Speech is built from sounds resembling the fundamental physical events of solid objects interacting in the world. Writing evolved letters and visual forms that look like the kinds of contours and junctions found among opaque objects strewn about natural scenes. Music is structured to sound like the movements and expressive behaviors of another human being.

These are not arbitrary conventions laid on top of a generic brain. They are extraordinarily sophisticated culturally evolved technologies, shaped over generations to exploit the perceptual and cognitive instincts nature had already given us.

And their power is difficult to overstate. They literally helped transform us from ape to man — from Human 1.0 to Human 2.0.

Indeed, what we take ourselves to be is so thoroughly entangled with these technologies that we barely identify with the creature underneath them. Strip away language, writing, music, art, and the rest of accumulated culture, and what remains is a clever ape — but not remotely the creature we recognize as ourselves.

This is part of what I call Zoocentrism: to understand humans, start by remembering that underneath all this cultural machinery we are animals, equipped with ancient animal instincts. The great trick of civilization has been to build technologies that fit those instincts so well that they come to feel natural.

Which brings me, strangely enough, to units of measurement.

Our traditional systems of measurement were not dreamed up at a committee meeting. They culturally evolved over millennia. And, like language, writing, and music, they are positively reeking with design — accumulated solutions shaped around human bodies, human perception, human action, and the quantities that actually matter at human scales.

Metric, by comparison, is a base-10 ladder.

That’s basically it.

It has one enormous virtue: unit conversion is easy. And if you sat someone down on the first day of civilization and told them that easy conversion was the only design objective, they could invent essentially the same thing before lunch. Powers of ten. Move the decimal point. Done.

That simplicity is often mistaken for superior design.

But culturally evolved technologies tend not to be simple, because humans aren’t simple. They accumulate strange-looking structure precisely because they have been adapting to us — our bodies, our senses, our actions, our environments, our needs.

Traditional units deserve to be looked at through that lens.

The point of what follows isn’t merely to defend feet, inches, miles, pounds, cups, or degrees Fahrenheit because they’re familiar. It’s to show that beneath their apparent messiness lies something much more interesting: deep human-scale design.

We inherited these constructs after thousands of years of cultural evolution. Before throwing them away in admiration of a base-10 conversion chart, we might first ask what all that accumulated structure was for.

I’ve been struggling to diagnose the prejudice against the use of AI among many people, including when it comes to AI helping with one’s prose. I find the preoccupation asinine, superficial, and ultimately a signal of a lack of substance.

In my world, it’s all about the idea. As a theorist, I’m good at finding ideas, and at providing evidence and arguments for them.

Getting an idea written up in the best form for an audience is a needed further task. It’s important. It needs to be done. Over the years, I got good at explaining my ideas. But if I could make that task easier — with a writer always at my side whom I trusted, who grasped my work and could explain it really well — I’d leap at the opportunity. More time to work on what I’m all about!

And so AI comes along, and it’s currently really, REALLY good at grasping existing ideas, and also exceedingly good at explaining them to others. Sure, its prose often has its own look that we’ve learned to recognize, but I don’t care about that. Do you understand the idea? That’s what matters.

If you’re the type to fuss, “Did an AI help you write up your idea?”, you’re not the audience. You’re just about the presentation, and not even in a substantive sense. “Sure, the prose really does explain the point well. But, muh, I can tell it’s gone through an LLM.” 👏 👏 Good job missing the point — missing the substance.

If you write a piece, I want to (i) understand what you’re trying to say, and (ii) grasp what, if anything, the new idea is. What I do NOT give a crap about is whether you used AI to help get your point across better.

That is one of the brilliant consequences of AI: those interested in breaking genuinely new ground can focus more of their time on exactly that.

But, some may ask, isn’t AI also generating the idea?

No. Not even a little. Except in highly formal realms like mathematics, LLMs are NOT currently good at finding substantively novel ideas. And, as I have argued, LLMs are likely some ways from getting good at breaking genuinely new ground, especially in the sciences.

We’ve had spell-checkers since the late 1970s, grammar and style checkers since the 1980s, and increasingly sophisticated word processors helping rewrite sentences since the 1990s. Now we have AI that can help with prose at the whole-paragraph level and beyond.

The “AI sLoP!” screamers are the grannies bragging about how things were done in their heyday.

The interesting question was never whether a machine touched the words. The interesting question is: What did you contribute?

Did you discover something? Did you have an idea worth communicating? Did you find the evidence for it, work out the argument, see something the rest of us hadn’t seen? Or did you ask an AI to come up with the substance too?

Those are profoundly different things. Collapsing them into the single question “Did you use AI?” is about as superficial an assessment of intellectual work as one could imagine.

In fact, AI should push us in precisely the opposite direction. As producing competent prose becomes cheaper, we should care less about who physically assembled each sentence and more about where the ideas came from.

That’s the great opportunity here. Let the machines help with the parts they’re good at. Spend more human time on the parts they’re not.

Find something new.

Then explain it as well as you possibly can — no matter who does the explaining, your hired writer or an LLM.

Intelligent Design is widely regarded as an embarrassing cautionary tale about human reasoning. Yet the same pattern shows up almost everywhere. Again and again, we accept that a complex, decentralized, emergent process explains most of what we observe — until it produces something we find especially remarkable. At that point, we decide the process cannot possibly be enough, and conclude that something more must have been added.

~ Intelligent Design
Natural selection explains adaptation and ordinary evolution — but it cannot explain the bacterial flagellum, the vertebrate eye, human intelligence, and other seemingly irreducible biological innovations. There must have been a designer.

~ Central Planning
Markets can coordinate ordinary production and exchange — but surely an advanced economy’s most important sectors are too complex and consequential to be left to decentralized prices and decisions. There must be expert planners directing the crucial outcomes.

~ Conspiracy Theories
History unfolds through countless interacting people, institutions, incentives, accidents, and feedback loops — but events this consequential, or this seemingly coordinated, cannot have emerged on their own. There must be a hidden cabal.

~ Truth and Knowledge
Open inquiry, criticism, replication, and debate can uncover many truths — but surely they cannot be trusted on the most important or dangerous questions without authoritative institutions and fact-checkers deciding what is reliable. There must be arbiters standing above the debate.

~ Metaphysical Free Will
The brain deliberates, weighs reasons, and generates behavior — but that cannot really account for genuine choice. There must be a steerer behind the deliberation, somehow able to go beyond what the brain’s mechanisms and the laws of physics would otherwise produce.

