为什么历史上的每一个大脑隐喻都错了【特别版】

Let me tell you a little story. In the summer of the 1960s, a little kid named Carl was playing around in the back of his garden, and he noticed all of these woodlice crawling around—you know, the little insects that can curl up into a ball. What he noticed was that, depending on whether they were in the sun or in the shade, they would move faster or slower. They behaved differently. And that's it.

Carl grew up to be Professor Carl Friston, one of the most cited neuroscientists alive. He's been on this channel before, more times than I can count. That childhood observation about woodlice never left him. He spent decades developing what he calls the free energy principle, which tries to explain all of behavior with one equation: perception, action, learning, why you scratch your nose—all of it, Friston claims, comes down to minimizing a single mathematical quantity.

There's an old physics joke: assume that we can model a spherical cow in a vacuum. The joke is about how scientists grotesquely simplify messy reality to tame it. The free energy principle might be the ultimate spherical cow. It promises to explain self-organization, this bewilderingly complicated phenomenon, with something so emaciated we might as well call it tautological. Even Friston himself agrees with this, by the way. This is what he said to us last time we spoke with him.

Carl Friston

The free energy principle is not meant to be complicated or difficult to understand. It's actually almost logically simple. The whole free energy principle is basically a principle of least action pertaining to density dynamics—the dynamics or the evolution of not densities, but conditional densities. That's just it. This is before thermodynamics. It's before quantum mechanics. It's just about conditional probability distributions.

So what do we do with this? Has Friston actually found some deep truth about how minds work? Or is he doing what many scientists do, which is mistaking the simplification for the actual thing?

It turns out there's a philosopher who has spent an incredible amount of time thinking about this exact problem. Professor Marta Chirimuuta teaches at Edinburgh University. Her book, The Brain Abstracted, is basically about what happens when neuroscientists simplify brains to study them. What gets captured? What gets lost?

Marta Chirimuuta

One of the answers that might seem obvious to people is that we pursue science because we're curious. We just want to know how the world works. We want to reveal and discover the underlying principles of the universe, which apply in all cases.

But switching off the idea that you're just interested in nature for its own sake out of curiosity and saying, “Okay, how can we engineer these systems to actually do things that we want?”—getting them to behave in artificial ways—if those simplifications allow you to achieve your technological goals, there's no in-principle problem with oversimplification. If you're going to say, “I'm not just interested in nature for its own sake. I just want applied science.”

I should say, by the way, that The Brain Abstracted probably influenced my thinking more in 2025 than anything else. She's an inspirational lady. I look up to her very much, and certainly, thinking back on many of the episodes we've done in 2025, I can see her influence in the questions I ask and how I think about things.

So, here's her starting point. Scientists have to simplify. We're limited creatures trying to wrap our heads around systems way more complex than we can actually comprehend. Our working memory holds maybe 7 items. Our attention is more scattered than a group of toddlers with iPads. We die after 80 years if we're lucky, so we build models, right? We leave stuff out on purpose. We tell ourselves stories about how the world works.

But the question is, why does any of this even work at all?

Marta Halina

Science is a humanistic endeavor, right? The purpose of science in the universe is to make the universe intelligible to us, not to control it, not to predict it, and not to exploit it. Now, you can do all those wonderful things if you like, but in the end, as far as I'm concerned, science is no different from poetry in that we're trying to make sense of the world, trying to give it meaning in relation to our own existence.

If you'll allow the indulgence, I want to tell a little story. It's a boxing match. In the red corner: Simplicius. He thinks science works because the universe is actually simple underneath. Find an elegant equation and you've hit the real thing. Simplicity tells you that you're on the right track.

In the blue corner: Ignorantio. He thinks we simplify because we're too dumb to do otherwise. Our models work well enough for our purposes, but they're approximations—just useful fictions, if you like. The map, not the territory. Now, both of them agree that scientists need to simplify, but where they disagree is what that means about reality.

Simplicius had history on his side, or at least a certain type of history. Galileo, Newton, Einstein—they all believed pretty explicitly that nature was fundamentally orderly and that finding simple laws meant you'd found something true. Einstein famously said, “God doesn't play dice.” And no, he didn't actually think God had anything to do with it, but he was expressing faith that the universe is, at the very bottom, legible.

