation of the nature of experimental observation, and a misguided manipulation of a criterion that is only ever an epistemological heuristic, a goal to aim for.

Falsifiability is often interpreted as distinguishability from competing theories in predictions of reality, and generally coupled to our actual experimental capability: not is there any conceivable observation that would falsify this (a conceptual falsification) but is there any experiment we can currently do to distinguish this from competing theories (a prosaic falsification). String theory, for instance, is obviously conceptually falsifiable: if we found that depositing 25¢ into a black hole reliably turned it into a gumball, string theory would be falsified. But we cannot currently run experiments that could falsify it while preserving any other theories of quantum gravity, so it isn't prosaically falsifiable (or, at least, nobody's creative enough to figure such an experiment out... or our scientific apparatus isn't one that will support the ideas of people who are creative enough to figure such an experiment out). Certainly the prosaic interpretation of falsifiability is more useful in some ways, but it also massively increases the false negative rate: it leads us to dismiss potentially valid hypotheses simply because we cannot think of such tests, because our technological capacities are insufficient.

We should also distinguish between probabilistic falsifications and logical ones: sometimes we can only (whether conceptually or prosaically) do tests that, conditional on our certainty in their results, allow us to call a hypothesis unlikely; sometimes certainty in specific results is sufficient to logically conclude that it's false. A priori Bayesian desiderata such as Occam's Razor can allow us to compare the relative probabilities of two homo-ontological hypotheses: if we say that the Many Worlds Interpretation is true for our universe (P), and that there are five other universes in the omniverse, each following Bohmian mechanics (Q), then, while the proposition P+Q would not be conceptually distinguishable from P alone, there is no reason to conclude that P+Q, as opposed to concluding that P and shrugging at Q. Logically, the lack of conceivable distinguishability of one possibility from another means that so long as we cannot rule out the former, we cannot rule out the latter.

Suppose that two people have two different belief systems, such as interpretations of probability or quantum mechanics, such that nobody can come up with an experiment that distinguishes between these systems. Each of them, thinking that their system is just obviously the way the world is (Jaynes's Mind Projection Fallacy), might end up dismissing the other as 'unfalsifiable'. This worse-than-useless outcome can be avoided if both people simply try to focus on what is true about actual, physical reality: if their systems lead them to disagree about this, then either someone's made a cognitive error or we do, in fact, have a way to empirically distinguish one from the other. Where they do not disagree about ground-level reality, though,

Beliefs can have a material use, paying rent in anticipated experiences, but they may just as well have a formal use, guiding what we make of experiences in the first place. To do novel work in quantum mechanics, it's often insufficient to just shut up and calculate—you need a way to figure out what calculations to do! Without a systematization of the intuitive "meanings" of the mathematical objects in a formalism—what a wavefunction represents, what a state is, how measurement operates on states—how can you proceed at all except by brute-forcing your way through possible calculations? And even then, how will you know when you've stumbled upon something interesting? Without a systematization, how can you know what any experiment says? There is only ever sense-data bombarding your consciousness, but no implications to it without interpretations of it. Such a systematization is what is provided by an interpretation (or, rather, the sum total of implicit perspectives and mental processes which make up the vast majority of the registered sense-to-cognized concept process One part of this, equally as important as what we call interpretation, is sometimes called coordination: a coordination tells us how to obtain determined mathematical objects from sense-data (and knowledge of an experimental apparatus), such as how the outcome of a Bell inequality test is supposed to tell us whether the actual inequality has been violated in that test or not; an interpretation tells us how to cognize nature through formal mathematical objects, such as what a violation of the inequality is actually supposed to tell us about reality. ); when they lead to calculations with genuinely different answers, you can go to the lab and shoot one down, but when they don't—consistently don't—the most we can say is that they're just different cognitive toolsets for thinking about the same thing. But cognitive toolset matters.

Fabricating Things

conceptsconsciousnessrationalityterminologylarge

The process of fabrication of "things" is extremely dynamic and context-dependent at multiple levels, but, since it can rely on higher-level cognitive systems, it's generally coherent to analyze it in terms of the extent to which it does so — fabrication is perception insofar as it doesn't rely on such systems, conception insofar as it does. Fabrications are always mental "things": my phone has a purple case, so when I see it lying face-down on the table I see it as purple, and I see both the phone and the purple through perception (they're forms I'm intimately familiar with and recognize instantly), but usually it is just the phone that is taken as a "thing" and is therefore a fabrication/perception; the purple, unless I explicitly mentally isolate it as I have here, is not a "thing" and therefore not a perception, even though it is perceived.

The initial act of fabricating a representative of some non-legible thing occurs hand-in-hand with perception, which kicks the process off by registering certain fundamentally identifiable properties of that thing such as its borders and intensities (in space, time). Analysis takes these properties and analyzes them in terms of whatever context it has been given in order to come up with an initial set of characterizations; engaging in a perception-conditioned development of these characterizations, it bootstraps a fabrication.

My use of fabrication comes from the Buddhist 'sankhara', sometimes translated as "fabrication" but more often as "volitional formation" or "conditioned thing"—it fits the concept I'm looking for near-perfectly. In both that term and my own, there's a strong emphasis on the perceiver's fundamental role in the process.

As a substantive noun, a 'fabrication' is an artificial construct created by mental activity which represents some thing and makes it legible to analysis. As a deverbal noun, 'fabrication' is the act of creating such constructs.

To reach into the interconnected flow of psychological activity and make a singular fabrication out of some chunk of the activity, calling it perhaps a psychological process or a psychological state, distinct from others, necessarily prevents you from perceiving the entirety of the nature and function of the chunk underlying the fabrication. Nature does not totally segregate itself into natural kinds which humans could theoretically use as scoops to losslessly carve fabrications out of the activity (in other words, nature is non-kind); further, there's no particular reason that the equivalence of the observer and the observed in this case should facilitate a segmentation of the activity into fabrications each of their own natural kind, not in the least because the actual faculty of analysis which in this case does the fabrication is an artifice built at the very top of the activity. (This "faculty of analysis", by the way, is itself a fabrication).

Nevertheless, we must do some fabrication if we're to do any analysis, since the analytical faculty needs to identify legible marks in the material with which it is to work, and the non-fabricated activity does not present such marks to it. We have to put some initial fabrication in to get an analysis out, and undergo the dialectical process to work out the kinks introduced by that original sin.

When speaking of concepts, we have to take care not to screw up the map-territory relationship: if I were to write "this or that material concept", I would be introducing a hypothetical cognitive attribution of substance to a singular undetermined concept, the materiality of the concept for its cognizer being unrelated to the content of the concept, and yet unless I take the example literally as I write that sentence (as I have just now, thus making the "singular undetermined concept" material for me), it is the undetermined haver of the concept that is attributing substance to it, not me; I merely take it as a kind of thing which particular actual concepts can be, so that I cognize the noun phrase 'material concept' as a formal object. This brings up another point - an undetermined object hypothetically posited by a cognizer, represented by multiple different referents in a sentence (as above, where the phrases "'material concept'", "singular undetermined concept", "the concept", "the concept", "'singular undetermined concept'", "the concept", "it", "'material concept'" were all cognized as coreferential by me), may vary in its materiality or formality across each independent referent in which it is cognized. This principle usually only comes into play when we're being way too meta, as now.

Bijectivity fails in the other direction too, in the sense that there are multiple different ways in which some 'thing' can be a "concept". Consider: is "17" a concept for me? In one sense, no, since I don't have any idea of seventeen-ness like I do an idea of twoness, threeness, even sevenness (picturing a heptagon in a sort of indivisibility). In another sense, yes, since when I think it I think it as a single object.

Impressions

psychologyconceptsterminology

Impressions are like perceptions, but without the connotation of being directly sense-based. They include mental movements, for instance.

The Virtual Dimension of Impression: there are many different levels of intensity of virtuality of the visual (sensory in general) imagination that only loosely correlate with detail [?]

The Conceptual Dimension of Impression: After I clean the frame of the dead laptop's monitor, I turn my eyes away from the black screen via a saccade around 20° laterally to the right, and it goes from the near-mid portion of my peripheral vision to the mid-far portion; very shortly after, I get an impression of a white Apple logo superimposed over the screen in my far peripheral vision, a bit to the left of where it would actually be. It has a high luminous intensity, nearly matching the actual logo, but a very low qualitative intensity; it's very brief too, around a quarter of a second, but both my minor surprise at it and my noticing that it isn't real happen within that interval.
The phantom image was obviously conceptually tagged — the first thing a Macbook displays when you turn it on is that white apple over a black screen — [?]

The Conceptive Dimension of Impression:
A conceptive figure which is kind of like the juxtaposition of $\left(\frac{1}{\sqrt{2}}, \frac{1}{\sqrt{2}}\right)$ against $(1, 0)$ and $(0, 1)$ -- a half-figure which is in its halfness weaker than either of the things it is between

I have a sense-image-conceptive-figure of taking something which is a sum or integral each term of which is a multiple of two varying things $p(x)$ and $q(y)$ [feels like $f$ as much as $q$], and separating out the $p(x)$ to sum over it separately because the structure of the sum allows that and it-is-motivated-by this simpler sum being given in a predetermined/axiomatic fashion, e.g. if it is a probability distribution and the total sum includes a sum over every $x$ then the sum of the $p(x)$ is 1.

it's not even clear that we can use noumenality, since the -ality conjugation gives us a particular tense which is conceptive; tenses are conceptive like ready-to-pounce or powerful-striking or receptive-negative_space-capaciousness or applicative-substancy [i have a conceptive figure of a thing which can be applied, which snaps-meshes onto other things like a female-socket snapping onto a male-joint]. and yes, when i think of why i use the -ality i think because it's a generic pudding-like concept which i have to represent in that way, so we can't use noumenality...

The Possibility of Knowledge

psychologyinference

It's commonly claimed that humans can't truly know anything about the world as it exists outside of our experience, as all that we have to form knowledge from are the innate structures of our brain and the fabrications that it creates, filtered as they are through the distorting and lossy processes of perception and cognition. On the other hand, though, the computer I'm typing this on only works because humans have learned how to carve metal at the nanometer level so as to allow it to conduct electrons in certain desired ways. Given that — this gap between internal attributions of truth and the actual, "in-itself" state of things in the real world (let's refer to the latter as "Knowledge") isn't really doing too well at preventing us from having knowledge of the outside world, is it? While it's obvious that the argument holds (assuming that external reality is really external, and not e.g. constructed by consciousness as it is in dreams), it is only true in the absolute strictest of senses.

That is to say, it's true that I cannot know anything about the reality outside of myself, in the sense of rightly coming to a 100% certainty about the relation of some things beyond my experience — in fact, it may be the case that this particular moment of my experience is all that exists, the sense of flow and long-term memory and working-memory and so on being illusions, so I can't even Know that I'm even a thinking agent — but the fact of this lack of Knowledge is both irrelevant and, in itself, pretty uninteresting. In fact, it's often pretty misleading, insofar as it causes us to downgrade the complexity and subtlety of the process of gaining knowledge by convincing us that there is no such thing as knowledge; this is the primary reason I've emphasized the particular meaning of the word that is denied us as capital-K Knowledge. There is no Knowledge, but there very obviously is knowledge, and this knowledge can (as in the case of computers) come to approximate Knowledge. So, rather than boring ourselves with the non-existence of Knowledge, wouldn't it be much more useful and interesting to study the existence of lowercase-k knowledge — the conceptual schemata according to which we come to form an understanding of the world, and the manners in which this understanding evolves as it comes into contact with the world?

Explanation-Liveness=Rules

metatheoryrationalityconcepts

An explanation for why the sky is blue, if it is to be a true explanation, should hold only in those cases where the sky is actually blue. It should not, for instance, hold for any planets where the sky isn't in fact blue, or even in any earthly places or cases where the sky is some other color (though you might need an "and there isn't any other physical effect overpowering this one" clause to be sure of that). Trueness of the explanation obviously doesn't require that the explanation holds in all cases where the sky is blue, since the sky might be blue for entirely different reasons (so that taking the original explanation to hold would result in a Gettier case).

Generally speaking, the substance of an explanation is the network of causal relations it delineates — so they, like causal relations, get precise meanings via their counterfactual semanticsTheir meanings aren't directly conceptual nor even ultimately conceptual, only fundamentally conceptual in the same way that the Mona Lisa is fundamentally physical paint on canvas. We can say things we didn't mean and mean things we didn't say, fail to make our models clear or even legible to ourselves, and so on. What we mean when we give an explanation can only live inside the conceptual structure underlying the explanation as we assemble it in our mind, but the (theoretical) meaning of an explanation, on which the validity of its relation to the world stands, is to be found in how it differentially applies among a system of counterfactuals..

In reality, explanations are always given in a context, a background of what is taken to be known, what is taken to be under question, and what is taken to be black-boxed; more fundamentally, what ideas and processes are taken to be relevant in the first place. An explanation of why a given historical event happened, for instance, is generally understood to have taken place within a technological, cultural, social milieu unique to that time; we might speak of an incident in Roman history in generalities that do not reference these specifics, but since history is not something we attempt to learn from but something we attempt to find and refute talking points in, it comes naturally to us to (try to) find differences that shoot down the validity of explanations (Gettier behavior, right for the wrong reasons). But when offered scientific explanations, we tend not to think about this sort of thing (sometimes, they're offered as paradoxes/tests of true understanding). To go back to an example from ECM: one might give the single word "oxygen" as an answer to the question "why does apple flesh brown?", but to what extent can this be made into an actual explanation? It's understood that apples are 'apples' in the sense of being a particular kind of fruit (and 'fruit' here refers to the conceptual classification we learn as children, rather than the botanical classification which includes beans, nuts, rice, and so on), and so if I were to ask "why don't white people turn brown when exposed to oxygen?", you might assert that apples browned because of something about their being fruit that caused them to interact with the oxygen in some particular way, pointing to how peaches, pears, and bananas also brown in order to make a reference-class level statement like "browning is a thing that fruits do when their flesh is exposed to oxygen".

And we know how this game goes. I say "but lemon", breaking this second explanation, you make... some specification or caveat which discludes a class of fruits including lemons, or some subtle perspective shift that happens to disclude lemons while including some new kinds of objects, But isn't this just a rule? What's the difference between an if-then rule and a genuine explanation, except that explanations are conceived of as speaking about "alive" processes in reality?