~ The Hard Problem of Consciousness
The brain perceives, integrates information, remembers, attends, and reports — but that cannot really account for subjective experience. There must be a watcher behind the processing, somehow possessing a “what it is like” that goes beyond anything the brain’s physical and functional organization entails.

~ Human Uniqueness — Against Zoocentrism
Ordinary primate cognition, harnessed by cumulative cultural evolution, explains much about humanity — but it cannot explain language, music, mathematics, religion, science, and civilization. There must be something uniquely human that was added, whether a special package of human instincts or, paradoxically, a uniquely unconstrained blank slate.

The recurring intuition is always the same: the accepted decentralized mechanism explains ordinary cases, but cannot possibly be sufficient for the truly extraordinary ones. So we conclude that something special must have been added.

There’s a problem with this that I hadn’t really appreciated. A big one. What happens to the incentive to create the generative core? Today, if I spend five years developing a genuinely new idea and then another three years writing a book around it, copyright protects the book. Sure, you can understand my idea and use it yourself, but historically that only got you so far. If you wanted your own book, movie, or game built around it, you still had to do the enormous work of actually making one. AI is killing that barrier. Soon I may spend years discovering the idea, reveal it in my work, and you simply extract the underlying idea and ask AI to make something completely different from it — perhaps something you like better than mine. No copied sentences. No copied characters. Maybe no recognizable similarity at all on the surface. Copyright working exactly as designed. Except the valuable thing I discovered is now yours.

And that sets off a nasty cascade. First, why spend five years making the finished book or movie if bespoke instantiations are becoming cheap? Fine — that was my earlier essay. Forget the finished project. Develop and publish the generative core instead. But then why publish the generative core if publishing it amounts to giving it away? And if that’s the case, why spend five years struggling to discover the generative core in the first place? AI starts out threatening the incentive to finish things. Follow the logic far enough and it threatens the incentive to find genuinely new things in the first place.

Oddly enough, we solved something like this problem centuries ago. Just not for books, movies, games, scientific conceptions, and most other kinds of ideas. We solved it for inventions. We call the solution a patent.

Patents and copyrights protect quite different things, and historically the distinction made a lot of sense. If you invent a new machine, process, or technology, a patent can protect the invention across different implementations. You don’t just own the particular prototype sitting on your workbench. In fact, you don’t necessarily need to have built a prototype at all. If the invention has been sufficiently conceived and disclosed, and meets the other requirements for patentability, the intellectual achievement can exist before any finished product does. Someone else can later build it at another scale, with different materials and all sorts of engineering changes, and still be implementing your invention.

Why protect something so abstract? Because the expensive part may have been figuring it out. You spend ten years discovering how to do X. The day you reveal X, I understand it and start making my own version. You bore the cost of discovery; I got much of its value essentially for free. If that happens systematically, fewer people bother making those costly discoveries. So we make a bargain with inventors: tell us how your invention works rather than keeping it secret, and for a limited period we’ll give you certain rights over implementations of it. Patent law has long recognized that the important intellectual creation need not be a particular artifact. It can be the idea that generates the artifacts.

Copyright made almost the opposite bargain. We very deliberately do not let people own ideas. You can copyright your novel, not “two lovers from hostile families.” Your movie, not “aliens invade Earth.” Your game, not “civilizations compete for territory.” And thank God. Ideas are the raw material for more ideas, and culture would suffocate if someone could plant a flag in every concept and demand payment from anyone who wandered nearby.

But there was also a practical fact making this division work, so obvious that we barely had to think about it. Having the idea was nowhere near having the thing. An idea for a novel isn’t a novel. An idea for a movie isn’t a movie. An idea for a video game isn’t a video game. Between conception and finished work stood an often brutal amount of writing, rewriting, programming, acting, filming, composing, drawing, editing, designing, financing. We could leave ideas free because there was still an enormous amount of difficult work between having one and producing the thing people actually wanted.

AI is destroying that practical fact.

The distance between idea and instantiation is collapsing. Suppose I spend years developing some genuinely new conceptual structure — call it G. From G one could make a novel, movie, game, television series, interactive world, whatever. Until recently, knowing G didn’t remotely give you all those things. Every one of them was another mountain of work. Now that mountain is starting to look more like a button.

So I finally reveal G in a novel. You read it, figure out what makes it tick, and tell your AI: “Take this underlying idea, but don’t copy his book. Different setting. Different characters. Different prose. Make it funnier. Make the protagonist more like me. And actually, don’t make it a book — make it a twenty-hour interactive experience.” The result might infringe nothing. That’s the point. I’m not worried here about AI getting better at copyright infringement. I’m worried about what happens when AI gets very good at NOT infringing copyright — at extracting the generative idea from a work and then going somewhere completely different with it.

Now our old intellectual-property arrangement starts looking strangely backwards. We protect the artifact just as artifacts are becoming cheap to generate, while leaving the generative idea unprotected just as the idea may be becoming the scarce part. AI may be turning what used to be copyright problems into the kinds of problems patents were designed to solve.

This doesn’t mean we should simply start patenting ideas. Horrifying. Someone patents “a detective who turns out to be the murderer.” Someone else gets “a society controlled by AI.” Another person owns “a multiplayer game about territorial conquest.” Within six months human culture is a minefield of infringement claims. The old rule that ideas remain free exists for very good reasons.

But maybe “idea” and “finished work” aren’t the only two categories we need anymore. There’s something in between. That’s where the generative core comes in. Not a theme. Not a genre. Not a one-sentence premise. The minimal but sufficiently precise conceptual machinery that makes an idea distinctive and allows all sorts of different things to be generated from it.

AI makes this less philosophical than it sounds. Give it the core and it can instantiate it. Give it an instantiation and it may be able to infer the core. Two works could share almost nothing obvious on the surface and still derive from the same underlying machinery. So “Did you copy my work?” may eventually miss the interesting part. Another question starts to matter: “Did you instantiate the generative idea I discovered?”

That sounds a lot like the question patents already ask.

I don’t know what rights, if any, should eventually attach to a generative core. But there is a much simpler place to start: attribution. We need some way to say, clearly and publicly, who originated a sufficiently specific new idea and when. Even if that initially carries no legal rights whatsoever. Before we can decide whether people should have any claim over the downstream instantiations of their ideas, we at least need a serious way of keeping track of where those ideas came from.