Now, Chirimuuta has gone all-in on Ignorantio's position. She thinks successful science tells us we've become good at building useful simplifications, and that doesn't prove that nature is simple. The philosopher Nicholas of Cusa had a phrase for this attitude: docta ignorantia. Basically, learned ignorance. You study hard, you learn a lot, and what you learn includes what you don't know.

When we interviewed Chirimuuta, she had been following François Chollet's videos. For those of you who don't know, François is a friend of the channel. He's our mascot. He's one of my heroes. He's got this idea called the kaleidoscope hypothesis, which is basically that the universe is made out of code. Underneath all of the apparent gnarly mess that we see, there is intrinsic underlying structure.

François Chollet

Everyone knows what a kaleidoscope is, right? It's like this cardboard tube with a few bits of colored glass in it. These few bits of original information get mirrored and repeated and transformed, and they create this tremendous richness of complex patterns. It's beautiful.

The kaleidoscope hypothesis is this idea that the world in general and any domain in particular follow the same structure: They appear on the surface to be extremely rich and complex and infinitely novel with every passing moment. But in reality, they are made from the repetition and composition of just a few atoms of meaning.

A big part of intelligence is the process of mining your experience of the world to identify bits that are repeated and to extract these unique atoms of meaning. When we extract them, we call them abstractions.

Now, she's not saying that Chollet is wrong. She's saying that he's making a philosophical bet. It might be right, it might be wrong. It's the same bet that Plato made.

Marta Halina

Seeing that, as a philosopher, I thought, “That's Plato,” because François precisely says we have the world of appearance. It's complicated. It looks intractable. It's messy. But underlying that real reality is something neat and mathematically decomposable.

Now, I feel like I should defend Chollet a little bit here, because obviously we love Chollet. He's not making any weird metaphysical claims—at least I don't think he is. If scientific theories actually explained reality the way it is, you would expect fewer U-turns.

Now, the biggest simplification in the 21st century, the final boss of simplifications, is this idea that the mind is a computer, or that the mind is running a software program. So, we have inputs, we have processing, and we have an output. This metaphor has become so established in the collective zeitgeist that no one even questions it anymore. It barely even registers in our brains as a metaphor.

So, is it or isn't it a little bit weird that computation is this abstract formalism—an automaton that makes these state transitions, something completely nonphysical—and we're describing the mind as if it is that abstract thing? That sounds a little bit weird. There are many movies made about this that talk about uploading their minds into the Matrix. Neuralink talks about interfacing with your brain's software. Joscha Bach thinks that consciousness is a software program running on your brain.

Joscha Bach

The idea is that this is the universal: You have these invariances in nature, patterns that have causal power, that have the ability to reproduce themselves, that have the ability to shape reality. They are invariances that you cannot simply explain more simply by looking at what atoms are doing in space. You have to look at these abstract patterns to make sense of them.

Every other explanation is going to be more complicated, in the same way as money is going to be impossibly complicated if you try to reduce it to atoms. So, you have to look at these causal invariances, and spirits are actually such causal invariances. They are actually disembodied, right? They're not bodies. They're not stuff in space. They're not mechanisms in the same way, but they are causal mechanisms—abstract mechanisms.

And so we put the spirit back into nature using the concept of software. A lot of people think that's metaphorical, but I don't think it's metaphorical at all. It's the literal truth. Software is spirit.

We're all just talking about this stuff without even batting an eyelid. Where's the skepticism, man? It just sounds so plausible to us, so we assume that it just has to be the case. There is something super interesting about computers.

What a computer ultimately is is a causal insulator. The computer is a layer on which you can produce an arbitrary reality—for instance, the world of Minecraft. You can walk around in the world of Minecraft, and it’s running very well on a Mac and on a PC. If you’re inside that world, you don’t know what you’re running on, right? It’s not going to have any information about the nature of the CPU it’s running on, the color of the computer’s casing, the voltage the computer is running at, or the place where the computer is standing in the parent universe—our universe.