Dimensions of Consciousness

consciousnessphenomenologyterminology

Timelike dimensions

1. Rt, the retentive/protentive definition of time
• The singular consciousness is merely the "loudest" influence in retention memory, which is itself in a sense
• Retention memory has a differing legibility spectrum across Rt
• Cognition as it occurs is fundamentally illegible, and is only made legible for retentive (high-Rt) consciousness
2. Xt, the processing dimension of time
3. Pt, the physical dimension of time

"A second time dimension measuring the rate of progress along one's own worldline". Rt and Xt to some extent can be embedded into Pt, but with complex warping and pinching and blurring patterns. (The blurring is one of the reasons why only "to some extent": the spatial structure on Rt and Xt is not the Euclidean metric on the real line).

"nested" dimensions of Rt? at time X1 I'm conscious of having been at time Y conscious of being in state S1 at time Z; at time X2 I'm conscious of having been at time Y conscious of being in state S2 at time Z.

Spacelike dimensions

1. Dx, the depth dimension of consciousness
• towards low Dx is the preconscious, high Dx limited by cognitive temperature (or level of "magnetization")

Finite vs Infinite Minds

It seems to me that minds limited by their finite extent in this universe are pretty much necessarily sensory-intuitive rather than systematic-abstract -- I'll start with the postulate that minds, in order to carry out cognition at all, i.e. in order to have mental representations and a coordinated play of representations among one another, must be able to computationally grasp their internal activities as constituting representations, must be able to in some way perceive characteristics of these representations by means of which they can be related to one another, and must be able to computationally carry out the formation of these relations and use this formation-process to condition their further continuation.

It follows that there must be some facets of these representations that minds hang on to in order to recognize and employ them in their capacity as representations, and and some facets of these representations that minds hang on to in order to combine them in the cognitive process and analyze this combination to draw conditionalization from it. In finite sensory-cognitive minds, these facets are derived from the physical inherence of mind — in humans, for instance, mind is carried out by a complex series of electrical and chemical signals in a particular ball of flesh — and the simple fact of the interaction-according-to-rules of the underlying states of the sensory-cognitive mind, which are limited by their physicality, is the driving force by which sensory-cognitive mind can cognize anything at all.

To reiterate: in sensory-cognitive minds, again, these means are provided by simple physical interaction, which can provide a means because mind-states of the sensory-cognitive mind are fundamentally qualified by their physicality; in fact, the rule-basedness of physicality and the evolutionary-ideatic order of the physical instrument from which cognition arises are the necessary and fulfilled precondition by which sense-cognition can indirectly grasp the infinite, because the infinite somehow innervates physicality.

In contrast to this whole situation, the states of an abstract-systematic mind are not limited by their physicality, pretty much by definition (not that I've defined these terms) — how, then, could a finite abstract-systematic mind obtain a means for selecting such-and-such a particular representation, such-and-such a particular means of interaction among selected representations, such-and-such a particular schema for conditioning the continuing cognitive process upon the fruit of this interaction?

Their finiteness disallowing any infinite reach of their minds and any infinite rapidity of processing, finite abstract-systematic minds simply have no way in which they can cognize any one thing rather than any other thing — as such, they must be taken to be always cognizing everything-nothingness, making them no different from a rock. So, I conclude, finite minds are necessarily sensory-cognitive; contrapositively, abstract-systematic minds are necessarily infinite.

There's one clear prima facie loophole here, which is abstract-systematic minds directly acting on sensory-cognitive minds, and I'm not sure how to handle this. My first thought was that in order for such an action to take place in a manner beyond a merely contingent arising of a different series of sense-cognitions than would have arisen without the influence of the subject on the object, i.e. in order for it to communicate some aspect of abstract-cognition which fundamentally distinguishes it from sense-cognition, the object must have in their mind some faculty capable of receiving and recreating such non-sense-cognitive representations, and this is impossible because finite sense-cognizers simply cannot do anything with infinite (=unconditioned, universal) representations; I'm not too sure this is true, though: even finite sense-cognizers can in some way indirectly perceive the infinite, and this word "indirectly" has a whole universe of vagueness in it I can't unpack which could genuinely allow for productive action by infinite entities on finite entities. Actually, the Absolute itself, considered as the source of the universality with which physical laws act, and considered as the driver of the ideatic evolution of physical cognizers subject to natural selection, might be considered as acting upon us in order to show us the infinite. I'm not sure, so I don't want to say that sense-cognizers are necessarily limited to cognizing the finite; all that I feel vaguely okay saying for now is that finiteness lends itself to sense-cognition, and infiniteness to abstract-cognition.

Modes and Despairs of Categorization

Points of Characterization: Kinds of Character

1. Cognitive-Linguistic Character

If the thing is an action, property, or relation: what is the (linguistic, not hedonic) valence of the thing — how many other things are required to be combined with it in order to situate it in the real world?

(Note) Linguistic Valences: For ditransitive verbs, there is a subject, a direct object (the thing the verb directly acts upon), and an indirect object (the "recipient" of the verb) — for instance, with "I gave Mary a puppy", I am the subject, puppy is the direct object, and Mary is the indirect object. For transitive verbs, just the subject and direct object. For intransitive verbs, just the subject, and for impersonal verbs, nothing. For instance, "perception (n.)" is divalent: it is a perception of some thing (1) by some one (2). "put (v.)" is trivalent: some one (1) puts some thing (2) some where (3). Some general vocabulary: avalent = impersonal (to rain), monovalent = intransitive (to sleep), divalent = transitive (to kick), trivalent = ditransitive (to give). (note that in determining the valency of a verb we're counting the "core" arguments (who is sleeping?) rather than the "oblique" arguments (e.g. where are they sleeping?)).

What linguistic category does it fall into? verb? adjective? and so on. How can it be treated in terms of the other categories? (e.g., hypostasis).

2. Dimensional Character

What are those dimensions along which the thing can vary? Dimensions don't necessarily have to be copies of $\mathbb R$: they're manners of quantitative variance in general. Can the kind of thing be idealized by a finite collection of distinct states, such that any given element of that kind can be described as some particular combination of those states? If so, then how do those various states interact with one another? What are the patterns of simplicity and complexity in this multichotomy? If the thing is a property or descriptor, is it binary? is it continuous? What is its geometry? E.g., how is "propaganda" used — is something propaganda according to whether it meets certain criteria, or is it more propagandistic insofar as it satisifies certain criteria? What to do with understanding that a given dichotomy is simple or complex?

3. Functional Character

Find the right level and context to characterize the phenomenon. How to mark it as a substance, property, relation, etc.? How to, having marked it in this way, form versions of it serving other properties (e.g. hypostasis)? What are the manners in which it is applied? Can it be, or is it already implicitly, primarily known through these manners?

4. CFMP

(Taken from pdf) Here, we take some thing, and delimit it in terms of:

1. Its characteristic, or salient quality;
2. Its function, the concrete task it performs, thing it does, and/or goal it achieves;
3. Its manifestation, the way it presents itself within experience;
4. Its proximate cause, the principal condition upon which it depends.

I think it might be useful to divide characteristic into fixed characteristic, the salient qualities common to all members of its kind, and variable characteristic, the salient qualities common only to some members of its kind or which vary in intensity among members of its kind.

5. Manners of Definition

Up, down, and lateral definitions (from Aella):

• Up-definitions are pragmatic, and directly point to a manifestation of the thing in the world, or the role the thing plays in the world.
• Lateral-definitions offer synonyms, giving the thing a semantic location.
• Down-definitions offer a sense of what the thing "really" is, or is made of.

Points of Characterization: Questions to Ask

1. How many instantiations of the thing are there? Is there one or many?
2. How should the thing present itself to consciousness?

Points of Characterization: Understanding the Thing

• Perspectives of teleology: why must it be this way? Is there any essential nature to it? Any function it serves that it must be some way it is to serve? What are its contingent aspects?
• Investigation into what a word truly refers to, what the natural terrain of its concept really is, and rebasing from that more natural perspective.

Movements of Characterization

BinExt: a set of properties which are viewed as binary and mutually exclusive are instead viewed as continuous and not necessarily mutually exclusive. Ex., "is it like this or like that" becomes, after B/E, "how much is it like this and how much is it like that".

(On Renaming) When figuring out how to replace an old, less useful label, e.g. "knowledge", characterize the kinds of things that go under that label, and the reasons why they may do so.

Despairs of Characterization

1. The Characterization is Not Real: the real is unknowable
2. The Characterization is Not Complete (across different manifestations, contexts, etc.)
3. The Characterization is Not Stable: we move from a thought inducing a label to a label inducing a thought, which is not continuous with the original thought; the frame inducing the original thought is unique to the mind that had it, which existed at precisely one point in time and then changed into something else. Not only "meaning is ineffable", but "'meaning is ineffable' is ineffable".

Despairs of Understanding

1. Physical Understandings are Fake: evolution seems to do stuff, but this isn't its intent, and so on. identical to "the characterization is not real"
2. The True Physical is Too Complex to Understand: different stimuli constantly merge with and condition one another
3. The True Physical is Too Imperceptible to Understand: can't communicate symmetry-visuals in an image

Formal Despairs

Obstacles to knowledge inherent in formal systems. Gödel's incompleteness theorems, the halting problem & Rice's theorem, the Entscheidungsproblem, Tarski's undefinability theorem, in some cases the Lowenheim-Skolem theorem

Inherence-eternality-substance

phenomenology

There's this inbuilt category of inherence-eternality-substance through which i instinctively view the world, especially when I perceive it rather than merely sense it. When I place my hand on a table with intent, this instinct guides me to, rather than simply attending to the undecorated sense of that contact, reify this sense as being a manifestation of the Forms of the things involved: it's Me [Who Is!] Placing [Placing!] My Hand [Tactition! Perception!] On The [Solid! Corporeal!] Table. A play of eternal forms of which the now is a direct manifestation.

But that's not the way it really is. Rather, there's just a particular tactile sense, just a particular visual sense, just a particular volition and coordination and perception and attention having a little get-together to make an experience which a brain could tag as "me touching a table", but which tagging is really extraneous and unnecessary; there's nothing inherent or eternal or subsistent in it, it just is for that short while and then it isn't. There is a continuous flow of something that could be called a self — a succession of states of get-togethers of various senses combining in various ways, each state rigging the next to form a chain of agent-ness, but it doesn't have to have any sort of Internal Ultimate Meaning As The Self [Me! Look! Me!].

But this is such a hard thing to say correctly. What is the nature of this category of inherence-eternality-substance? For now, i want to say -- it's like a reverse Plato's cave: we think we're staring at a play of forms, when really we're just staring at ephemeral flickering shadows cast by fire...

The Mathematical Imperative

mathematicsmetatheoryrationality

The mathematical imperative, metatheoretically, is to formalize: all conclusions about things must be expressed in terms of the necessary properties through which they obtain.

The mathematical imperative, psychologically, is to turn all patterns into utilities: what shows up

1. You can't really say these are different ways of saying the same thing, because there is no thing that they are independent of the cognitive processes deploying them (and they're clearly separate within said cognitive processes: if this weren't the case, we wouldn't have two things which we could say are "the same") — the same thing is a focus imaginarius, and if you think that they're the same thing there's just you thinking "same" in reference to some things. Sherlock does not do anything and is not anything, there is merely ink on a page which we shape into "Sherlock does this, performs this action", video footage of hominids in costumes which we shape into "Sherlock is this, exhibits this property" —— in the same way, the things are not anything, there is merely cognitive processing which we shape into "it is this". Until this is burned into your brain you'll just keep on eating the onion.
2. This is a fundamental skill to introspection. Lots of mentation is really just disordered slop not amenable to questions like "why did you do that", and in fact it immediately illuminates itself as such, it presents as disordered slop, but tailing-retracing conceptual cognition tends to assume that there is a coherent legible reality underlying it, which there just isn't.
3. But be careful with it. There is genuine transcendence, and overemployment of voidness will sweep it under the rug. Use it as an escape route rather than a bludgeon, honing your "doesn't quite fit"-sense and training yourself to instinctively step back and look at the concepts you're using as concepts whenever a situation (inconspicuously conceptually phrased) doesn't really seem to be shaped right — like when a few errors in a jigsaw puzzle cause the surface to warp!
4. (I am aware that what I mean by 'void' lines up with the Buddhist notion of sunyata (sometimes actually translated as voidness, but I didn't know that until after I picked the word); I am also aware that if I commit to any relation less ambiguous than "lines up with" then I'm gonna get dogpiled by people puffing their chests and saying you'd have to be a moron to believe that's what sunyata really means. So I wish all Buddhists a heartfelt "go away").

Moral mathematizing then looks like the following line of thought: if eating sausages is wrong, there's a particular reason that it's wrong — maybe because it tends to cause animal suffering, in which case "eating sausages is wrong" is only a general heuristic, and sausage-consumption causally separated from animal suffering (e.g. a sausage just appears due to ridiculously improbable quantum bullshit, and you eat it) is fine. And there's obviously a whole garden of forking paths here. One could respond "it's still bad because it cultivates the habit, i.e. raises expected future animal suffering" or "it's still bad if you don't know that it's not causally separated, because morality obtains at the level of decision making, not consequence" or many other things, and each of these would merit its own multiplicity of responses.

Dual-Person Pronouns in Old English

languagefacts

Phenomenal Field Enrollment

phenomenologyterminology

FIELD ENROLLMENT: Is a process with two main periods: Lower enrollment, by which the brain forms stimuli into a flow of senses in particular fields, applying schemata like "flow" and "this field". Transcription of stimuli into neural sense signals. Upper enrollment, the process by which the brain organizes sense into a perceived world of objects with properties in relations. (e.g., seeing a book on a shelf as a book on a shelf).

FIELD REGISTRATION: A perceptual classification of some field stimulus that guides its enrollment into that field. E.g., walls and some furniture are commonly visually sensed in a background register, in which they serve as relative (pro-prepositional) determinants for more salient objects. But this varies as attention shifts from considered object to considered object, e.g. if a book is placed on a small table itself situated awkwardly in the corner of the room, it might happen when seeing the book that it is registered as foreground as the table is registered as background concurrent with the perception of what is to the brain a book on a table and perhaps the apperceptual recognization of that book (giving attention to it). But if attention shifts from the book to the smallness of the table it's on — the table having initially entered salience as the background for the book — that table might be registered as foreground as the salience of the book is dropped or deferred entirely and the corner (walls and floor) it's in are registered as background concurrent with the thought {that table in the corner is small}, i.e. the apperception of the table in the corner as small. Then attention may shift to its relation to the corner as a location, and in particular its awkward situation, [?]