Credit is an incentive too. And once generative cores can be stated, dated and attributed, we can begin arguing about whether anything more should follow. The important first move is recognizing the core itself as an intellectual contribution rather than treating only its particular instantiations as things with authors.

Because the old bargain is coming apart. Copyright arose in a world where we could leave ideas free partly because turning them into valuable finished things remained hard. Patent law addressed domains where the discovery itself might be the hard part, and where, once revealed, it could become relatively easy for others to build from it. AI is making more of human creativity look like that.

And that could be wonderful. I meant it when I wrote that the disappearance of the finished project could unleash entirely new forms of creativity. We might spend more of our lives developing deep new ideas and less of our lives grinding out one particular realization of each of them.

But only if people still have a reason to reveal those ideas. And before that, a reason to spend years finding them.

If AI gives everyone an astonishing machine for instantiating ideas, while we have no meaningful way even to preserve who originated those ideas, we may end up with exactly the wrong incentives: enormous demand for new ideas, almost unlimited ability to exploit them, and less reason to do the difficult human work of finding them.

We invented patents because, at least once before, we realized that was a bad bargain.

I’ve been working for several years on a new, deeper theory of the origins and meaning of music.

In Harnessed, 15 years ago, I described how music sounds like a human moving evocatively in your midst. But I did not then have a theory of why we tend to like the usual (diatonic) scale, what chords mean, what melody means, or how they all fit together.

But now I do. And it’s breathtakingly beautiful.

And, by the way, AI hasn’t helped a bit. Not that I didn’t try to get useful help — it’s just nowhere near ready to make these sorts of creative leaps in science. Maybe next year. Lol.

The theory — MOTION SPACE THEORY — in its most unifying (and least reader-friendly) statement is this:

Motion Space Theory proposes that the diatonic scale’s Thirds Space has the geometry needed to represent the changing directions of human movement along a trajectory past an observer. The bulk motion of the extended body through this space gives rise to chords and harmony, while the oscillatory motion of its articulated parts gives rise to melody. The diatonic scale is unique in its ability to enable all of this.

What I find most surprising is that chords and melody — two seemingly very different pillars of music — fall out of the same underlying theory. And they get grounded in exactly the thing that could explain why music is so emotionally evocative, and why it is so deeply connected to dance: the detailed dynamics of a human moving near you — their trajectory, parallax, rotations, and the oscillations of their head, limbs and other body parts.

Working on the final article now. Or book.

Although, as I’ve discussed recently, I’m not sure there’s quite the reason to write books anymore. Once I write the paper, you’ll be able to have your favorite AI create a bespoke book for you about the theory. 🤷

(The diagram is a little peek under the hood. Don’t worry, you’re not supposed to understand it yet. 😎)

My eyes were open. I could think clearly. I knew where I was. But I could still feel the object I had been holding moments earlier, in a dream that had ended. I couldn’t seem to move my arm into view, and I felt almost nothing from the rest of my body.

For a few extraordinary moments, what we ordinarily call “being awake” appeared to have come apart into its constituent pieces.

This happened after I had been dreaming that I was holding a gun. As I lay there in bed, eyes open and apparently awake, I noticed I could still feel the gun in my hand. I remembered the sensation clearly: it felt plastic and fake, even though in the dream it had been functional.

So now I was confused. Why was I apparently holding a gun here in bed—the same one from my dream?

I moved my arm downward to bring it into my line of sight. But no matter how much I tried to move it, my arm never appeared. It remained up and to the right, out of sight.

The sensation was unique: moving a body part, feeling oneself move it, yet never seeing it enter the visual field. It was not as though I felt the arm pass invisibly across my vision. Even my proprioception, although it told me the arm was moving, never registered any change in its position.

So there I was in bed, having awakened from a dream, trying to see what I was holding that felt like a gun—yet no amount of motor effort could make my arm budge from where it seemed to be. I was perplexed, a little annoyed, and perhaps mildly concerned. Why was my motor system failing? And what was in my hand?

Then a wave of sensation spread across my body. I could feel it coursing over my skin. I believe this was the moment my somatosensory and motor systems woke up and I could once again feel my actual body.

What returned was merely the ordinary sensation of having lain in the same position for a long time. We barely think of it as a sensation at all. It is part of the baseline we normally ignore. But before it returned, I had felt none of it. My bodily sensation had apparently been switched off, and so when this mild, boring feeling of almost nothing came rushing back, it felt intense.

Only then did I realize that, before that moment, I had felt nothing across my body at all.

This appears to have been a variety of lucid dreaming in which my visual and cognitive systems had awakened, while my somatomotor system—and perhaps even parts of my visuomotor or dorsal-stream system—had not.

I was half awake, but not in the ordinary sense of occupying some fuzzy boundary between sleeping and waking. Rather, half of my brain seemed fully awake while the other half remained fully asleep.

Whether that interpretation is exactly right or not, the experience left me with a deeper appreciation for just how many independent systems have to work together to produce the seemingly simple experience of being awake. We think of wakefulness as a single switch. For a few unforgettable moments, I experienced it instead as a collection of distinct faculties that had briefly fallen out of sync.

Musk’s latest Tesla spectacle reminded me of something I’ve been meaning to write about for years: fashion shows.

Many people don’t get them. “Who’s going to wear that?!?!” That misses the point. A fashion show is demonstrating the heights and limits of their new design: colors, patterns, fabrics, cuts, fit, etc. If we can do ALL THIS, imagine what our designs can make possible for the clothes you can realistically wear.

Not that most of us can imagine — we’re not fashion designers, and are not creative in that way. But it garners confidence in the audience: They’re going to bring some radical innovations to real clothes — I can’t wait to see them.

It is also meant as a spectacle. A joy to watch, not a meeting with the board of directors about their 5 year business plan. (And it’s art! The most ancient art: decorating the human body, transforming it, because one perceives clothes not as occlusions, but as the body itself. But that’s a whole other story!)

Same for Musk / Tesla’s demo. Almost word for word as in fashion. “Who is actually going to want their car to do that?!?!” As before, that misses the point. Tesla is demonstrating the heights and limits of their new design innovations: the suction and propulsive capabilities. The point isn’t that your car is going to do that. The point is: We have whole new revolutionary toolboxes at our disposal, so imagine what we can do for your everyday car.