The computer is insulating this world of Minecraft from our world. It makes it possible for an arbitrary world to happen inside this box. Our brain is also such a causal insulator. It’s possible for us to have thoughts that are independent of what happens around us. We can envision a future that is not much tainted by the present. We can remember a past that is independent from the present in which we are, and that’s necessary for us. Our brain has evolved as such a causal insulator to allow us to give rise to universes that are different from this one—for instance, future worlds, so we can plan for being in them.

Joscha Bach says that money is an example of a causal pattern. It’s not the ink on a banknote. It’s not the electrons in your bank server. It persists across various physical instantiations: paper, coins, gold, digital ledgers. And yet they say money causally affects the world. It gets you fed. It starts wars. It builds cities.

He says that software is the same. A program is an abstract pattern that can run on many types of chips, maybe even neurons. That pattern has causal power because it controls whatever substrate it’s running on. The same algorithm produces the same effects regardless of what physical stuff implements it. So, the invariance—that sameness across substrates—is the causal mechanism, the pattern itself, at least according to Joscha.

He even accepts that physics is causally closed. He says that the abstract description and the physical description are 2 ways of looking at the same causal structure. Neither is reducible to the other. Both are real.

But I’m pretty sure Chirimuuta would ask who identifies that invariance when we say the same algorithm runs on different chips. Completely different things are actually physically happening, right? Different voltages, different electrons doing different things. The sameness is something that we impose. It exists in our description, not in nature.

As for the money example, money only works because of human interpretive practices. If you take away the humans and their agreements, it’s just paper, right? Money is just paper, and the causal power is actually in the social substrate that participates in it.

Now, I think Joscha has taken a useful way of talking about complex systems and promoted it to metaphysics. And that’s Simplicius all over again, right? Mistaking the elegance of our descriptions for the structure of reality itself. Maybe information really is more fundamental than matter, but that’s another philosophical wager. We’ve made these bets many, many times before. Just look at the history of all of this.

Descartes thought that the nervous system worked like the hydraulic automata in French royal gardens: fluids pumping through tubes, pushing levers. That was the high-tech metaphor of his day. Later, when scientists figured out that nerves carry electrical signals, the brain became a telegraph network. Then it was a telephone switchboard, with signals traveling down wires and operators routing calls. Now, in our era, the brain is a computer.

To be precise about what we mean by physical, everything has to be physical, because even GitHub has to store its data on some sort of hard drive or in a magnetic field, or whatever technology it uses. It’s not storing it in nothingness. Knowledge and information always have this form of physical embodiment.

I think we tend to think about it as nonphysical because it is a thing that is not a thing, which is the same as temperature. You wake up, you look at your phone, and you see the temperature and decide how you’re going to dress. Nobody has any doubt that temperature is something that can be measured. But it took about 2,000 years for us, as a species, to figure out what temperature was and the fact that it could be measured.

There were 2 fundamental difficulties that made it difficult for us to understand temperature. The first is that people initially thought hot and cold were 2 separate things, so that temperature was a mixture of the 2. It’s like when you make green out of blue and yellow. It took a while for people to understand that cold was the absence of heat, and not that cold and heat were 2 different quantities that were tempered together or mixed. So, temperature actually means mixture—not what we now mean by temperature.

The other thing that was very difficult to understand is that people thought temperature was a thing, some sort of fluid that grabbed onto things. Let’s say you had a steel rod that was hot. They thought that the steel rod had this invisible fluid, heat, and they had good reasons to believe that it was an invisible fluid because it could flow. You could connect that rod to something that was cold, and that cold thing was going to warm up because the fluid was going to flow in that direction, and so forth. They thought that it had a physicality as a thing.

A brilliant Englishman, Joule, figured out that this was not the case. They did it through an observation involving how cannons used to be built. If you grab a piece of sheet metal, make it into a cylinder, and try to make a cannon out of it, the moment you shoot the cannon, it’s going to open up like a flower in a cartoon—a Looney Tunes type of situation.

So, what they would do is make these solid cylinders of metal and bore a hole in them to create the cannons. Boring those holes released an enormous amount of heat. Joule thought, “How come all of that heat is there? It’s like an infinite amount of heat. If I continue to bore a hole in a piece of metal for an infinite amount of time, I’m going to—” It cannot be a thing, then.