On Batman

large

I've been playing through the Arkham series of Batman games recently, all of which have roughly the same structure: shit goes down somewhere, and, over the course of a single night, Batman has to fight his way through and around the open world, defeating between seven and sixteen of Gotham's many supervillains. It's a series where lots of bizarre and unnatural things happen, and where Batman's saved time and time again by unrealistic deus ex machinae, but I can generally keep these from breaking my immersion. But what instead struck me was how absurdly hypercompetent, resourceful, and tenacious most of these villains are. Half of them are described as "being trained to physical perfection", generically intelligent characters are invariably prodigies with eidetic memories, and many of the supervillains happen to be world-class experts in multiple fields While looking through the in-game character biographies, I also noticed that almost all the male characters' heights are above 6', and the women are almost all 5'7" or 5'8". In terms of realism, there obviously is some selection effect which would bias towards characters being tall—the same effect that causes U.S. presidents to be unusually tall—but it wouldn't turn out like that. .

In Arkham Knight, the Riddler takes three months to build an army of semi-intelligent humanoid robots capable of hand-to-hand combat against Batman—meanwhile, Boston Dynamics has spent years building robots that can... sometimes run without falling over? Scarecrow, former professor of psychology, develops an aerosolized psychotropic capable of instantly inducing extreme terror with a single whiff (and it's so good that gas masks don't work!); give psychiatrists thirty years to do the same and you might get a pill that works slightly better than placebo (p=0.043). If they use firearms, they happen to be deadshots; if they fight physically, they happen to be gymnasts with black belts.

The rest of humanity is just normal. Normal people with normal physical and mental skills and normal technologies, entirely incapable of doing anything to stop these titans Part of this is because the skills that characters display are made to fit whatever actions the writers want to depict them doing, rather than the other way around (as it is in reality, where skills come before actions). It would be uncool if Batman just !c{flashbanged}{or tased, tear gassed, etc.} the Joker before breaking all his ribs with a flying kick—so he's got to have incredible resistance and fighting ability (and pain resistance/healing factor when he eventually does get kicked, so he can run away and come back next week perfectly fine)... . In this sense, I think of the Batman world as "K-selected", in a sense similar to the evolutionary one. (This isn't really the best term, but it's the one that's stuck in my head despite persistent attempts to replace it). The course of the world, rather than evolving according to large-scale trends embodied by governments, militaries, technologists—or societies, cultures, nations, peoples—is guided by a few select super-people, more or less sporadically chosen from the populationMetal Gear Solid offers a really interesting twist on this r/K narrative dichotomy; in most games the protagonist is a single legendary soldier who somehow ends up preventing all-out nuclear war by single-handedly fighting some larger-than-life supervillain, and you'd suspect K-selection from this, but... one of the core questions MGS repeatedly asks is why people do what they do, and the answer Kojima keeps coming back to and reinforcing is that they're reacting to the ideas, stories, roles, and purposes thrust upon them by the very times in which they live; they're making sense of the world in a culture-bound way, and cannot help but to relive and reinforce the narratives of the culture itself. And the Patriots, by MGS4, simply are the times themselves; they are the r become K..

When the world's narrative is K-selected, it does make a bit more sense to attempt to use your billions to fight crime by training yourself to physical and mental perfection and developing a bunch of tech with which to directly fight people. But it's still obviously wrong.

[...]

Let's differentiate between unrealities that are fixable while keeping the moment-to-moment narrative intact, but just there for gameplay value, and unrealities that are not so fixable. Playing as Batman, you have an infinite stock of what ought to be disposable gadgets (explosive gel should deplete, batarangs should run out), but this is a simple fix: you really don't have to use many gadgets in any given encounter, and it would be simple for Batman to set up caches or configure the Batwing (a bat-shaped jet used for quick transportation and reconnaisance) to launch supply drops. We can imagine a world where this unreality is fixed while Batman still does more or less the same things. Another unreality: these gadgets don't take up storage. His suit isn't thick enough to conceal so many gadgets in such readily reachable places while also being fitted throughout with bulletproof armor, but we can't imagine that they're on the outside and just not rendered, since no rifle bullet ever prematurely detonates a freeze blast, jams the batclaw, or destroys an electronic device (as would happen frequently if he stored so many gadgets outside his suit at center mass). Yet this is another simple fix: we can imagine a world where the suit is slightly bulkier so that some gadgets can be slotted or embedded in it, where the gadgets are designed to fail gracefully and detectably so they can be replaced wholesale via supply drop, where some of them are combined into weird but space-saving multitools, where Batman is slightly slower to grab a gadget he rarely uses because the more commonly used ones get priority on easily reachable places.

Meanwhile, what is an unfixable unreality? Here's one: Batman just has a health and armor bar, and they both regenerate after every major encounter. Over the course of Arkham Knight, you'll be shot probably hundreds of times and hit with iron bars, blades, stun rods, and riot shields hundreds more; all you need to do is survive each individual encounter, and you'll be in perfect condition for the next. The game canonically takes place over a single night, so there's no downtime, let alone time for surgery or recovery. Even if Batman blows through bulletproof suits like toothpicks, this is unrealistic: a rifle round will still bruise (it's almost always a rifle, shotgun, minigun, or machine gun firing at you—handguns only appear in cutscenes!), and by morning Batman would just be pudding in a suit. There's really no way to fix this except for Batman to just be consistently perfect to the point of precognition and paranoia, since getting shot at is kind of just a thing that happens when you're Batman, and the only realistic solution is to not be Batman against hundreds of armed soldiers in a single marathon.

Folk Neuroscience

conceptsrationality

If we did not start with a 'folk architecture' of mind, neuroscience would be entirely meaningless. We'd see cool waves and patterns, but connecting these to the kinds of phenomena we're capable of conceptually dealing with, such as "cognitions", "emotions", "memories", and so on would be beyond us. The best we could do is notice that some subjective phenomenal states seem to correlate to some objective neural patterns, and this would be done essentially randomly, since we refuse to extract data from introspections except as arise through pre-established correlations with neural data.

Actually, it would be worse. The stronger norms against using introspective approaches are, the more people conceal the approaches they use, from not just others but themselves. The objectivity of a conscious phenomenon is immediately convincing to a consciousness that has objectified that phenomenon, and in fact it skips over interrogation to automatically make itself available to further cognition as an object; thereby do we think that "thinking is an objective thing people do, it is an objective function of the brain, it is objectively the thinking of certain thoughts, it objectively proceeds from prior thought to following thought", when really the carving out of "thinking" from the whole of mentation is largely linguistic happenstance driven by introspective awareness. If you really want to cut introspection from neuroscience, then you should want to destroy neuroscience altogether.

The Transcendental Impossibility of Magic

metatheoryrationalityquantum

It initially seems obvious that the discovery of 'magic' tomorrow would not violate physics per se—it would violate known physics, but ultimately extend the project of physics. For, whatever magic is, it would behave according to rules, and we can learn these rules well enough to understand the nature of magic as well as we understand any other part of nature.

One irritating feature of this argument is the way it relies on the way we cognize the world, rather than any feature of the world "in itself". Why would magic have to act according to rules? "Because everything else we've encountered in this world does, and so by induction magic must be the same!" is such a pathetically weak argument; (a) the only things you say we've encountered in this world are the things we conceptually delimit by virtue of their existing within systems of rules, (b) would you even try to recognize if there were something in this world that didn't act according to rules, or would you just pretend to ignore it?, (c) couldn't an aspect of this world that didn't behave according to rules appear to us in the same way that the lack of a fundamental theory of physics, or fundamental interpretation of quantum mechanics appears to usTo be clear, I'm not saying that "the world doesn't fundamentally operate according to rules". I suspect the truth is something more complicated (see the thought "when humans create physics"), and that indexical facts (the fact of my being me, of being this me that sees the quantum coin flip come out to zero instead of one) are resolved by a sort of symmetrical reductionism—if the wavefunction has another me that thinks themselves to be this me that sees a one instead of a zero, then indexicality is just an illusion, a matter of artificial perspective. I'm saying that we can't fundamentally be so sure that we are built so as to be able to cognize all of this correctly via our schemas of 'rule', 'cause', 'reason', 'law', and so on.?

Here's one interesting possibility: we're in a simulation, on a computational system in Outer Reality that proceeds according to that reality's physical law and which computes our reality according to our physical law (churning out processes according to some algorithmically encoded ToE); for convenience, let's say that the physical laws are the same, and the computer is made of transistors and programmed by humans with neurons, et cetera. But one of the sysadmins is katy t3h PeNgU1N oF d00m, who likes to make ad-hoc modifications to the bit-level encoding of our universe whose effects are t3h raNd0mz as f. So occasionally the LHC will collide two beams of heavy ions and get a fistful of live caterpillars instead of a quark-gluon plasma—they'll see the nucleons assemble into cells and everything—and so on. Such a possibility would ultimately follow physical law, since katy's decisions are ultimately decided by neural signals that proceed according to physical law, but the pattern of these decisions can't be correctly explained by any physical theory that doesn't at least model these neural signalssince that's just how the brain works. it doesn't follow large-scale laws like a gas in thermal equilibrium does—at least, not any laws with perfect predictive power. knocking a single synapse off balance can lead to a butterfly effect which might at least slightly alter the timing of one of the random interventions.. And we could make the problem much worse by uploading katy as a !c{daemon}{As in Linux machines, not Solomonoff inductors}—a background process parallelized throughout spacetime that just constantly adds little bits of concept-laden randomness wherever seems like a fun place to do so. Such a process would, for any intelligence below AIXI's level, be totally indistinguishable from 'magic'. We can hardly figure out our own laws of physics, let alone that of a reality only made available to us through virtualization and then occasional randomization. And the problem gets worse if katy decides to be adversarial for the lulz. Maybe you find that you can actually bend spoons with your mind, unless there's anyone watching or recording, so that you can't persuade anyoneOf course there are ways around this. Have the skeptic put the spoon in a fiberglass safe, or paralyze you with curare and a straitjacket, or hook you up to an electromyograph, or blindfold you and secretly make the spoon red-hot, or tether it to the ceiling of a large, empty room, or all of the above, before leaving the room for thirty seconds. But katy operates by her own rules, not the image of them you're building and trying to work around; she could just cause the bending to fail whenever it's done with the intent of convincing someone else, and she could cause it to go horribly wrong if you started using it systematically to make industrial processes more efficient. More likely, though, it'd be something more r4nD0m than either of these already-existing tropes—something not so easy to catch on to and predict.. Maybe katy specifically screws with the processes through which we form our understandings of physics, such as by targeting experiments to make them resolve as though phlogiston and the luminiferous aether existed, or inserting the word 'not' in the worst possible place of the occasional scientific paper. And so on.

There's another aspect of the situation which is due to the limitations of our own conscious experience—it is not capable of experiencing physical law as physical law. To take an extreme case, suppose that I'm in an Everett timeline where the air on my desk spontaneously manifests into a tropical parrot—or, rather, that my consciousness, which is more or less consciousness of a single temporally ordered sequence of events, becomes aware of physical consequences that irreversibly bind it to such a timeline. From the third-person point of view,

Language of Abstractions

languageterminology

A field is a formally subformal object — it is a thing given with, whether explicitly or implicitly, some larger thing (its possessor), which derives a specific determination (value) from the concrete determination of its possessor. For instance, the generic person has mass as a field; this or that concretely determined person has a particular value giving their mass.

An -ity (-ivity, -ality, -icity, etc.) of a thing is a manifold in which instances of that thing may easily be delineated and used to characterize the manifold°, but which does not necessarily present naturally as a union of such singular instances. So we can speak of a conceptuality, or a manifold which may be characterized by the concepts we can delineate from it.
The countable noun x-ity is different from the uncountable noun x-ity: the latter resembles the -hood or -ness suffixations, and (generally) references the formal conditions underlying the presence or employment of x.

° It is sufficient that we can conceive of the characterizing delineations of the x-ity as being given in the same moment, or shortly after, the conceptual formation of the x-ity itself. For the concepts from which the x-ity is built may evaporate or change very rapidly, as might the x-ity itself. This is one reason why the x-ity may not seem to present naturally as a union.

• It (in itself; external, fictional, and unreachable), the itivity from which we delineate it, our delineation of it ('it'), the delineativity from which we delineate this delineation of it, our delineation of our delineation of it ("it")

The materiality of a material object necessarily rests on its being a this, and consists of the manners in which the mind cognizes it through the framework of bare virtual physicality.
This is separate from whether the object is being projected on to some (formally or materially) actual state of affairs, or obtention-context, or materialized formal determination. Insofar as it is projected in such a way, it is said to be concrete; insofar as it is not, it is said to be abstract. Insofar as a concrete object is conceptualized as being received rather than generated (e.g., the mental object I label "memory of my hairbrush", which I conceptualize as deriving its properties from the corresponding real-world object), it is said to be a (concrete) model object.
If I think "a particular schema", I am thinking of a materialized formal object, but it is still abstract since there is no materialized formal determination of it; when I hold a given manifestation of a schema as a concrete model object.

A particular schema, this schema, is a materialized formal object in the moment I think of it as being a particular schema with a particular extension. When I zoom back into it, neglecting the extension to focus on the intension, it just becomes the schema, and it loses its materiality.

• To go from "X is a set with at least one member" to "Let x be a member of X" is known as existential instantiation.
• To go from "this x in X has this property" to "some x in X has this property" is known as existential generalization.
• To go from "an arbitrary element of X has this property" to "all elements of X have this property" is known as universal generalization.
• To go from "all x in X have this property" to "this x in X has this property" is known as universal instantiation.