Again, most of us know little about car engineering, and our creativity doesn’t take us very far at predicting what a real car is going to be able to do. But it breeds confidence: “They’re going to bring some radical innovations to real cars. I can’t wait to see them!”

When I asked ChatGPT — because I’m not a car engineer or designer — “What possible applications could that kind of technology have, even if they look nothing like the feats at the Tesla demo?”, it gave a variety of suggestive guesses. Floor suction could mean radically better cornering, braking, acceleration, wet-road grip, and stability — even at low speeds. Propulsive thrust could mean assisted acceleration and braking, skid and spin correction, control when the wheels lose contact, and emergency maneuvers beyond what tire grip alone allows. Who knows which of these, if any, will matter. That’s the point: these are new engineering powers, and most of us aren’t the people who know what to do with them.

These spectacles aren’t product demos. They’re FLEXES. They’re saying: “Look at the new powers we have at our disposal.” What eventually gets built with those powers may look nothing like the flex that showed us they were possible.

Getting these animals tipsy and stabbing them through the eye with a stake turns out to be much less impressive than when Odysseus did it.

For vertebrates, we find no cyclopses. On some fish, frogs, lizards, and relatives, there is a third eye (a “parietal eye”) directly on the tops of their heads (middle photo below), which is cyclops-like in not being paired with another eye. However, although these eyes are light-sensitive, they are not really eyes at all, because they are not capable of forming an image of the outside world. Instead, they seem to be involved in thermoregulation, although no one is quite sure.

The closest thing higher vertebrates have to cyclopses are anencephalic infants, like the kitten shown below. But that’s nature’s mistake.

One reason for the scarcity of cyclopses is that they can’t see what’s behind them.

Most of the Earth’s animals with vision, however, can see much of what’s behind them, and presumably need to.

They do this by having two eyes facing in opposite directions.

This is true, for example, of squid, insects, fish, reptiles, dinosaurs, birds, and many mammals like rabbits and horses (see photos below). You can’t sneak up on these animals from behind.

Let’s take notice of something so obvious about the animals pictured above that one is apt to overlook it.

These animals have eyes facing in opposite directions, as I mentioned, but more specifically their eyes are on the sides of their heads, so that the eyes point sideways, as in the drawing of the animal on the left below.

But this is not the only possible way of having panoramic vision.

One could instead place a single eye on the very front of the head and another on the very back, as illustrated below and on the right. You never find this in nature – and not even in fiction, as far as I know. Why not?

The reason has to do with the fact that those sideways-facing eyes don’t just see opposite sides of the animal.

Instead, they typically also have some overlap: parts of the world that both eyes can see, called the binocular region.

Seeing a part of the world via two eyes has certain advantages, advantages that animals typically want to bring to bear in front of themselves.

For animals with sideways-facing eyes, the binocular region does tend to be in front.

For example, for the drawing of the animal above and on the left, you are in its binocular field because you can tell that each of its eyes can see you.

For animals with one eye on the front of the head and another on the back, however, any overlapping regions of the visual field would be along the sides of the animal, not the front.

And animals don’t tend to want their especially powerful vision directed to the sides.

Our vision is a variant of having eyes on the sides of the head, in the sense that our eyes are placed on the left and right of our midline, and so our binocular region is also in front.

What are these advantages of binocular vision for seeing the world?

The typical answer given by anyone who knows anything about vision will be stereo vision, or the ability to see in depth. The study of binocular vision is so historically wrapped up with stereo vision that courses and books about binocular vision tend to be largely about this special kind of depth perception.

Contrary to conventional wisdom, however, in my research I have come to appreciate a power of binocular vision no one has yet noticed: the power to see through stuff. As I discuss in my book, The Vision Revolution, understanding this “X-ray power” is crucial for understanding why we have forward-facing eyes.

But in this piece I want to point out a special, very fundamental kind of “X-ray vision” binocular vision gives us, one that helps further drive home the advantages of panoramic vision via sideways-facing eyes rather than via front-and-back-facing eyes.

It is the power to see through yourself.

Before an animal can worry about how to most ably see what’s out there in the world, it has to make sure it can actually see what’s out there. A fancy eyeball placed on the inside roof of its mouth would not see much of the world.

But where should you put the eyes?

Most spots will be better than the roof of the mouth, but bodies tend to have gangly parts that could possibly block an eye’s view.

And, in fact, one typically wants some of those body parts to be out in front of the eye: body parts like a mouth, nose, whiskers, or hands that are designed to physically interact with the things out in front. Animals want these appendages out in front of the eye not only because these appendages are good at touching the world and eyes are not, but also because it is useful to see one’s own appendages so as to help guide the appendages’ interactions with the world.

And that leads to a riddle that animals with vision had to solve.

Namely, how can they put appendages out in front of their eyes, yet not occlude their view of the world?

This is a riddle that video game makers have had to grapple with as well. How do you let the player see his own character without obstructing the view of the game?

If you play games in first-person perspective mode, then if your character is holding a gun or sword, its arm and weapon occlude a significant portion of the screen.

If you play in third-person perspective mode – where you see the entire character from behind – then there are also significant parts of the view missing due to the character’s whole body blocking the view.

That’s why many games allow the player’s viewpoint to roam around the character, rather than being locked with a view of the character’s back. Allowing our eyes to float, Steadicam-style, around our bodies and relay information to our brains would, then, be a potential solution to the problem of seeing past our own appendages.

Nature found a different solution, however, and you can see it for yourself by looking at your own nose.

If you close one eye and wiggle your nose, you’ll see it in the bottom corner, blocking that eye’s view of whatever is beyond it, as illustrated in the pair of upper images in the figure below.

But if you now open the other eye, you’ll see what’s behind your nose via this other eye, as illustrated by the lower image below.

In fact, you’ll perceive your nose still to be there, but perceive it to be transparent, with the scene beyond visible through it. Each eye sees the nose, and each has its view blocked by the nose. But the views blocked by the nose are different for each eye, so the pair of eyes, in total, don’t miss a thing – and yet can also see your nose.

If, instead, an animal’s eyes were on the front and back of its head, then the appendages out in front of the front eye would simply occlude its view.

Having eyes on the sides of the head – or on either side of the midline, like ours – rather than on the front and back is not only useful for placing the powerful binocular region in front of the animal, for the purpose of better perception of what’s out there.

In addition, having eyes on the sides of the head is crucial for seeing past one’s own bodily appendages.

And not just seeing past one’s own appendages, but simultaneously seeing them and thereby gaining the advantages that come with having this visual feedback.

Seeing through oneself may be one of the most fundamental reasons why animals have eyes on either side of their head and possess a binocular region.

This was adapted from Chapter 2 of The Vision Revolution (BenBella, 2009).

There is a very different way to think about the foundations of physics.

The usual picture is familiar. The universe begins in some initial state, and the laws of physics determine how that state evolves from one moment to the next. The initial state supplies the starting point; the laws supply the transformations. Each new state follows from applying those laws to the preceding state.

In computational terms, the laws of physics are the algorithm, and the initial state is the input. The algorithm computes the next state of the universe, that output becomes the input for the following step, and the process continues indefinitely.

But computation can be organized in another way.

Consider the distinction between an ordinary Turing machine and a universal Turing machine. In an ordinary Turing machine, the transition function is specialized for the particular computation being performed. The machine itself contains the algorithm, while the tape supplies the input.

A universal Turing machine works differently. Its own transition function is fixed once and for all. Written on its tape are both a description of another machine—the program to be executed—and that program’s input. What had previously been built into the specialized machine has now been transferred onto the tape. The universal machine itself becomes a general-purpose interpreter, repeatedly carrying out the same fixed machinery on whatever program and input have been encoded in its initial state.

The contrast can be seen even more cleanly with one-dimensional cellular automata.

Imagine a row of cells, each of which can be either on or off. At each time step, every cell changes according to a local rule that looks only at that cell and its immediate neighbors. There are 256 possible elementary local rules of this kind.

The ordinary procedure is to choose one of those rules, choose an initial row of cells, and then watch the system evolve. This mirrors the conventional picture of physics:

Initial state + transformation laws → history of the universe

Different rules produce very different characteristic forms of behavior. The rule therefore appears to carry most of the explanatory burden.

But one of those elementary cellular-automaton rules—Rule 110—is special. Stephen Wolfram identified its remarkable behavior, and Matthew Cook proved that it is computationally universal. With a sufficiently constructed initial configuration, Rule 110 can emulate any Turing-machine computation.

This means we no longer need to select a different local transformation rule for each different computational world. We can hold Rule 110 fixed and encode both the program and its input into the initial configuration itself.

We could therefore have arbitrarily different simulated universes, exhibiting arbitrarily different effective laws, while every one of them evolves according to exactly the same primitive local rule. All of the differences between those universes—including what would ordinarily be called their laws—would be encoded in their initial states.

The universal update rule would still be necessary, of course, but it would contribute almost none of the information specific to any particular universe. It would simply execute whatever had been encoded into the starting configuration. The real explanatory work would be done by the initial state.

This flips the usual way of thinking about physics on its head.

Given a universe exhibiting some particular laws, there are two very different ways its evolution might be represented.

In the conventional representation, the universe has a relatively contingent initial state together with substantive transformation laws. Those laws explain why the universe unfolds in its characteristic fashion.

In the alternative representation, the universe has a fixed universal transformation rule, while its initial state encodes both its effective laws and what we would ordinarily call its initial conditions. The universal rule serves merely as general computational infrastructure. The explanation for why that universe unfolds as it does lies in the structure of the initial state.

These are not necessarily two different universes. They can be two computationally equivalent descriptions of the same unfolding. What is built into the transformation rule in one representation can be transferred into the initial state in another.

This kind of equivalence is well known in computer science. A specialized machine can have a particular algorithm built into its transition function, or a universal machine can keep its transition function fixed and receive a description of that algorithm as part of its input.

Applied to physics, however, the second representation is much less intuitive. Physics normally treats the laws as the deep explanatory objects and the initial conditions as the comparatively accidental facts to which those laws are applied. The computational reformulation permits the exact reverse: the transformation rule becomes nearly content-free, while the initial state contains the structure responsible for the universe’s apparent laws.

I find this interesting because, although I am an outsider to academic foundations of physics, I am not familiar with physicists adopting this second representation as a central way of thinking about nature. Physics generally seeks the correct fundamental laws governing how the universe evolves, while treating the initial state as the contingent material upon which those laws act. It does not ordinarily begin by fixing a universal update rule and asking what principle could determine an initial state that encodes the effective laws of our universe.

The two formulations may ultimately be equivalent. But equivalent formulations need not be equally fruitful. They assign explanatory importance to different objects, make different questions appear fundamental, and suggest different routes toward a theory.

Universal computation shows that the distinction between law and initial condition may be representational rather than fundamental.

I have an idea for one. It would be about the evolution and meaning of Western tonal music — the strange cultural system built around the diatonic scale and the chord progressions with which we’re all familiar.

And writing books is hard! I’ve written many. I know.

So shouldn’t I be more motivated to write the next one today? Producing a book is becoming easier than it has ever been. Now, if I actually wrote it today, I’d write it myself, without AI, because I’d want it in my own voice. But set that aside. I’m interested here in the motivations that are going to increasingly take hold as AI gets better.

Until now, writing a book about a scientific discovery required me to be good at two very different things. First, I had to make the discoveries. Then I had to turn those discoveries into a book that was organized, readable, entertaining, accessible and hopefully difficult to put down.

Why should I have to be good at both?

I’m a scientist and theorist. Perhaps I should concentrate on the ideas themselves and let my writing staff explain them to the masses. The fact that my new writing staff happens to be AI just means I no longer need to employ a room full of people to do it.

So I should write my next book with little compunction!

Right?

No.

There are two obvious reasons why the coming ease of creation might actually make creators lessinclined to create.

First, if anyone can do it, the mere fact that you did it loses much of its value.

Second, if everyone can suddenly turn every half-decent idea into a polished book, movie, documentary, album, lecture series or whatever else, there is going to be a massive surge of completed visions competing for everyone’s attention. That alone is quite a disincentive to spend your time making another one.

But neither of those is the big reason.

I’d be willing to compete. My ideas — I think — are better, and one needs to get them out into the competition. If they’re actually better, they should fare well in time.

The trouble is a third and much deeper reason.

And you may already be exhibiting it yourself.

Suppose someone posts an interesting article on social media. Do you always click through and read the whole thing?

Or do you sometimes ask your favorite AI to summarize it?