Affix and Word Choices

Prefixes

• Fate
• Depending on chance: aleat[/o]-, tycho-, temer-, sor[c/s/t]-
• Related to time: chrono-, horo- (conn. a unit), tempo-
• Work
• Governing, ruling: agon-, arch-, dom-, gov-, magist-, poten-, regi-, tyran-
• High: alt[/i/o]-
• Open, public: agora-
• Struggle, contest, contention: agon-, ath-, athl-
• Defending, warding, keeping off of: alex[/o/i]-
• Rough, difficult: asper-
• Stiff, unmovable, adhering: anky[/l/lo]-
• Do work: ergo-, ergasio-, labor-, oper-, pono- (conn. toil, suffering), urg- (also -urg[e/ia/ical])
• Perform a function: funct- (also -funct)
• Order: taxo-
• Power, strength: dyn[/a/am]-
• Life
• Human: anthropo-, homo-
• Female: fem-, gyno-, mulie-, virgo- (conn. marriageability/maidenhood), anil- (conn. elderliness)
• Male: andro-, vir-, arrheno-
• Animal: faun- (also -fauna), therio- (also -therid) (conn. wildness), zoo-
• Budding, sprouting, flowering: anth[o/er]- (also -anth[y/ic/us/ous/emous])
• Living breath, soul, conscious mind: anim[/a]-, noo-, spiro- (also -spire), pneum-
• Life, aliveness: bio-, vita-, viva-
• Growing, nourishing: alesc-, alit-, olesc-
• Ancestor's ancestor: atav-
• Physiology
• Related to the head: cephalo-
• A body: corp[/o]-
• Related to cells: blasto- (also -blast)
• Breath, breathing, air: pneum-, pneo- (also -pne[a/ic], spiro-, psycho- (ancient Greek), hal- (also -hal[e/ing/ant/ation])
• Remedy, cure: ac[o/eo]-
• Felt, perceived, sensed: [a/]esth[e/esio]-, senso-, pass- and pati- (conn. enduring), path[/o]-
• Pain, hurting, injury: alg[o/esi]-, dolor-, noci-, odyno-, poeno- (conn. punishment)
• Sharpened, pricking: ac[u/ut/uti/uto]-, agu-
• Bending: flect-
• Carry: fer-
• Sensation
• Smelling of something, or the act of smelling: arom-, odor-, olfacto-, osmo-, osphresio-,
• Pleasant: ameno-
• Sharp, acrid: acerb-, aceto-, acid-, oxy-, pung-
• Dry: arid-, xero-
• Cold: cheimo-, [c/k]ryo-, frig-, gel- (conn. congealing, freezing solid), hiber-, pago-, psychro-, rhigo-
• Dirty, disgusting: rhyp-, rup-
• Stinking: brom-, ozon-
• Burning: caut- and caust-
• Leaning: clin- (also -cline)
• Curved, crescentlike: menisc[/o]-
• Lacking strength: asthen[/o]- (also -asthen[ia/ic])
• Large, spacious, sufficient: ampl[/i]-
• Bottoming out or bottomless: abysso-,
• Raylike, radiating: acti[s/n/no/ni]-
• Dark, hazy: achl[u/y/yo/ys]-
• Deep: batho-, bathy-, bentho-
• Oceanic: thalasso-, pelago-, halo-, mar[e/i]-
• Boundless, infinite: apeir[/o]-
• Unconcealed, real: aleth[/o]-
• Psychology
• Mental faculty, intellect: menti-, noo-, nous-, noe- (also -noia), phren-, psych-, thym[/o]- (also -thymi[a/c])
• Memory, remembrance: memor-, mne-
• Forgotten, obliviated: letho-, oblivio-
• Relating to belief: cred[/o]-
• Love, fondness: agape-, amat-, philo-, vener-, amor-, am-
• Deranged, crazy, mad: deliri-,
• Deceitful, fraudulent: apat[/e]-
• Dull, stupid: ambly-, moro-
• Uneasy: anxi-
• Cunning, crafty: astu[/t]-
• Acute, sharp, perceptive: ac[i/ies/icul]-
• Friendly, bearing goodwill, peaceful: ami-, amic-
• Faith: fid-
• Relations
• Unequal, asymmetric, unlike: aniso-
• Uneven, abnormal, irregular: anomal[/o]-
• Incomplete: atel[/i/o]-
• First in time, original, primeval: arch[a/e/i/ae/eo/aeo]- (also -arch).
• Beginning, first in sequence: alpha-
• Causing: aitio-
• Directed towards, added to, near: ad-
• Setting in motion, acting, conducting: a[g/gi/git/gen/ct]-
• Directed away from: a-, ab-, abs-
• Separate from: ap-, apo-
• Lacking: a-, an-
• Before, prior to, in front of, forwards of: ante-, antero-, anti-, pre-, pro-
• Opposed to, preventing of: anti-, ant-
• Change, alteration, exchange: amoeb[/a/i/o]-
• Segmented, chopped up, point of articulation or joint: artic[/ul]-
• Highest point, prime, pinnacle: acme-
• Peak, extremum: acr-, acro-, apex-, apic[/o]-
• Suitable, appropriate, fastened, joined: apt-, ept-
• On both sides, around, surrounding: amb[/i/o]-, ampho-, circ-, circum-, gyro- and cyclo- (conn. spinning around), peri-
• Uncovering: apert-, aperi-
• Upwards, backwards, against, again: an[/a/o]-, am-
• Ringed, ringing around: annul-, anul-, ano-
• Numbered: arith-, arithmo-
• Another, alternate: ali-, allo-, alter-, allelo-, hetero-, mut- (conn. change), poikilo- and vari- (conn. varied)
• Attitudes
• Good: agath[o]-
• Best: arist-
• Abnormal, perverse, alien: allotri-, barbar-, xeno-
• Earth
• Of the land, ground, earth, soil: agr[a/i/o]-, chthon-, lut-, pedo-, pel-
• Place, land, territory: choro-, epeiro-
• Fire: ars-, ard-, flagr-, flam-, foci-, fulg- and fulmi- (conn. glowing, shining), gehenna- (conn. hellfire), phleg- and phlog- (conn. heated, inflamed), pyr[/o/eto/ito]- (also -pyrexia), volcan-, vulcan-, ign[/i/is]-
• Burning, kindling, lighting up: cau[/s/t/st/ter]-, crema-, aether[/o]- ff
• Ashes: cine[/r]-, spod- (conn. waste)
• Shining, glowing, bright: aether[/o]-, fulg- and fulmi- (conn. burning), luc[/o/i]- (also -lucen[t/ce]), lum[/en]- (conn. also lamp, heavenly body), lun[/a/i/u]-, lust[/r]-, pheng[/o]- (conn. sunlight), can[/d]-, cend-
• Lightning and/or thunder: bronto-, astrapo- (conn. hurling lightning), ceraun- and keraun- (conn. crushing boom of thunder), fulgur- (conn. flashing lightning), tonitr[o/u]- (conn. thunder)
• Month, moon: men[i/o]- (also -mena), selen[/o]- (conn. moon)
• Wandering, drifting, planetlike: plan[/o]- (also -plan[y/ia])
• Star-like, related to space: ast[r/er/ro]-

Suffixes

• Having the quality of: -ace[an/us/ous], -acity
• Inclined to or abounding in: -acious
• Highest point, prime, pinnacle: -acmic
• Raylike, radiating: -actinal
• Sharpened, pricking: -cusis
• Having the state, quality, or condition of: -[a]cy
• An action done, or the product of a process: -ade
• Quality of, process or product or location of: -age
• Leading, guiding: -agog[/ue/ic/uery/y/ic]
• Pertaining to, characterized by, belonging to: -[/i/e]al
• Divining, prophesying, interpreting: -man[cy/cer/tic]
• Feeling, perceiving, being afflicted by: -path[ic/ology/etic/ize/y]
• Hurting, suffering: -alg[y/ia/ic/etic/esic/esia], -agra (conn. sudden flaring)
• Specializing in something: -an
• Connected with, belonging to: -an, -ane
• The quality or state of being or fact of obtaining: -ance, -ancy
• Inhabiting, living in or on, occurring in: -col[a/e/as/id/ent/ine/ous]
• The agent or tool for some action: -ant
• Of the nature of, or the agent of: -ar
• Someone who regullarly performs or is characterized by an activity: -ard, -art
• Related to or connected with something: -aria
• A person that is part of something, with a certain connection to something, or in a certain state: -arian
• A place for or connected to something: -arium, -aria
• An adherent of a certain doctrine or custom: -ast
• A medical state or condition: -asis
• Having the character of, or pertaining to: -ary
• Inferior, small, shallow, insufficient: -aster[/ing/ism/y], -astry, -astress
• Office holder: -ate
• Cause or instrument of something: -ate
• Tending to: ative
• Pertaining to, or of the nature of: -atic
• An action, process, state, condition, or result of doing something: -ation
• Warlike: -machy

Collocations

• Collocations
• Type: kind, variety, class, category, genre, genus, species, family, order, strain, style, model, cast, mold, pattern
• Identity: standard, nature, style, manner, model, cast, mold, archetype, pattern, paradigm, prototype, identity, likeness, semblance, image, character, shape, essence, form, design
• State: state condition predicament circumstance situation accident
• Disposition: temperament attitude outlook perspective
• Faculty: faculty apparatus capability facility potential equipment aptitude ability proficiency propensity potential capacity power wherewithal instrument tool mechanism device competency aptness skill talent department means provision outfit setup system organ
• Tendency: tendency proclivity proneness inclination disposition predilection susceptibility habit
• Configuration:
• Complex: compound amalgam mixture concoction bundle apparatus organization faculty
• Property: quality characteristic attribute trait element aspect feature facet detail factor consideration circumstance setting value
• Object: thing artifact contrivance item entity being substance material body structure
• Image: depiction profile account impression idea view conception formation sense intuition apprehension notion outline figure precis characterization likeness representation rendition portrayal frame shape form configuration construction structure arrangement assembly
• Outline: silhouette, profile, figure, shape, contour, form, delineation, configuration, framework, diagram, abstract, essence, sketch, scheme, trace, plan, recipe, idea, blueprint, frame, build, model, order, system
• Model: representation, reproduction, facsimile, type, style, mold, template, framework, pattern, design, guide, blueprint, paradigm, illustration setup, system
• Pattern: design, decoration, motif, marking, ornament, adornment, embellishment, shape, influence
• Disposition: temperament, nature, character, constitution, makeup, grain, humor, temper, mentality, turn of mind, kidney, arrangement, arranging, disposal, ordering, positioning, placement, organization, configuration, setup, lineup, layout, array, marshaling, mustering, grouping, gathering, dressing, disposition
• Manner: way, mode, means, method, system, style, approach, technique, procedure, process, methodology, modus operandi, form, routine, demeanor, air, aspect, attitude, appearance, look, bearing, cast, deportment, behavior, conduct, comportment, mien
• Context: background setting condition climate milieu habitat
• Cause: constrain control condition govern determine decide affect effect act on change alter modify transform form shape guide bias treat prepare prime temper process acclimatize adapt adjust season

Crystalline Lens Exercises

perceptionphenomenologyfacts

Find some small object on a wall between one and three meters away, focus one eye on it and close the other, and place the tips of your pinkies around eight inches away so that they're aligned on either side of the object. Then, shift your focus between the midpoint of your pinkies — focusing on both pinkies at once — and the object in the background. I think it's important that the exercise require that you not focus on any specific pinky, but instead guide your mind to focus on "both at once", so as to develop the capacity to bend the crystalline lens without having a specific object to focus on.

Autonarrativization

phenomenologylarge

One of the prominent effects of intense sleep deprivation is, ime, the tendency for mind to "narrativize itself". There's a tendency for new thoughts to, instead of performing their intended function, encapsulate and reflect on what is perceived to be the immediately prior thought/experience as though it were a beat in a story. This is very similar to a phenomenon which I've noticed happens to me during meditation when I'm very tired. To describe the phenomenon: suppose I watched a movie earlier that day that involved a lot of talking between characters A, B, and C. During meditation, I might have a normal thought, like "oops, getting distracted, I'll get my attention back to the breath" (distraction being the precondition for such thoughts, of course); this phenomenon causes the next thought to implicitly insert me into that set of characters as my own character D, and insert that previous thought into a flow of conversation, to which the other characters naturally respond. So I'll think "okay, I should go back to focusing on my breath", just in the ordinary first person, and then the narrativization takes over without my realizing it, with A responding "Perhaps if you focus on your breath more, it'll help us solve this mystery!"; B goes "I think it's over in the kitchen, I'll go look for clues"; C's face lights up in some sort of flash of realization; D says "Hmm, but maybe I shouldn't focus on my breath...", and I'm just watching the conversation instead of focusing on my breath.

Notably, the talking of D often isn't really my own talking, but rather a vague simulation of my own talking in the same way that I non-consciously and vaguely simulate the speech patterns and characteristics of A, B, and C in order to make them talk. Phenomena that happen in two separate contexts generally aren't accidents — usually when we look closer we find some reason that those contexts incline the mind towards those phenomena. The one common factor that I can think of here is the sense of "looseness", which is really important to describe. As a first approximation, I'll say it like this: the dexterity with which attention grasps onto objects is greatly reduced. If we metaphorically picture attention as a part of the body (seeking to represent its varying kinds of dexterity), then I'd say that sleep deprivation tightens it up, in the same way that excessive cold tightens up the hand, whereas sleep deprivation loosens it up, in the same way that mild numbness loosens up the hand.

Why should tightening/loosening of attention cause narrativization? Actually, first I need to find a better phrase than "tightening/loosening" which views them both as opposite moments of the same kind of movement... but what is that kind of movement? This is the factor of sustained attention (vicāra), I think, which i have elsewhere diversified into a continuous axis of sensation-properties parametrized by voluntariness, calling focusability and fixativity the tendency of a given sensation to voluntarily/involuntarily sustain the attention of mind; extending this further along the low-high intensity continuum and considering only those contributions to focusability/fixativity due to the state of the mind itself gives us a quadrant of global modulations of attention. Here, tightening is a global raising of fixativity, while loosening is a global lowering of focusability, neither of these being "good"; the other diagonal, which runs from global lowering of fixativity (equanimity) to global raising of focusability (concentration), is "good". I think that the tightening/loosening dichotomy can further be unified by developing an insight into this quadrupole moment that turns the diagonals into straight lines.

The insight is obvious — the good path involves increased control, the bad path involves decreased control. So we can simply say, "why should decreased control over attention induce autonarrativization" (in certain contexts, and in me specifically)?

When is Selection Bias a Problem?

inferencemetatheoryrationality

Consider: when is selection bias a problem? To clarify, let's suppose that sampling method X is used for a study that aims to characterize Y. Why — and, therefore, when — do we think that X might have an effect on Y?

First-order answer: when we notice a conceptual similarity between some of the causal factors behind each. If we are looking at political polling by calling random landlines, we first conceptually set up X = "landline ownership" and Y = "political position" † [I'm not sure why this dagger is here. An ambient threat?]

Really, both are finer — it may be the case that landline response rate has an effect on political position, and that landline ownership correlates with poll question interpretation, so that you must add these two factors to the concepts of X and Y — but then there are even more factors delineable in the differential, and when you add those there are more, ad infinitum. X and Y are exactly what they are, no more and no less, and they are not what you conceptualize them to be: even if you take their general form and the instance they are to be applied to are given conceptually, and you can hold the form of the instantiation conceptually, you cannot actually evaluate it without falling into error. The concept of "f" and the concept of "x" allow you to form the concept of "f(x)", but they do not allow you to actually find the thing that is f(x), because the true evaluation of this thing must inevitably interact with the real world which is beyond any conceptions of it.

(c.f. Kant's "the concept of '12' is not to be found in the union of the concepts '7', '+', and '5': it's our synthesis of these concepts, utilizing the prerequisite aesthetic conditions for the possibility of mind such as the imagination of a communing of separate amalgamations of discrete things differentiated via our imagining them in space, which derives the concept of '12'; and because these fundamental conditions of mind hide themselves from us by virtue of their not being reducible to any smaller components through which we can recognize them as they exist immanently in our minds, we're forced to deduce their presence transcendentally, in absence of which deduction we're liable to mistake '7', '+', and '5' as giving rise to '12' analytically, thereby mistaking '7+5=12' for an analytic a priori proposition when it is really synthetic a priori".

Mind Twists Itself Into Knots

rationalitypsychologybiology

The mind will spontaneously and effortlessly tie itself into knots, these knots being the things it implicitly assumes must exist, the imaginary foci. Because the world does not intrinsically have these knots, instead being untangled in its self-equivalent immediacy, the mind causes suffering and error for itself through these knots as it attempts to add to the world what was not there in the first place.

Were the world truly meaningless, this would be a comedy, but it's more like a tragedy — the self-equivalent immediacy of the world gives rise to form which is simultaneously both fictional and non-fictional and neither fictional nor non-fictional. These knots don't just form randomly, and often point at meaningful patterns in the world. But they are knots all the same. If you want to understand the world, you have to start by looking at it with fresh, unassuming eyes: by detangling your mind.

What is the meaning of this? What is the purpose of this? What is the nature of this? What are the characteristics of this? The habit of spontaneously forming such questions and then tightening the knot indicates a confused mind as much as any delusion. It's not that you should never ask them — never go full Buddhist. They are useful questions to ask when approaching a situation you wish to understand, but — just as when cleaning with bleach — you need to be aware enough to protect yourself from their toxic effects.

Evolution obviously gives rise to purposeful structures. Look at the anatomy of the eyes: if you don't know how the crystalline lens works, look it up now. This order is obviously not fictional. And yet it's not real either. Evolution is fictional: there is only the moment-to-moment causation through which some organisms live, some reproduce, some die, some change. (PAY attention to what I am saying here. If you throw 2014 atheist talking points as though I were calling evolution fictional in the same way a fundamentalist does, I'm gonna make you fictional).

Did you seriously think reality would make things easy for you by being Aristotelian with respect to your made-up adjectives? It won't do gymnastics to contort itself to your preconceptions. What makes industrial machines scary is that we are to them as ants are to us: you can push against a pneumatic press as hard as you want, but it'll keep moving as though you weren't even there, crushing you without effort, passion, or knowledge. In this sense, reality is the greatest machine of all.

The Void, The Mute, @

phenomenology

Determined vs Hypostatic Properties

metatheoryconcepts

Q: Where is the saltiness of salt located? In the salt, or in us?

In answering this question, we ought to first identify the key things and relations of things within it, both those posited and those to be determined (without which it wouldn't be a question), stabilizing these relationships within our minds so that we have a clear system of relata; to the extent that we don't do this, we're clashing against the question without order, making our answers that much less justifiable.

To start off with, we're essentially taking an object, salt, and another object, "us". but who is the "us"? humanity in general? surely there are some people for whom salt doesn't taste the way it does "for the rest of us". at least, given our uncertainty about the matter, let's postulate such a person, Walt. Instead of being bitter, salt tastes sweet to Walt, due to let's say a mutation affecting the binding affinities of his tastebuds; salt activates the sweet-sensing tastebuds, not the bitter ones.

What I'm doing here is simply showing that the anomaly has its own determination, its own specific properties and cause. What the cause is I could care less — until some aspect of it comes into play in our analysis, I don't really care what it is, and postulated the tastebud cause just to demonstrate that there's a cause).

For Walt [? presumably part of a hypothetical, but didn't record previous part], salt would still have "saltiness" in the sense that salt inherently has the property of tasting like salt — which reveals that the question as originally posed is underdetermined, could be one of two specific questions. First, it could mean by saltiness the tendency to induce that specific flavor we taste; second, it could mean by saltiness the tendency to taste like salt. Call the first notion determined saltiness, or d-saltiness, and the second notion hypostatic saltiness, or h-saltiness. If next Monday all cardboard suddenly tasted like salt does today — say a demon watching over humanity decided to, as a prank, toy with the neural signals sent upon tasting something so as to do this — it would suddenly have d-saltiness and h-saltiness. If on Tuesday all salt began tasting like isopropyl alcohol, it would lose its d-saltiness while keeping its h-saltiness, cardboard would lose its h-saltiness while keeping its d-saltiness, and isopropyl alcohol, which was never d-salty, would suddenly gain h-saltiness. But saltiness could've just as well been named qwertyness while salt remains named salt, so let's take this to be a coincidence. you might argue that the presence of a bitterness-detecting kind of tastebud allows us to render objective the notion of 'saltiness', [?]

In our world, everything that is d-salty is h-salty, because salt is d-salty.

The Core of Phenomenology

phenomenologyterminologyconceptsrationalitylarge

[Seemingly written for something adjacent to ECM?] Loosely, the core understanding I'm trying to build is this:

• (Orientation) To analyze the nature of thought, it is improper to analyze the (a) re-presentation, or post hoc recall, of (b) singular and discrete (c) thoughts occurring in sequence (d) as such sequences appear to the highest levels of conscious awareness, generally via their linguistic aspects; instead, one must analyze the (a) direct, immediate presentation of (b) the explicitly and fundamentally phenomenologically conditioned flow of (c) conceptual content of all perceptible types and cognitive functions, rather than just thoughts, (d) through every reachable dimension, including emotive, volitive, and sensory aspects as well as (virtual) affordances.
• (Thesis) The property of conceptuality, no matter how abstract, consists at a fundamental but generally subconscious level of the bearing of particular configurations of sensory symbols (instantiations of what I call virtual schemata, as opposed to conceptual schemata, which exist within the realm of concepts proper). These configurations exist within and evolve through a network of relations between actual exteroceptive and interoceptive sense, virtual sense such as comprises the schemata, and internal dispositions such as emotion, volition, and so on.
• Over the course of this writing I'll attempt to introduce a framework for seeing and interpreting the schemata and the network they evolve in through direct perception, for, ultimately, there is something interesting that I'm observing and that I want other people to perceive, if not to directly sense (for I only gained the ability to do the latter after significant training, as mentioned below). However, because of some critical complications — that the thing being seen is too subtle and prima facie alien to most conscious experience to be explored without the creation of an interpretation to serve as a sort of "foothold" for perception; that the thing thereby determines the ways in which it can be perceived; that we must figure out how to go beyond it in order to establish an understanding of it from the abstract, social points of view of science and philosophy — the story can't end with mere observational directions. Instead, we must go through the "hero's journey" whereby this tantalizing glimpse into our conceptual faculties crystallizes into a global self-reflective orientation regarding the forms and functions of conceptuality. This journey has to be carried out with extreme caution: nobody who's ever attempted it has succeeded, and those aren't exactly good odds. Nevertheless, I'll give it a shot. Since "understanding" obviously isn't something we should anthropomorphize or reify as a distinct thing, I'm going to situate it in phenomenology, which, while more difficult to talk about, is more auspicious a starting point (I'll emphasize: starting point). I'll take as a starting point my own phenomenology (not like I can have yours), and will therefore start off by talking about the means of investigating phenomenology.

One annoying aspect of this journey: it is not traversed through time in the same way that an actual journey is. Rather, a hazy outline comes together first, before being filled in slowly with more and more detail which it warps to accommodate as it sharpens, and as a result there is no single aspect of it which is clear until the entire thing becomes clear.

[...] I can't really communicate the thing I wanted to get across in a single picture. It's one of those things where — and I really need a name for this, because it happens all the time — I take the abstract form of one representation to be "representative" of an entire class, but either (a) it turns out that the semantic content of the abstract form is very far from being the kind of set of material beings and relations that can be unified in a single image, such that no image I think to create seems to me to capture a "True" representation of what I want to represent, or (b) it turns out that even when I do have a single such recreatable picture in mind, the semantic content I naively attribute to it actually derives almost entirely from how I slot that picture into my various mental schemata, the picture just happily managing to be such that the various mental components of the thing I want to get across can all hang out together in that one picture, using it as an instrument for my mind to more effectively work with the thing as being a thing, rather than a representation of the thing.

Improvements in Physics as Suggestive of Reality

physicsKant

It's a lot harder to argue for the nonexistence of a reality independent of mind when confronted with the bare misses of modern physics: Newton's simple and intuitive theory of gravitation was just barely wrong (at human scales), and we had to adopt the extraordinarily strange principle of general covariance as a successor. What was it that we just got barely wrong?

In general, it's clear that what we experience isn't directly reality; while Kant's Transcendental Aesthetic is an amazing way to articulate how we must necessarily assert our independently homebrewed forms of intuition on reality in order to understand it in any way at all, I think he vastly overshot the degree to which our physical understanding of reality is therefore within ourselves. Like, the modern response to his proclamation that physics is due to how we see the world might be to point at the radical nonintuitiveness of relativity and quantum despite their being accepted parts of physics, and the fact that modern microprocessors, which wouldn't exist without relativistic quantum theory, do actually work—and I think this is basically a correct way to argue that we do really know something about the world outside of ourselves. Perhaps it's easy to say that physics is within us when 'physics' is just a few systematisms concerning astronomy, mechanics, and optics, as it was in Kant's time, but, now that it's grown to tell us why almost all aspects of (physical) reality are the way they are, I personally find physics much too large to contain within myself.

What makes reason apply to the world?

conceptsrationalityKantlarge

If we wish to use our faculties of reason to understand the world — and by this I mean not just the phenomenal world given to us or the physical world we suppose underlies it, but anything that may be said to exist in any sense, whether it be the nature of morality or something interesting like the structure of the prime numbers — it would be useful to understand what capability we have for understanding such things at all, the manners through which reason is able to “hook on”, as it were, to the world. That it can is clear by the existence of the computer I'm typing on, the human construction of which evinces the human capacity to understand solid state physics (among many other things), but how does it do that and why can it in the first place? Don't answer by discussing the scientific method — don't answer at all, because you don't know the answer.

A guiding question: just what exactly is it that makes my everyday processes of reason fruitfully apply to the world? There are multiple possible interpretations of this question, as there are not only multiple moving parts but multiple modes of interaction between these parts.

Essentially a self-reflective reformation of the entire faculty of thought -- a consequence of which is that I can't truly detail what exactly I'm chasing without first having captured it. It might be taken as starting off with the question "just what exactly is it that makes my everyday processes of reason fruitfully apply to the real world?", and immediately splitting into several families of problems depending on what exactly is meant by this very question:

• If I'm asking about what allows reason to deduce valid facts about the world given certain information, I now have an investigation into the nature of logic, both subjective, regarding the natural systems of statement manipulation within the mind, and objective, regarding the structures of valid implication that subjective logic may be idealized as “attempting to approximate” — or may be considered as relating to, in any case.
• If I'm asking about what must necessarily be true about the world such that it has valid facts discoverable by reason, I have an investigation of what logical structures may inherently exist in the world, and the extent to which they may be rediscovered or at least recast by reasoners such as ourselves.
• If I'm asking about how reason gets its information at all, I have an investigation into the structure of mental content — neither just the form nor the content but the manner in which the two interact, and neither just inbuilt nor given information but the manner in which the two interact.
• And if I question the validity of the question itself, I have an investigation into the epistemic limits of reason; should I attack my ability to answer it, I have an investigation into the epistemic limits of phenomenal experience.

Kant seems to have been one of the few people to actually take this problem seriously—the Critique of Pure Reason opens, and closes, by pointing out that Reason absolutely has to come to terms with how it is able to know anything at all if we are to understand how we are to come to conclusions about what reality exists beyond ourselves His examples were primarily metaphysical--the existence of a god, the immortality of the soul, etc., but his discussion does not in any way limit itself to these examples, as demonstrated by his distinction between noumena and phenomena, the explanation of physics given in the Transcendental Deduction, and the entirety of the Transcendental Aesthetic. (The rest of this note went to 'Improvements in Physics as Suggestive of Reality').. But other 'philosophers'—they never actually "love !c{knowledge}{or wisdom; sophia is a word from a different era}", only the image of loving knowledge—generically do not care to ask such questions; if they do, it's usually insofar as they're prodded into responding to Kant's legacy, and not out of such autonomous compulsion as is required to motivate the deep, critical investigation required; when it is, the questioners have generally not learnt enough science or mathematics to get a feel for the rhythm of truth; when they have, they are generally incapable of such introspection as is required to get a feel for the rhythm of cognition.

Stag Hunt

decisionmathematicslarge

the stag schema: among choices R and D, group GD conceptualizes R < Z < D and group GR conceptualizes Z < R, perhaps not thinking about D; so GD sees GR as choosing a bad option instead of a good one, while GR sees GD as trying to block a perfectly good option. (Z is not someone's point of zero utility; the phenomena we can describe with it are high-level enough that nothing like utility can be said to exist in our minds. There are merely patterns of mental factors)

whence the possibility of conceptualizing "R < 0 < D" or just "0 < R" for a game with clear utility payoffs?

1. zero points are primarily emotive backdrops, not utility judgements,
2. even if people agree on the object-level content of the outcomes, which they generally don't, they value things differently
3. humans don't have consistent linear preferences anyway
• it's perfectly normal, if problematic, for a stag-hunter to see a non-hunter and think "that lazy prick", and then to see a rabbit-hunter and think "fuck you jackass i hope a mountain lion fucking rips your throat out and eats your stupid little rabbits"

calling some issue a stag-hunt really points more at its general coordination structure: this is independent of the number of players, unlike the payoff matrix, which has 2^N entries, almost all intractable. it's a lot easier to characterize the payoff matrix by looking at the generic person's utility varies by decision when everyone else's decision is fixed and this fixation is itself characterized by a set of regimes (e.g., most cooperating, most defecting, whatever regimes the object-level details suggest might be good to look at). of course, this is just a way to characterize the general coordination structure through a series of tractable mathematical approximations…

there's a sort of statmech way of considering the large-N "local correlations" continuous-time stag hunt

• $H = \text{ human plane }, I = [0, 1]$

we can think of people as organized in a plane s.t. closeness represents possibility for communication/coordination, and then — treating the plane as continuous — conceptualize a field $P(p, t)$ = propensity for person $p$ to go stag at time $t$, which is continuous-like since close people have close opinions. ($P(-,-): H \times R \to I \cong P \cong P_{-}(-) = \lambda t.(P_t = \lambda p.P(p,t))$). And a weird but vital functional-kernel thingy: $K[P_t](p, q)$ = vector telling us how much each of q's actions affects p's utility when the board is in state $P_t$ (curried time). $K: (H \to I) \to (I \times I \to \mathbb R) \cong K: I^H \times I \times I \to \mathbb R \cong K: I^{H+2}\to\mathbb R$. especially important is $K[P_t](p, p)$, or p's own utility function (vector). no reason $K[P_t]$ should be symmetric, but it seems safe to treat it as such, and as a decreasing function of distance from p, like $1/r$ or $1/r^2$

$P(—, t)$ depends on $P(—, t-dt)$, since cooperating is better in the localized stag hunt when other people are already cooperating — we can use $K$ to set up a partial differential equation for $P$ and characterize the dynamics of the localized stag hunt. mental image: domains of (ferromagnetic) aiignment spontaneously forming in ising model: temperature is dependent on metric structure of plane, and external field is a modification to the relative utility of choosing stag over rabbit across the plane. super powerful metaphor, since magnetism concepts carry over to societal-game-theoretic notions... but not quite right, since the two halves in the ising model act identically — not so with stag and rabbit (latter is the disordered/paramagnetic "background", former is the ferromagnetic "foreground"). also, ising behavior changes greatly between dimensions 1, 2, and 3, whereas there's no good reason to make $H$ 2d instead of 3d...