Someone sends you a 45-minute YouTube video. Maybe you ask Grok to give you the skinny on it first.

If the summary sounds interesting, perhaps you then go back and read the entire article or watch the whole video.

Or will you?

Because you’re already talking about the idea with your AI. It knows what the creator was trying to say. And now it can explain the idea specifically to you.

At your level. Given your background. Given your interests. Given how much time you have. Interactively. With more detail wherever you want it and less wherever you don’t. With objections and counterarguments. With diagrams or other visual aids whenever they would help.

Why go back and consume the creator’s particular instantiation at all?

Sometimes there is an obvious answer. If the instantiation is the art, then of course you want the instantiation. You don’t ask an AI for a summary of a painting and figure you’ve now seen the painting. For some creations, the finished artifact may itself be the shortest specification of the idea.

But very often that isn’t true.

And in those cases, you may never return to the creator’s project at all.

There would seem to be a problem with this argument. The creator still had to make the article or video in the first place. That’s how your AI learned what the creator had to say.

Except it didn’t have to happen that way.

Just as I could now hand AI the substance of a scientific discovery and ask it to expand that into a book, the AI on your side doesn’t actually need my book in order to explain my discovery to you.

It needs the generative core from which my book would have been made.

And if your AI has that, something very strange happens.

I don’t need to write the book.

Your AI can write the book.

Not my book, exactly. Your book.

It can instantiate my idea on the client side, specifically for you.

Which means that even though AI has made it spectacularly easier for me to turn my idea into a completed project, there is increasingly little reason for me to bother doing so.

Everyone else can instantiate it themselves.

The End of the Finished Project

This will happen most obviously when the central value of a work is logical, informational or otherwise compressible rather than residing in its exact instantiation.

Educational books and videos are paradigm examples.

Why should an expert turn an idea into one particular 300-page exposition intended for some imaginary average reader?

One reader needs 40 pages. Another needs 600.

One already understands the mathematics but needs the biological background. Another knows the biology but needs the mathematics explained from scratch.

One wants diagrams. Another wants exercises. Another wants an adversarial presentation that spends half its time trying to destroy the theory. Another wants a two-hour conversational audiobook for a drive.

Why should the creator choose?

But the same thing can reach considerably further into what we now think of as creative works.

Consider a novelist who prides himself less on the precise sentences than on an ingenious story: its world, characters, revelations, causal structure, intricate plot and character development.

That structure might be specified much more compactly than a 400-page novel.

If it’s good, a million people could take that same canonical story and have their own AIs instantiate it as a novel of whatever length and style works for them. Or as a movie. Or a ten-episode television series.

Every one of them might experience a somewhat different artifact while still meaningfully experiencing the same story.

That isn’t even as alien as it sounds.

We’ve been doing something like it with novels for centuries.

You and I can read the exact same novel, yet the worlds generated inside our heads may be radically different. My character has a different face. Your house has a different layout. We hear different voices. We imagine different lighting, timing, expressions and movements.

Then Hollywood makes the movie and produces yet another instantiation.

Yet we have no difficulty saying that we all experienced the same story.

AI simply externalizes and vastly expands the process of instantiation that our brains were already performing.

And that leads to the paradox:

The easier AI makes it for us to instantiate our projects, the less reason there is for us to instantiate our projects.

Instead, we’ll increasingly be incentivized to create just the generative core of the project — the canonical specification of the vision.

Those interested in it can then have their own AI create a bespoke instantiation, tailored to their interests, needs and level.

Historically, cultural production has looked something like:

idea → instantiation → client

discovery → book → reader

AI increasingly allows:

idea → client → bespoke instantiation

discovery → reader → book tailored to that reader

Notice what has happened.

The instantiation and the client have switched places.

For most of cultural history, I had to instantiate my vision before knowing you. I had to choose the words, length, examples, pacing, visuals and medium, freeze all of that into an artifact, and then send that artifact out into the world in search of people for whom those choices happened to work.

Now the client can arrive first.

Only then does the instantiation need to exist.

So What Should I Write?

Which leaves me with the distinct impression that my next book — perhaps this new idea about the cultural evolution and meaning of Western tonal music — shouldn’t be a book at all.

I should create the thing from which the books would be generated.

That thing would have to be designed differently.

It should be optimized not primarily to be read straight through by a human, but to be completely and accurately grasped by an AI. It should contain the full structure of the theory: its claims, definitions, evidence, logical dependencies, examples, qualifications, implications, boundaries, objections and replies.

It would need to be generatively complete.

In some ways it might resemble a patent. A good patent is not merely one particular implementation of an invention. It attempts to specify what the invention is, across a range of possible implementations, distinguishing what is essential from what is incidental.

A generative intellectual object would attempt something analogous.

I might still generate one particular instantiation — perhaps even something resembling a conventional book — to introduce the idea to the world.

But that wouldn’t really be the product.

The product would be the generative core.

Interested?

Grab it.

Put it into your AI.

Tell it who you are, what you know, what you care about and what you’re trying to accomplish. Have it grow the idea into whatever form best allows you to understand, interrogate or use it.

For someone it becomes a book.

For someone else, a five-page explanation.

For another, an undergraduate course.

For another, an adversarial dialogue.

For another, a documentary.

For another, an interactive simulation.

For another, a two-hour conversation.

And eventually the same principle will apply to stories, movies, television series, games and creative forms we haven’t invented yet.

This would mean the decline of the fixed final instantiation.

But hardly the decline of creation.

Quite the opposite.

It would put greater emphasis on identifying exactly where the genuine creative contribution resides — what the actual idea is, what belongs to it, what is essential, what may vary, and where its boundaries lie.

And surrounding every worthwhile idea could be an effectively unlimited cloud of instantiations, varying in form, length, depth, style and function, generated anew for every person and every purpose.

AI may make it possible for creators to finally complete almost any vision they can imagine.

And precisely because of that, we may stop completing them.

Instead, we’ll be incentivized to create just the generative core of the project — the canonical specification of the vision.

Those interested in it will then have their own AI create a bespoke instantiation, tailored to their interests, needs, and level.

Historically, it’s been:

idea → instantiation → client
(discovery → book → reader)

AI will increasingly make it:

idea → client → bespoke instantiation
(discovery → reader → book tailored to that reader)

But is that actually true?