Interfacing with Causation

causality

"root" cause is relative to consideration of causal structure — some combination of salience and proximity which for these purposes can be clearly delineated

the generalizing schema: dilemma over whether salient thing B with effect E actually belongs to background thing A; whether E ought to be treated as an A-problem or an B-problem. and its dual, the cogeneralizing=specifying schema: when B can be narrowed down in a given case to specific thing C, should E be treated as a B-problem or a C-problem

responsibility interface: A and B are agents, with B causing E conditional on A's inclining B to do so. subinterface 1: A rendered E triggerable, and B incidentally caused it; subinterface 2: A indisposed B to trigger E, and B intentionally caused it. factorization interface: A and B are processes, with B the root cause of E and A a generalization of B. Is the default analysis of the effect of B on E given as a specification of the general effect of A on E, or is it custom-made?

coresponsibility interface: B and C are agents, with B causing E by manipulating the actions of C. subinterface 1: B rendered E triggerable in a situation where C would incidentally cause it; subinterface 2: B took action to railroad C into triggering E, and C did not intentionally not trigger it. cofactorization interface: B and C are processes, with B the root cause of E and C a sub- or emergent process through which B effects E. Is the default analysis of the effect of B on E given as a specification of the general effect of A on E, or is it custom-made?

New Shade of Blue

perceptionphenomenology

I feel strongly that we need a new commonly used color name for a certain segment of the color space commonly called blue. There should be a secession: the name 'blue' will be retained by a more intense-deep-radiant color to ultramarine-Egyptian blue-sapphire, while the breakaways will resemble a more subtle-quiet-practical one closer to... azure? people seem to define the colors in this range pretty wildly, but I've made a diagram below. I've intentionally chosen not to name this new color, not even in my head, for reasons I will explain later. Instead, I'll just refer to it as the new color (though 'new color' is not its name, as 'proposed color range', 'novel shade', and so on are conterminous with my use of 'new color', but are clearly not equivalent to the name 'new color'. If something called 'cat' can also be called 'member of the species cattus cattus', it's more likely that it's a cat rather than that it has been named 'cat' in particular, and that its being a cat is why 'cat' is being used to refer to it. You can call a cat Cat, but that's a choice you have to make).

The differing fonts attempt to evoke the global synesthesia I have regarding these colors:

• The rightmost blue, which has its ideal in ultramarine, is rich, juicy, like a perfect apple. It's a vivid, deep color. Like, look at this ultramarine pigment. It's just so... it's not that I want to eat it, or even touch it or use it to color anything, but... it's just so intense!
• Cyan is kind of awkward, like a baby flamingo (I am not going to check if baby flamingos walk the way I naively imagine they do), but it's also bright and sweet. It's not that important.

Now, why do I feel this way? Why do I seem to think that there ought to be a new color in between cyan and ultramarine? [on subtle differences between peoples' chromatic sensory arrays, the linkings to their visual cortices, the ways they learn to create and interpret the experience of color and of this or that particular color; on all the subtle differences that grow within this series of processes all procedurally generated in any given human and only imperfectly standardized by culture]

--- Discovery (several months after writing the above): Russians actually have this distinction—it's native to their language! There's голубой (goluboy), which is more or less equivalent to the middle blue/new color I proposed above, and синий (siniy), which is the right blue. The Wiki article on blue-green color distinctions describes the former as "sky blue, light azure, but does not equal cyan" (right!!), and the latter as "'true' deep blue, like synthetic ultramarine" (!!!).

The Structure of What You Can Remember

perceptionconsciousness

When I refer to the explosion of your modeling faculty, perhaps you imagine a hallucination of the space around you being rapidly tessellated by hot dogs, or (if you have nonzero intelligence) a constant chaotic irruption of color and noise, but this is not what I mean. You cannot imagine such an explosion, and, if one occurred to you, you would not be able to remember what it was like, for the same reason I can't remember a sequence of Arabic characters (even with Chinese, I have determinate mental formations that can occasionally allow me to reproduce the characters; to say them would be something like "crossed seven, double christmas trees, stake window" (these are not the formations themselves))--but Arabic 'just' looks like squiggles. Illustration: imagine that you're playing a video game written in a different language, and some character comes up and gives some instruction you invariably can't read, but depending on what it is you have to do something different. If the instructions are in Polish, you have lots of things to hold on to -- specific letter sequences, vocalizations, and so on, which vary exactly with the meaning. And there's a whole world of such things, even if you're not conscious of it. If in Japanese, you might hold on to various aspects of the density-openness and curviness-angularity (exemplified by its splits between kanji/kana, hira-/kata-).

Having and Recovering From Blindness

perceptionphenomenology

Though I've never been blind, I can tell you what (congenital) blindness is not like—it is not blackness. What do you see out of your right eye when you close it and it alone? It is not blackness. "No light" is still an input, and is the input given by closing both eyes, but when just the right eye closes, the mind expunges the entire input channel, and you see through it about as well as you see through your hair.

what must recovery from blindness be like? first, vision is a culturally-guided locating: someone in our culture would have an experience which they characterize in ways which remind them of that vision thing people talk about, recognize their experience as that vision thing or something similar, and interpret it thereby; but if everyone else had always been blind ("why do we have eyes? animals lose their straight-line-telepathy without them, but i guess they're vestigial in humans?"), what would occur to them in lieu of the experience-structuring-concept of vision? an awareness of an overwhelming something, which when probed folds in on itself to reveal a gradience of intensities congruent with spatial extension, a manifold of spatially varying qualities. To learn to recognize particular people would be an extraordinarily difficult task—though, like with a video game, the work mostly happens below the level of awareness. If you play, say, Mortal Kombat all day every day, your brain will develop its own little dialect to correlate perception with action: you don't think "he's moving like this, so he's probably gonna do that", but you experience the thinking of that (insofar as your brain infers the correct action, it does so in a probabilistic framework, even if you screw it up every time you try explicitly thinking), and get the felt expectation-of-possibility/desideratum-for-next-actions.

Thinking of Chickens

intelligenceconceptsphenomenologylarge

If you simultaneously shock a rat and ring a bell multiple times over several weeks, the rat will learn to correlate the sensation of shock with the sensation of bell-ringing; in this manner, minds correlate sensations. The rat can be said to have acquired a bell/shock "sensory formation"; further, the rat will have acquired two notions of particular kinds of senses: the sound-sense of bell and the touch-sense of shock; it's not relevant here that the two are correlated, just that the rat learns to take them each as distinct "kinds". When the rat has such a kind — say, bell-sound — in mind, it can not only recognize new sounds as being bell-sound, but as being like bell-sound, where the exact criteria for likeness obviously depend on those criteria which are involved in the formation of the kind and which distinguish it from other kinds...

A single cognition is really just a cluster of such sensory formations, and the act of cognition is the evolution of sensory formations through a hyper-excited and recurrent correlational network of such formations. For instance, when I ask myself "why did the chicken cross the road", I immediately and non-voluntarily break that sentence down into the clusters of sensory formations evoked by each word, or by each idiomatic combination of words / template, and then my mind just starts correlating and evolving this whole mass of formations according to the network it's learned. Obviously, I (virtually) picture a chicken and a road, and the "virtual motion" of crossing — sort of like a series of tangent vectors perpendicular to the road — by which the chicken moves. But then the "why" part comes in, and plays upon this sensory formation in its own peculiar way: it primes me to consider the crossing as volitional, a specific aspect of the chicken-universe complex, and primes me to consider the chicken as intelligent—or, rather, it primes me to attribute my own sense of cognitive sense to the chicken. This particular complex of sensory formations (virtual-chicken-cross-road/self-chicken-cognitive-projection/chicken-universe-complex-driven-volition) is then evolved by my urge to answer the question; answer is explanation is backwards cognitive-sensing, and primes me to use my internal network's criteria for such backwards cognition to find something which I can see as itself evolving into the volitional crossing of the road by the chicken.

Apparently, in my mind, this answering force translates to looking for an environmental cause; because of this, i'm driven to find some sensory formation linking to the universe as it stands in the chicken-road-universe-complex, and my network evolves this reverse-causal formation in this specific chickenic context into a list of cognitions which look like "motives". my mind wants to split motive up into carrots and sticks, and thereby flicks through a series of formations each of which could be construed either as desire for the me-embedded-in-the-chicken to get to the other side or as an aversion for the me-embedded-in-the-chicken against staying on the current side. Because of the chickenic context, my mind when it looks through such motives evolves chicken through the network to see what things are related—so i might end up thinking "slaughterhouse", and finalizing that into "the chicken crossed the road in order to get away from the slaughterhouse"; this makes it clear that it's really the me-embedded-in-the-chicken which is doing all the thinking, as it's not like an actual chicken would even know what a slaughterhouse is. Or i might say "the chicken crossed the road in order to eat the seed on the other side"; again, a perfectly normal answer coming from within my network, but maybe not accurate for a real chicken's sense of life. I've glossed over the critical role of language in the development and facilitation of cognition in this analysis, though.

Chakra Spin Tensor fields

phenomenologyperceptionmathematics

(From when I was playing around with mind-body oddities in the first half of 2022—meditation, kundalini, chakras, and so on.)

Each chakra seems to have its own spin to it — or, rather, spin operator, since there isn't one canonical feeling of a spin but rather a scheme by which different directions and orientations of spin seem more correct than others, and through which a set of eigenspins (or just maxima?) each capable of satisfactorily describing some chakra is given. I need some coordinates to describe them: common medical terminology calls the left-right axis the (F)rontal axis, the back-front axis the (S)agittal, and top-bottom the (L)ongitudinal. For orientation, say left/back/bottom is negative and right/front/top is positive. Think of F, S, and L as constant vector fields pointing from negative to positive along their particular axis. There's also an (E)xterior vector field, which points outwards from the skin at all points, and its additive inverse, the (I)nterior vector field.

Take the median plane Make a T-pose. F goes from left hand to right hand, S goes from back to chest, and L goes from feet to head. The median plane splits you into two symmetric halves. (orthogonal to the F axis) and intersect it with the surface of the body to get a closed loop around the body. With positive orientation (right-hand rule: counterclockwise around the F axis), this path travels up the ventral midline, backwards across the crown of the head, down the dorsal midline (spinal cord), and forwards across the perineum. I call this oriented path the primary medial loop. All the chakras are on this loop, so we can speak of the extent to which they spin around the tangent vector the loop provides each of them. All the chakras are also on the skin, so the E field also has a tangent vector at each location—and this vector is never collinear with the PML tangent vector, so each point's spins around each are linearly independent. So we can take the the E and PML tangent vectors to provide a basis in which to represent spin—the right things are more or less linear, I think, so that the felt spins of the seven canonical chakras can really be thought of as samples of a chakra spin tensor field across the entire PML.

I can't be more precise than giving signs, but for me the idiosyncratic spin values look like (chakra, +PML spin, +E spin): (7, -, -), (6, +, -?), (5, +, -), (4, +, -??), (3, +??, -??), (2, ???, ???), (1, -, -)There are many different systems for enumerating the chakras, but since they're not objectively real (just psychosomatic inclinations), I'm fine using the sevenfold system: The first chakra (1C) is at the very bottom of the spinal cord; 2C is just below the navel; 3C is at the solar plexus; 4C is next to the heart, 5C is at the throat; 6C is between the eyebrows; 7C is at the crown of the head.. Sometimes the question marks represent bistability—I can stably picture the chakra's spin as being either positive or negative relative to the given axis (tangent vector), and sometimes they represent obscurity, where I can't really make it out either way.

Manipulating my body's energy flow also intensifies the idiosyncratic spins of those chakras I'm giving attention to, and in this way I can "strengthen" them. My impression is that strengthening 6C increases the flicker rate/temporal precision of attention, while strengthening 7C lowers the rate at which... breadth takes away from depth. And this isn't just breadth spatially, but, as usual with attention, across different qualities of different things.

Third Eye Effect

perceptionphenomenology

An interesting thing has been happening over the past week: sometimes I'll get a specific feeling between my eyes, a sort of light but noticeable ballooning pressure, like the coincident flesh is ever so subtly expanding. It seems to be triggered mostly, but not entirely, by the feeling of my hair resting on that area. Or, rather, that physical feeling is a possible cause which isn't necessary to any perceivable degree but which is often sufficient. Anyway, whenever this feeling arises, I find that I have increased access to a state of mental quietude. It's the same feeling of being in the swing of meditation, or more accurately in the highest jhanas — but it's a sort of shortcut I can take; the state makes it easier to simply tune out perception and cognition, though it tends to come with the fuzziness-feeling and great spatial extension of the fifth jhana... Actually, maybe it's best describable as an arupa modifier?

I occasionally do a sort of eyes-open meditation where I focus on individual aspects of phenomenology in turn in a manner that simulates very light forms of the four rupa jhanasIn the eight-jhana classification, 1J-4J are 'rupa', material, and 5J-8J are 'arupa' (non-material)., but when this feeling arises, the eyes-open meditation seems to want to go through the four arupa jhanas instead.

Start in the fifthThe first arupa jhana (5J), "infinite space", requires that you unbound your spatial sense, to be aware of what feels like an infinite space; 6J, "infinite consciousness" requires you to unbound your proprioceptive sense to feel coextensive with this space; 7J, "nothingness", has you let it fade away and your sense of self with it, to perceive nothing; 8J, "neither perception nor non-perception", has you let the sense that perceives nothing fade away, and is indescribable. by extending the fuzziness out to the entirety of the boundless homogeneous space underlying my actual visual field, go to the sixth by turning my focus to my proprioceptive sense of self and expanding it along with the fuzz (useful to visualize a pin pricking empty space as pricking me at the same time), go to the seventh by just rejecting the sixth, eighth by rejecting the seventh.

Also, my ability to hear music in my head seems to have become significantly better, though that's probably a more temporary thing.