One way to approach the question is to separate authoritarianism itself from the narrative used to justify it. Some regimes explicitly ground their legitimacy in religion. Others justify authoritarian rule through nationalism, socialism, anti-communism, revolution, race, military necessity, or other secular narratives. If our interest is understanding the political tendencies of civilizations, both kinds matter. A population that repeatedly falls under authoritarian rule is exhibiting an important historical pattern regardless of whether that rule is wrapped in religious or secular language.

And, in my view, these narratives are often largely post hoc rationalizations: political vocabularies that evolve to legitimize power, hierarchy, repression, and group loyalty after the underlying authoritarian dynamics are already in motion. Religion may supply that vocabulary. So may nationalism, socialism, race, revolution, or security.

The table above surveys selected authoritarian regimes since 1800 across the Muslim world and Christendom, classifying each according to its dominant legitimating narrative: religious, mixed religious–secular, or secular.

The broad pattern is striking. Both civilizations have produced numerous religiously justified authoritarian regimes, numerous secular authoritarian regimes, and many hybrids. In both worlds, secular or mixed narratives account for at least as many examples as overtly religious ones.

None of this is to deny that much of the Muslim world faces greater authoritarian problems today. It plainly does. But much of the non-Western world also remains economically and institutionally behind the leading Western countries—perhaps by something like half a century in some respects. And half a century ago, many countries now treated as securely liberal and democratic were themselves emerging from fascism, military rule, dictatorship, political violence, or profound institutional instability. Latin America and Eastern Europe, though part of the broader Christian world, have likewise lagged behind Western Europe and North America and have experienced much more authoritarianism. A developmental gap is therefore easily misread as a civilizational essence: societies are behind, and the explanation offered is simply “Islam.”

The larger lesson is that authoritarianism appears to be a recurring political phenomenon that different civilizations justify using different vocabularies. Religion is one possible narrative. Secular ideologies are another. Neither Christianity nor Islam, by itself, adequately explains why societies repeatedly fall into authoritarian rule.

As I’ve written before, there’s something deeply backwards about the word “Zionist.”

Israel EXISTS. It has existed for generations — for 78 years. Yet we have a special ideological label for someone who thinks Jews should be allowed to maintain a nation in their indigenous homeland.

Why does that need a name?

We don’t have special words for people who think Egypt should continue to exist as a homeland for Egyptians, or Iran for Iranians, or Turkey for Turks. Their continued national existence is simply the default.

But give the Jewish version a special name — “Zionism” — and suddenly Jewish national existence sounds like a HYPOTHESIS. A political proposition. Something requiring argument and defense.

And with that little linguistic trick, the burden gets placed on exactly the wrong people.

“Are you a Zionist?”

“Why do you believe Israel has a right to exist?”

“Why should Jews get their own country?”

No.

Israel already exists. Jews already have their nation. They have had it for generations. Jews should no more have to continually justify the existence of their country than other peoples do.

If anyone needs a special ideological label, it’s the people advancing the extraordinary proposition: that this particular nation should be dismantled, and this particular people should once again be rendered stateless.

And there’s a rather important historical fact here.

The Jewish state was born in the immediate aftermath of the Holocaust and became home to huge numbers of Jewish refugees and survivors. It meant that never again would a persecuted Jew’s survival depend entirely upon whether some other country was willing to open its doors.

And ever since, there have been people demanding that this refuge cease to exist.

So maybe THAT position deserves the special word.

I propose:

Holocauster.

Not because every person who wants Israel dismantled consciously wants another Holocaust.

But because they advocate stripping Jews of precisely what the Holocaust demonstrated, with almost unimaginable horror, that Jews needed: a nation and refuge of their own.

Let THEM carry the label.

Let THEM defend the proposition.

The Jews are no longer on trial for wanting to exist.

The Universe Without Arbitrariness

One of the deepest aspirations in science is not merely to explain things, but to explain them from as little as possible.

Usually, this means grounding some important phenomenon in more fundamental premises. If those premises seem reasonable—or at least probably true—then the theory has done its job. It is not necessarily the theory’s problem to explain why those premises themselves hold. Newton could explain Kepler’s laws of planetary motion from deeper laws of motion and gravity. That was already a monumental achievement, even if Newton’s own laws were left standing as axioms.

Physics is filled with triumphs of this kind. Peculiar, apparently unrelated phenomena turn out to follow from a small number of deeper principles. Planetary orbits follow from gravitation. Light, electricity, and magnetism become aspects of one electromagnetic framework. Vast domains of the world that once seemed disconnected collapse into a compact mathematical structure.

Even more satisfying are those cases in which a phenomenon seems to emerge from almost nothing at all—from mathematics applied to a more basic description. Thermodynamics once appeared to require its own fundamental laws. Much of it can now be understood through statistical mechanics. Temperature, entropy, and the familiar directionality of macroscopic processes arise from the combinatorics of enormous numbers of microscopic states. In an important sense, the higher-level regularities are not extra ingredients. They are what almost inevitably happen.

That is what one would ideally like to do for the whole universe.

One would love to discover that Lorentz invariance, general relativity, quantum mechanics, and even the peculiar (U(1)\times SU(2)\times SU(3)) symmetry structure of the Standard Model simply had to be that way. The fundamental laws would not be substantive facts about our particular universe. They would fall out of mathematics itself. Reality would have no room to be otherwise.

But it increasingly does not look as though this hope will pan out.

Whatever final unified laws physicists eventually reach are unlikely to be nothing but mathematics. They will probably contain substantive structure: particular symmetries, particular fields, particular constants, particular rules. They may be extraordinarily elegant and extraordinarily deep, but they will still seem to describe one way a universe could be organized among other conceivable ways.

Many physicists claim not to care. The universe is what it is. There is no teleology in physics. Reality does not owe us an explanation that makes its laws inevitable.

And perhaps the demand itself is incoherent. There is no such thing as an axiom that justifies itself. Once one reaches the bottom, what could possibly lie beneath it?

Yet anything short of that leaves us destined to have no answer to the child who asks, after all the equations have been written down and all the forces unified,

“Yeah, but why is the universe set up that way?”

If the answer is simply, “Those are the laws,” then centuries of physics may have relocated the mystery without dissolving it.

This longing has a long philosophical pedigree. It is closely related to the Principle of Sufficient Reason: the intuition that there ought to be some reason why reality is this way rather than another. Whether or not one accepts the principle in its strongest form, it captures something fundamental about explanation itself. We do not merely want to know what the facts are. We want to know why those facts obtain rather than others.