Note: the third eye pressure eventually led to (what I'm pretty sure is) a temporal lobe seizure, which led to me calling off my meditation practices—I'm not epileptic, and have not had a seizure since then. This seems like a rare phenomenon, but it's evidently possible.

Why Hypnic Jerks?

perception

I closed my eyes and tried a bit of meditation just now, but it didn't work: as usual for this level of sleep deprivation, I just enter the hypnagogic mode of rapid manifestation and relation of absorbing situations. When I opened my eyes to break free from this, though, I felt the weird sensation of awareness re-filling my body from my head down to my hands over the course of maybe half a second to a second. I wonder if that's a normal thing.

(later) Ok, it does happen consistently, and it doesn't seem to spread from my head but rather appear in my upper arms from my head a few fractions of a second after opening my eyes, spreading down my arms from there. Sometimes if I'm especially deep there's a small involuntary physical jerk upon breaking free of it; also, it seems to be able to spread to my heart as well, though I generally have to focus on that for it to happen.

Seems like the strength of the physical jerk is dependent on strength of reflection and, it seems, content of thoughts (...I feel like i'm not supposed to be doing this). Is this the reason for hypnic jerks? there's a sudden physical reaction on the transition from hypnagogia-mode thinking to normal thinking whose strength is dependent on the depth of absorption in the latter, such that when you're really absorbed the jerk can really spook you?

Visual Resets

perceptionfacts

While writing my huge thing about phenomenology, I had some questions about the feeling of dizziness. So I got up and spun around in circles, did a small test in the aftermath, and just saw the answer for myself—braver than every cognitive scientist put together. The question: how much of the visual perception of dizziness after spinning is given by the falsely persisting sensation of a rotational velocity in the direction of the spin? And the answer: pretty much all of it. The proprioceptive sense combines with the vestibo-ocular reflex to produce a really cool effect where, if you fix your eyes on one point while extremely dizzy, the world seems to keep moving in the direction of the spin but "resets" maybe... two times a second? so as to keep your eyes on that point. It's literally like the entire scene keeps jumping in the direction opposite the spin.

The Multivocity of Truth

conceptsterminologymathematicslarge

The idea of truth itself may seem simple -- necessarily among the simplest of ideas, we might argue, due to its foundational nature -- but it's not. The reason why: there is no singular mental image of truth, but a multiplicity of fuzzy images a great many of which simultaneously coexist in each of us, all of which we call "truth". Our neglectful use of language leads us to systematically elide the fractal interspersed differences between the images of truth varying not only across but within each of us; this is one of the core reasons behind the persistent and hitherto irresolvable confusion underlying all language use. I'll refer to this property of the idea of truth, which is a basic consequence of its conceptual fictionality, as multivocity. This is in contrast with the more well-known univocity, and we can pin down these concepts by saying that a property P is univocal when all P things are P in the same way, and is multivocal when P things can be P in different ways. As an example, electrical conductivity is univocal, as all things which have that property have it by virtue of conducting electrons to some extentHad I just said "conductivity", this wouldn't hold. It's the word 'electrical' that makes it clear that this is a technical term with a precise meaning., whereas weakness is multivocal, as weak things can be weak because they are fragile or timid or powerless and so on.
In gauging multivocity, I think it's best to ignore merely quantitative distinctions — to me, the fact that some things are more electrically conductive than others does not make electrical conductivity multivocal — and focus on qualitative distinctions. Nevertheless, I think that it's best not to simply classify properties as either univocal or multivocal, but to gauge the quantity and form of their multivocity: it's not a simple binary proposition except in extreme cases, being in the generic case a mode of analysis. "Wild" is multivocalYour hair is !c{wild}{unruly/disorganized}, that's a !c{wild}{non-domesticated, connotes unruliness/hostility} dog, he's been acting !c{wild}{intense/frenzied/hostile}, he attacked !c{wildly}{intensely/unrelentingly/disorganizedly} but "feral" is, while still multivocalThat's a !c{feral}{non-domesticated, connotes danger} dog, he attacked !c{ferally}{aggressively/desperately/disorganizedly}., less so than wild.
That multivocal properties differ in how multivocal they are does not in itself make multivocity multivocal, since we're ignoring merely quantitative distinctions, but what does seem to make multivocity multivocal is the varying geometries of attributions different properties can take on: one property might be multivocal with one wide yet clearly singular range of forms of attribution, whereas another might be multivocal with several discrete manners of attribution. To me, "loose" seems to fall in the former category and "true" in the latter — loose means freedom of movement, but can have a wide range of different valences, implications, and substrates; I'll show shortly how "true" is in the latter.
Note: My use of the word "property" here is a white lie, since the concept of a property breaks down under sufficient analysis, but since it would take another ten thousand words to explain that, whereas the white lie seems sufficient for now, I'll just stick with it, providing as a makeshift alternative the following characterization: A property is "a continuous range of mental images united within a certain word". The level of distinction at which I treat two clusters of images sharing a common word as different properties (i.e., at which I take continuity to have broken) seems largely arbitrary — bright-intelligent and bright-luminous feel like different properties to me, but bright-intelligent and bright-auspicious feel like two extreme ends of what is the same property, since "he's bright" and "he's got a bright future" feel like they're continuously connected to me.

Multivocity and Perception

It's hard to say that a multivocal property can even be called a single property, or a single idea (again, won't go into this here). Nevertheless, by using a single word for the various ways in which we apply it, we tend to elide the differences between these ways, forcing our own faux univocity onto the word by means of which we think of it as being single rather than multiple. Sometimes this is intentional: one common, but obviously ridiculous, solution to the multivocity of, say, truth, is to merely pin down one formal account like the correspondence theory and treat uses of our intuitive idea(s) of truth not legible to that account as incoherent; not only are we simply discarding lots of information which almost certainly has some important relations to those aspects of our intuitive idea(s) of truth that are legible to the account, but we're simply fooling ourselves into thinking we've tamed truth. In reality, we'll continue to think and speak with those versions of truth not legible to the formal account, because any given formal account can only span a very small part of what humans mean by the word true, what genuine uses humans get out of these non-formal meanings; in our premature declaration of victory we'll already be sowing the seeds of the system's own demise by fundamentally and perpetually failing to act according to its prescriptions even as we build it.
One way of partially fixing the problems that unexamined multivocity causes is through examination, namely the process of categorification and separation: of discovering the individual components of the geometry of the multivocity of some property and the dimensions of continuous variation of each of these components. In other words: when we attribute some property to a thing, there is a particular image (in simpler cases we might call it a 'sense', and in more complex ones an 'understanding') of the property as it seems to arise in our perception of the thing. Fixing the property and letting the thing and image vary, what are the fundamentally different ways in which the image can vary, and, for each particular way, what are the dimensions along which the image can vary while still keeping within that way?
This is an investigation of the actual use of words at the actual time of their uses. It is insufficient to theorize about words without constructing actual, normal examples of their use and analyzing those and the mind states in which they are constructed, and it is insufficient to think of reasons for why a particular property was attributed to a particular thing rather than directly looking at the mental process as the property is actively being attributed to the thing.
Hence the necessity of personal phenomenological investigation. In the case of truth, we must peer into these images by examining the specific ways in which we take things to be true, the specific kinds of mental configurations by means of which things are marked as true. (I say "things" rather than limiting myself to "propositions" because our sense of truth does not limit itself to "propositions" — I might think that a portrait truly captures someone's likeness, that a fictional book says something true about the world even if I can't quite pin down what that is, and so on).

As a demonstration, I'll pick out some pretty normal beliefs I have that I take to be unambiguously true, and examine what's going on that makes them seem true to me. Again, this is an analysis of what's going on inside my mind: if I could have equally deep access to the minds of others, I'd examine theirs and try to formulate higher-order patterns, but as it stands I can only talk about my own idiosyncratic patterns of thought. They will not be the same as yours, even if they support beliefs that you share as well.

• (Truth Via Verification)
• Belief (a): Callisto is a moon of Jupiter. True.
• Really, this is basically just a symbolic proposition with little real import: I know nothing about Callisto except that it's a moon of Jupiter, so I can only really say that I take this to be true insofar as I'm aware and confident that this meaningless symbol is attached via the being-a-moon-relation to the very meaningful planet Jupiter.
  This awareness shows up in the intuitive expectation that I'll find "blah blah Callisto is a moon of Jupiter blah blah" should I go to look it up. In other words, I not only take belief (a) to be true, I take it to be the kind of belief that evinces itself through a social symbolic agreement rather than through a configuration of the external world: my taking of this belief to be true does not directly consist in my taking there to exist a whole moon with real but merely unknown properties, as if this were the case then I would directly attach belief (a) to the feeling of a findable moon around Jupiter, rather than passively infer the findability of such a moon as a consequence, and only once prompted to think about it.
  This division between the direct, immediately inferred consequences and indirect, incidentally inferred consequences of a belief may seem to be rather strange and secondary to the point, but it is in fact the key point here, as it tells us why exactly I take (a) to be true: I take (a) to be true because it is a rote syntactic relation I have learned. To see this process flipped, I might consider the statement:
• Belief (a'): Balsa wood has a density below 1 gram per milliliter. True.
• In this case, I immediately re-interpret the predicate as "less dense than water", and construct a vision of wood lying above water due to its density, rather than simply coupling the words "balsa wood" to the idea of floating on water. Perhaps this is because I have actually floated balsa wood on water myself; certainly such an experience wouldn't be enough to confirm that balsa wood is actually less dense than water, but I've looked it up to see that it really is, and therefore take (a') to be true because it has an intuited physical relation concordant with experience and research. (If a single search counts as research). To see an example yet between these two extremes, take:
• Belief (a"): Helium is lighter than air. True.
• I believe this to be true, and associate it with the experience of rising balloons, but that it's helium in particular in those balloons isn't associated with an intuition: I've never consciously seen or felt helium, and I haven't even inhaled it to talk in a $^\text{super high voice}$, so there's nothing physical I associate with it except for a vague sense of presence. (On the other hand, I feel air all the time — most obviously when it's windy or I'm out of breath — and do have a physical intuition of it as a sort of slightly dense-viscous permeating substance).
  Summarizing: One kind of way that I might attribute truth to some statement is by affirming that the relations that statement presents actually hold, which is done through mental contact with the images we have of the objects/properties being related, concurrent with analysis of those objects/properties.
• (Truth Via Concordance)
• Belief (b): I'm more attractive than a dead bird. True!
• This belief has a method of being true which points to a rather curious property of the object ("a dead bird"): there is no one dead bird I have in mind — in fact, I didn't even picture a dead bird when I first wrote the sentence — and the actual attractiveness levels of all dead birds existing at this very moment does not matter. Even if you managed to find a dead bird whose beauty makes the Venus de Milo look like a febrile child were given a papier-mache tutorial in Klingon, I'd still take it to be true that I'm more attractive than a dead bird, the focus being on the article "a" even though I never even pictured any particular dead bird.
  If, however, you splashed me with sulfuric acid as a prank and reduced my face to a hideous blob, my intuition would rapidly switch from taking (b) to be true to taking (b) to be false. So the truth of (b) clearly does depend on the state of the world insofar as it depends on my attractiveness. In fact, if dead birds suddenly became extremely attractive starting tomorrow, I'd stop believing in (b) starting tomorrow, so the truth of (b) depends on the attractiveness of dead birds, even though no dead bird was pictured and judged (un)attractive when I came up with the sentence and took it to be true.
  This, therefore, is a belief that I judge to be true through generic features: the abstract idea of a dead bird, to me, creates repulsiveness creates a total lack of attractiveness; my mind, after forming a context in which we're talking about attractiveness and we're talking about my attractiveness, reads "a dead bird" and forms this zero-attractiveness, and upon comparing it to my non-zero attractiveness finds that I am indeed more attractive than a dead bird. This model explains our observations: one beautiful dead bird won't change my abstract idea of a dead bird, but all dead birds becoming beautiful could, and so on. So, I take (b) to be true because it checks out when I analyze it with my own abstract forms.
  Summarizing: One kind of way that I might attribute truth to something is by affirming that the relations that thing presents are semantically concordant via contact and analysis of some mental images provoked by the sentence concurrent with semantic exploration contextualized by and started from the sentence.
• (Attributions of Mathematical Truth)
• Belief (c): $\int_{-\infty}^\infty e^{-x^2}\, dx = \sqrt{\pi}$. True.
• This equation displays the value of the Gaussian integral, which is something that probably most students of math and statistics eventually encounter, since it's important to defining the normal distribution and is generally evaluated through a mildly entertaining coordinate trick.
  Of course, here we're considering $\sqrt{\pi}$, $\pi^{1/2}$, and so on to be the "same thing" — as usual in mathematics, I implicitly take the categories which truths are truth of to be independent of any particular symbolic manifestation. "They're the same point on the real number line", I might say. There's a tension to this, in that not all symbolic representations of this point are the same to me: if the phrase "the area of a circle with radius 1" were written in place of $\sqrt{\pi}$, I'd take it to be a geometric quantity which can in some way be 'filled out' or 'swept through' by the integral, whereas I'd take $\pi^{1/2}$ to be an algebraic quantity which can in some way be 'expanded to' the integral. Nevertheless, this is just a tension, not an invalidation, as the experience of doing mathematics strongly suggests that mathematical objects can be treated as observer-independent, and therefore independent of the psychology of any particular doer of mathematics. This objectivity is broken when I, as a person, take this equation to be true. I've identified at least five different ways in which I may do so. Three of these ways (identical/coreferential/occupational) rely on an non-interactive image of sameness of, while two (procedural and dynamical) rely on an interactive image of arrival at, and one (ideological) relies on a modal image of reason for necessity of. I'll describe them in this order:
1. The identical way, in which the relation between the two objects my mind turns the equal sign into is something like "is the same thing as" — $\sqrt{\pi}$ is interpreted as a point on the real line, and the integral happens to evaluate to that same point.
2. The coreferential way, in which the relation that the equal sign is turned into is something like "picks out the same thing as" — $\sqrt{\pi}$ is interpreted as the label of a point on the real line, which point the integral also labels by virtue of evaluating to.
3. The occupational way, in which the relation that the equal sign is turned into is something like "happens to occupy" — $\sqrt{\pi}$ is interpreted as a position that a number can have, and the integral happens to occupy that position. To illustrate the difference between these three modes better, I'll give an analogy: when I say that "Joe Biden is the president of the United States" (it's 2023), I might interpret this as saying that:
• (Identical) the person Joe Biden is the person who is the president of the United States, just as the number $\int_{-\infty}^\infty e^{-x^2}dx$ is the number $\sqrt{\pi}$;
• (Coreferential) "Joe Biden" and "president of the United States" both refer to the same flesh-and-blood person, just as the label $\int_{-\infty}^\infty e^{-x^2}dx$ picks out the same number as the label $\sqrt{\pi}$;
• (Satisficing) there is an office of "president of the United States" which Joe Biden happens to occupy, just as $\int_{-\infty}^\infty e^{-x^2}dx$ happens to occupy the position of (the office of) $\sqrt{\pi}$.
4. The procedural way, in which the relation that the equal sign is turned into is something like "evaluates to" — $\sqrt{\pi}$ is a symbol, and a series of correct and unambiguous applications of rules, each of which preserves identity, transforms the expression $\int_{-\infty}^\infty e^{-x^2}dx$ into the "final answer"/"thing-that-it-is", $\sqrt{\pi}$. Were the right-hand side more complicated, I'd be less likely to think of the equation in this way, because it would be less clear to me that there would be such a procedure, or at least that there "ought" to be one.
5. The dynamical way, in which the equal sign is not immediately turned into a relation — rather, the sides of the equation are turned into some moving sensory image (almost always visual, but sometimes kinesthetic) which ends in agreement between the sides, this sensible agreement being the relation the equal sign represents. For this particular integral, I don't seem to take a dynamic approach since (a) my mind seems uncertain about how $e^{-x^2}$ looks (even though I know it's exactly the same shape as the millions of bell curves I've seen), and (b) my mind is not sure how to link the area swept through by the integral with the magnitude $\sqrt{\pi}$.
   A much better example is given by the equation $e^{i\pi} = -1$: in accordance with Euler's formula $e^{ix} = \cos x + i \sin x$, my mind intuits $e^{ix}$ to be the point $x$ radians counterclockwise from $1$ on the complex unit circle, and when thinking of $e^{i\pi}$ puts a point at $1$ and lets it zoom $\pi$ radians counterclockwise — precisely across the upper half of the circle — to arrive at $-1$. The particular manner in which I assign truth to $e^{i\pi} = -1$ derives from the arrival of this point at $-1$.
   There's a way in which this example is uncharacteristically strong, though: usually when mental images play out like interactive videos in my mathematical thinking, there isn't an actual complete motion of the objects in the image, merely a sense of motion. If you're playing catch with a dog, a few rounds of throwing the ball will cause the dog to stop needing to see the ball in motion to assume that it has moved and start tracing its destination and running to intercept it — at this point, you can merely move your hand as if you're throwing the ball, and the dog will run backwards to catch it, not knowing that you've hidden it behind your back. I, the watcher of the video, am the dog, and only need to sense that there is motion, rather than to sense motion, in order to trace and follow the destination of said motion — I pre-consciously know where it'll go; since that same pre-conscious knowledge imbues and constructs the image, I, unlike the dog (whose ability to trace the ball's destination has no causal influence on the destination itself), am never misled, and never need to acclimate to an untrustworthy partner.
6. The ideological way, in which some intuition for how the solution should look is activated by the solution presented to me. Even if some of the details aren't correct — maybe there was a small error somewhere and the solution is actually $1/\sqrt{\pi}$ — the given solution fits into a frame, whether for the particular problem (yes, I've seen this before, I know that $\pi$ is relevant) or for something much larger (yes, we should expect $\pi$ because the integral can be interpreted as a circle something something). This isn't the whole story — for instance, topic-specific, field-specific, and global problem-solving heuristics which are often alien to the frame may also be thrown in as well (yes, we should expect a small, closed-form solution; yes, this does appear somewhere in the normal probability distribution function), and these contribute in their own way to the manner in which we take the belief to be true.
   Once again, I'd like to emphasize that this is how my mind seems to fundamentally think about these things, not how I desire to think or believe I should think or have a metaphysical committment to thinking. I don't have a firm metaphysical stance on the map-territory distinction my mind seems to enjoy subverting, and my mind is very capable of swapping and combining these forms of truth at a whim (I'm certain it's a very complex multichotomy with many more dimensions than I can consciously notice). To make the situation even more complex, I find it very difficult to translate the raw language of mental movements into the English language; though I have a constantly-running internal monologue which largely utilizes English, and have a very large vocabulary, these mental movements clearly lie well below this level of consciousness. So, again, this is not a commitment (at least, not one that I'm intentionally making) but rather a description.

This exemplifies what might be called the idiosyncratic conceptivity of "truth" -- the fractally context-dependent space of concepts that I activate in tandem with the word 'truth'. Perhaps, after all this, I can now ask the following questionwithout being immediately dismissed as a postmodernist heretic (did I use enough math symbols? I heard you like math symbols! Let me throw in a few more to make you think I'm being objective! $\bigotimes \sum$ !c{$\coprod$}{woww is that an upside down pi?! so RATIONAL!} $\to \bigwedge$): What is truth?
There is... something about this physical reality as it appears to me which allows mathematical structure to appear in a sort of fuzzy, fractal way even at the level of macroscopic conceptual delineations. There is something about the conceptual manipulations I call "doing mathematics" that allows them to access a timeless framework of mutual in/compatibility. Everything that is truth seems to come out of these first somethings... but what are they, and how do they create truth?

Tuning Into A-Time

phenomenologyconsciousness

The mind, after falling back from the 8th jhana to ground level, looks around for the right inclination, and eventually "falls into" it, very much like someone walking through a forest stepping into a pool of jello thinly covered by leaves and dirt. What one falls into is a virtual nothingness that I can only articulate as "zzzxxxxccccccvvvvvbbnnnnnn" — a nothing that is something and yet clearly still a nothing as that something, being in this sense a vague intuition of nirvana.

There are persistent senses of "being in the now" and "non-doing" that I seem to be able to access. The first is, like, being tuned into the A-theory of time rather than the B-theory of time — naturally I (and, I assume, almost everyone) half-ignores the flow of time, their directed actions instead structuring time in their own way; right now I am typing, but the typing overwhelms the right now, becoming it so as to give the right now—which is originally just an infinitesimal—an extension over an intuited interval. That's (a phenomenology I'm analogizing to) B-tuning; A-tuning, on the other hand, is accessible by means of a peculiar mental motion I can't really describe, and consists of the retraction of this virtual interval to the infinitesimal at its core.

It's actually pretty demotivating, and I don't like it. The first time I got into it, some time yesterday, I happened to set myself a mental reminder to do something in two minutes—maybe the microwave, idk—and felt dismayed by that, because it felt like "in two minutes" was forever away. I realized that this is because my brain was modeling "in two minutes" as "in two minutes from now", and this now was constantly moving forward, pushing the two minutes along with it! Whereas normally the brain places a marker like that at a certain point in 2-time relative to the current chunk, periodically increasing its salience as that point approaches. The second is just a persistent 40% sense that whatever is done is not being done by me, whatever is being felt is not being felt by me; obviously this is just the insight of non-self given a permanent semi-residence, but it's still very strange to examine as I go about life. Theravadans describe a stage along the path where there's a constant sense of non-autonomy and the "THIS is IT [sic]", which I guess corresponds to these?

-ness-ception

phenomenologyperceptionconsciousnessterminology

Yesterday, I was trying to approach NSNirodha samapatti, a state beyond 8J which involves the "cessation of perception, feelings and consciousness". I never really got to inhabit it before I put a stop to my meditation practice. by shutting off experience, which I sometimes did by moving my awareness to some phenomenon, linking my consciousness to that phenomenon (by making it consciousness of that phenomenon), and then dissolving the phenomenon by viewing it in its impermanent non-existence, thereby seeking to rip the carpet from under consciousness. This didn't really work, as consciousness just became consciousness of that non-existence, rather than non-consciousness of the non-existence (...), but I did get an extraordinarily weird sensation. A gap forms in the middle of phenomenal space, segregating near from far but connecting the top and bottom, and the near buckles away, falling behind me as I'm transported to the far, which becomes the new near. Ingram says "The rarest no-self/suffering variant is hard to describe, and involves reality becoming like a doughnut whose whole outer edge rotates inwards such as to trade places with its inner edge (the edge that made the hole in the middle) that rotates to the outer edge position, and when they trade places reality vanishes. The spinning includes the whole background of space in all directions." — this seems to match.

Now, closing my eyes and concentrating feels a bit different — focus is on the shape of the inclinations of mind and the play of fundamental intensities in the visual field. How do I describe it... There's a sort of "-ness-ception" that the mind is capable of, and it can be subverted by consciousness to a small extent. An example:

if you close both eyes, you can still see something if you learn to grasp that sensory interface in itself, regardless of what it's displaying. This thing is dark and detail-less at first, yes, but it's obviously there nevertheless. You can examine the slight colors and lights and movements that show up (of course, when you focus on them, they can get significantly stronger, but even when they don't, that they're still there is the point). Why is it, then, that if you close one eye, the mind blocks it out? there must be visual input of the same sort that causes most of those closed-eye phenomena, so why aren't they seen? It's this -ness-ception, I think. If I focus on it, I can see what's in the visual field of the single closed eye, but it's very high in non-ness. It's not that "it's not really there", but that there isn't very much to be there or not there in the first place—the non-ness is in the it itself, not in its being present or not being present. This might be the same non-ness underlying the imagery of the state of neither-perception-nor-non-perception — a transparent window into what nothing is seeing?

Autotelekinesis

phenomenologyperceptionfacts

(written while intensely sleep-deprived, playing around with sleep deprivation)

I've figured out a very weak form of auto-telekinesis: I can will some part of my body to move in any direction and, if I'm willing it strongly enough, which is extremely difficult, it'll move slightly in that direction. My performance of the action seems to be correlated strongly with subtle non-conscious movements of my eyes and mouth in the indicated directions, but these aren't necessary: I can still do the trick, albeit more weakly, while suppressing them.

I keep nearly falling asleep as I write this. And I can't really see the screen well because I think I hurt my eyes straining them to develop this thing, which works in tandem with my sleep deprived state... but one really interesting thing about this autotelekinesis is that if I close my eyes, it becomes significantly more powerful but also loses what reality it has. Like, I put my left and right hands over the keyboard, indices on f and j as usual; after lifting them up a bit, I close my eyes and will them to move apart: they seem to start to move rapidly, but even after letting this happen for a while, I open my eyes and they're still just hovering over the f and j keys.

I think this is a really, really important clue: it suggests that what i'm calling autotelekinesis is just directly sending predictions to my interoceptive kinesthetic faculty, and that movement is primarily the reading and filtering of this faculty, i.e. of the carrying out of predictions that the mechanisms responsible for choice have calculated are safe.

Rational Otherwise Spiritual Practices

rationalityperceptionconsciousnesslarge

I'm still not spiritual in the slightest, but I've seen enough to understand that people often have very compelling reasons for turning down the spiritual path. Right now I'm trying to figure out how to release the "Kundalini energy coiled up like the snake Shakti in the base of the spine" — I've seen enough to know for a fact that it, or something very close to it, is real. Science, or, more accurately, scientific culture, just wants to completely ignore such effects due to their extremely simplistic physicalism.

The main problem I'm grappling with is causation: could these phenomena have been said to have existed in me as they are now, yet below my perception, before I started any sort of insight or concentration meditation; could they be said to have been latent before and awakened now as a result of practice? could they be said to have not existed before, instead being created as a byproduct of the practice? I'm starting to come to the conclusion that for most of these phenomena it's between 2 and 3, and far closer to 3 than 2, as if it were 1, you could just... show people the way without having to make them pliable with previous practice, and kundalini/siddhi stuff would be common knowledge. This would explain most of the odd state of spirituality.

I think many people are accurately reporting their experience when they say things like "I saw an infinite plane of gods, shot bubbling electricity out of my nose, and felt like I could read the thoughts of every being that ever existed—and we were only five minutes into the work meeting!". But the brain has a MASSIVE bias to take what it perceives as being real no matter how ridiculous — this goes for thoughts too, as made clear by all the people who come out of LSD with "realizations" — and this settles cozily into a nice pre-existing framework where they really do have powers and really are special in this way.

Some people have incredibly strong natural tendencies towards these things (by "these things" I mean kundalini, siddhis, chakras, devotionals, etc.), and some people have incredibly strong natural tendencies towards other things which make the mind pliable towards these things, and these people will seem to figure that everyone must have these perceptions, they're just culturally inclined into dismissing it &c. — the latter is actually a perfectly reasonable inference from the former, and the former is while not that reasonable very understandable. We just tend not to understand how amazing human brains are at picking up on social things and cultivating strange perceptions. So it's extraordinarily usual and predictable that many practices will go on to spread communities of people showing their non-real powers to each other and believing in their reality; we've had tens of thousands of years to evolve verbal communication, but millions of years to evolve non-verbal communication and influence, so how could you possibly be surprised when the mind, leashed in its cognition by its use of language, subliminally spreads these shared understandings through implicit nonverbal communications that it is not capable of legibilizing and therefore exposing to itself or others?

I'm starting to think awakening (in the Buddhist sense, translation of "bodhi", commonly less-accurately-translated as enlightenment) is another one of these cases. People aren't realizing that they don't have a self, they're EXCISING their selves; they look inside after awakening only to find no-self, conclude that the texts were right that we simply need to realize no-self, tell their students that they simply need to realize no-self, the students that do genuinely have selves then excise them under this advice, and, becoming teachers, look inside after awakening only to find no-self; the cycle repeats for 2,600 years. The Buddha saw the first domino in this chain under the Bodhi Tree and knocked it right over; meme spreads throughout a series of cultures longing for a single religion with a single doctrine, a single figure to devote themselves, a simple set of practices, a simple end-goal; no-self long con propagates through the Sangha, and here I am in 2022 CE holding up this single domino before I know it's safe to let it fall over.

Through sheer luck, I plunged into the depths of no-self and managed to come back out okay, and have consequently seen that full attainment of no-self is a distinct possibility but also a fundamental change from how I currently am. Full awakening is a distinct possibility that I could clearly achieve if I wanted; on the other hand, I may retreat out of caution, halting all insight practice permanently, taking the concentration and existing insight attainments as a consolation prize in lieu of an awakening that may very well alter my cognitive ability for the worse.

Attention in the Blind Spot

phenomenologyperceptionfacts

Now I can finally make things disappear by looking slightly away from them! I had to learn to see the part of my visual field corresponding to the fovea, which I did entirely by accident during an especially intense battle with my attention... It's weird, and I can only do it because I've gotten so good at decoupling eye movement from visual mental attention, but it's a sort of... attentional energy that distinguishes these two parts. The fovea is "where attention wants to go", such that I can naturally and easily focus my visual mental attention on it, but the blind spot is "where attention doesn't want to go". If I'm probing a constant visual field by moving my visual mental attention around, the fovea is like a sort of magnified space, attention comes in and it can just move around freely and hang out comfortably without coming out, but the blind spot is like a sort of non-space, attention just naturally flows around it as running water naturally flows around a rock.