I do not have an answer to this dilemma. But there is a remarkably similar problem in another foundational area: the Problem of Induction.

Why are we justified in believing some things rather than others?

The trouble is that what one ought to believe is never uniquely determined by the evidence alone. Suppose thousands of observed points lie perfectly on an upward-sloping straight line. The natural conclusion is that the underlying function is that straight line.

But infinitely many curving functions could pass through every one of those points. None would contradict any observation made so far.

The evidence therefore never uniquely determines the “smart” hypothesis.

What one concludes depends, in a nutshell, on one’s prior.

If you and I begin with similar priors, then after seeing the same observations we will probably agree that the straight line is overwhelmingly more plausible than the fantastically contorted alternatives.

But a prior is what one believes before the evidence. At the ultimate limit, it is the probability assigned to every possible hypothesis before having learned anything at all about the world.

How could that possibly be justified?

By definition, the prior was not obtained by checking reality. You and I could begin with arbitrarily different priors, encounter exactly the same data, and end up believing radically different things.

The similarity to physics is striking. We are unsatisfied if the character of the universe rests on peculiar laws for which no explanation is possible. And we are equally unsatisfied if everything we are justified in believing ultimately rests on peculiar probabilities assigned before any evidence was available. In both cases, the explanatory chain appears to terminate in pure presumption.

Thirty years ago, Tim Barber and I proposed what we came to call the Paradigm Theory of Induction. The basic idea was to go one level below the prior.

What lies beneath a prior need not be another probability assignment. It can instead be a conceptual framework—a paradigm.

Before one has any data about which hypotheses are probable, one can still notice properties and relations among hypotheses. One may notice that some functions are simpler than others. One may distinguish polynomials from non-polynomials, smooth curves from discontinuous ones, low-degree expressions from high-degree ones, or hypotheses possessing one symmetry rather than another. None of this yet favors any hypothesis. The paradigm merely notices.

But noticing some distinctions inevitably means failing to notice others. Every conceptual framework has limited resolution. It may distinguish one broad family of hypotheses from another while leaving many hypotheses within each family indistinguishable.

The result is a collection of hypothesis clumps. The hypotheses inside a clump are genuinely different, but not as far as the paradigm is concerned. The conceptual framework has no resources for expressing their differences.

Once those clumps exist, two simple rationality principles do almost all the remaining work. Each clump should receive equal probability in the absence of any recognized reason to favor one over another. And within each clump, each hypothesis should receive equal probability, because the paradigm cannot articulate a distinction among them that would justify unequal treatment.

This is simply Laplace’s Principle of Indifference applied at two levels.

The prior then follows from the paradigm.

The foundation still has no empirical foundation. It cannot. But it is no longer a naked probability assignment. One can instead say,

“I see the space of possibilities like this.”

And that alone is enough to determine the empirical assumptions implicitly being made.

Have we merely pushed the arbitrariness down one level?

Not quite.

We possess genuine intuitions about sensible and nonsensible ways to carve a space at its joints. A paradigm can be natural, mathematically motivated, and structurally simple—or it can be an ill-gerrymandered, grue-like construction designed solely to force a predetermined conclusion.

As Tim and I showed, so long as the paradigm is not pathological in this way, different reasonable paradigms tend to generate similar priors. There simply is not enough freedom in ordinary ways of conceptualizing hypotheses to justify arbitrary empirical commitments.

Someone can still ask, “What justified you in conceptualizing the hypotheses that way?”

The answer is not that the paradigm itself was empirically verified. It is that nearby, non-gerrymandered paradigms tend to lead to essentially the same place. To obtain radically different priors, one must begin conceptualizing hypotheses in ways that no mathematician would regard as natural.

We had done something that initially seemed impossible: provided a kind of conceptual justification for an empirical foundation.

Now return to physics.

Suppose there exists a space of possible universes—or equivalently, a space of possible fundamental laws. Suppose further that there exists a natural way of conceptualizing that space: a kind of physics paradigm. Like the paradigms in induction, it notices properties and relationships among possibilities, but not every conceivable distinction.

The possible universes are thereby partitioned into symmetry classes.

Imagine there are only five possible universes.

Under the natural conceptualization, one universe stands entirely alone, while the other four are symmetric with one another. As far as the paradigm is concerned, there is nothing principled that distinguishes any member of that four-element class.

Now suppose the singleton universe is the actual one.

That would hardly seem accidental. Had one of the other four been actual, an immediate question would arise: Why that one rather than one of its three indistinguishable companions? Making one member of a perfectly symmetric class actual would require an arbitrary choice.

The singleton universe would require no such choice.

It would be the only universe that could be actual without an arbitrary selection from among equals.

Now imagine infinitely many possible universes instead.

Suppose that under every natural, non-gerrymandered conceptualization, almost every possible universe belongs to some larger symmetry class, while exactly one universe always occupies a singleton class.

If our universe turned out to be that singleton, no one in their right mind would dismiss the fact as a coincidence.

And here is the crucial point: the universe would not be logically necessary. Other universes would remain perfectly conceivable.

What we would have instead is contingency without arbitrariness.

Our universe could have been otherwise in the ordinary modal sense. Yet every alternative would require reality to distinguish arbitrarily among possibilities that, at the deepest natural level of description, have no principled distinction among them. Our universe alone would require no such unexplained symmetry-breaking.

The Principle of Sufficient Reason could therefore be satisfied in a subtler way than philosophers have usually imagined.

The sufficient reason would not be that every other universe was impossible.

It would be that every other universe required an arbitrary selection from among conceptual equals, whereas this universe alone did not.

Perhaps, then, the deepest laws of physics need not be derivable from nothing in the strict sense. Perhaps they need not be unavoidable theorems of pure mathematics, nor self-justifying axioms.

They need only pick out the one universe that requires no arbitrary choice.

Perhaps the final laws of physics will define a universe occupying a unique position under every natural way of carving the space of possible universes. The alternatives would remain mathematically conceivable, but only as members of symmetry classes whose internal differences possess no principled significance. For any one of them to be actual would require reality to break symmetry for no reason at all.

Our universe would be different.

The universe need not be the only possible universe. It need only be the only non-arbitrary one.

—

Our Paradigm Theory:

https://www.changizi.com/uploads/8/3/4/4/83445868/changizibrain25000chapter3.pdf