That leads him to realize that temperature is actually something that has to live in things, but it’s not a thing itself. It’s related to the kinetic energy of the particles in the thing, but it’s not a thing itself. It doesn’t have its own particle. There isn’t a temperature particle. Temperature is a property that matter has. It holds on to things.

Knowledge is similar, in that it holds on to you, to me, and to the collective in order to exist, but it doesn’t have a physicality in itself. It always exists in some sort of physical medium or substrate. In that sense, it’s always going to be physical. No matter how virtual it gets, it has maybe a different type of physicality. Even electromagnetic waves that are transmitting data from your Wi-Fi router to your laptop are technically a physical embodiment.

Now, I spoke with Professor Luciano Floridi a few years ago, and it was actually one of my favorite-ever episodes of MLST. I think very highly of him, which is why we’re going to show some clips of him in this show, because it’s very apropos. This is what he had to say about it.

Luciano Floridi

Ontology, on the other hand, is how we structure the world, in the sense that we think that that’s the way it is. With the kind of eyes we have and the kind of light around the world, those are the colors we perceive. But certainly, a world full of colors is the world which I take to be the world. That’s my ontology.

Reontologizing means changing some of that particular nature. Allow me a distinction. I hope it’s not too confusing. Reality in itself—call it the system—is one thing. The description of reality as we perceive it, enjoy it, conceptualize it, and live through is the model of the system. Ontology, to me, is the ontology of the model; it’s not the metaphysics of the system. I hope I haven’t made a complete mess here.

Okay. So, metaphysics, no? The system, whatever the source of the data that we get—fantastic—the data don’t speak about the source. The music of the radio is not about the radio, but there is a radio, of course. The music is what we perceive. The music has its own ontology, structure, et cetera. The model is, at that point, what we enjoy.

Why? The digital revolution has changed the nature of the world around us, not metaphysically but ontologically. So, reontologizing, because some of the things that we have inherited from modernity—a sense of the world that is now being restructured, and a certain understanding of the world. So, re-epistemologizing that world as well.

We go back to this temptation of talking about reality as if it were something that we need to grasp, catch, portray, hook, or spear, when in fact the way I prefer to understand it is as malleable and understandable in a variety of ways—something that provides constraints. It doesn’t mean that you can interpret it in any possible way, but it leaves room for different kinds of interpretations.

So, if the flow of data that comes from whatever is out there—and again, I’d rather be sort of agnostic about it—can be modeled in a variety of ways, one way is to, especially in the 21st century, given the technology we have, et cetera, interpret that as an enormous computational kind of environment. It’s perfectly fine, as long as we don’t think that there is a right metaphysics, the correct ontology for the 21st century.

This is not relativism because, on the other hand, different models of the same system are comparable, depending on why you're developing that particular model. Let me give you a completely trivial example. Suppose you ask me whether that building is the same building. That question has no real answer because it depends on why you're asking that question.

If you're asking because you want directions, I'm going to say, “Oh, yeah, that's the same building.” So, the same building? Yeah, absolutely not. Go there, turn left. No traffic lights. But if your question is, “Is it the same function?” I know it's a completely different building. It was a school; now it's a hospital.

Next question: Is it or is it not the same? That question is the mistake—an absolute question that provides no interface, what computer scientists call a level of abstraction, chosen for one particular purpose, so that I can compare whether an answer is better than another.

Let me crack a joke for the philosophers who might be listening. Is it the same, or is it not the same? Who is asking? Why? If it is the tax man, the tax man, you're doomed, man. I mean, there is no way you can play any game. “Oh, I changed every plank.” You're going to pay their tax. It's the same ship. I don't care.

But if it is a collector, that ship is worth zero. You changed all the planks? You must be joking. It's worthless. So, is it or is it not the same? It depends on why you're asking that particular question. Tell me why, and I can give you the answer.

No. Why? In other words, no frame within which we have chosen the interface that provides the model of the system, no potential answer. So, the question, “Is the universe a gigantic computer? Yes or no?” is meaningless. Is it worth modeling the universe as a gigantic computer for the purpose of making sense of our digital life? Oh, yes, definitely, because we are informational organisms.

Aha. So, metaphysics? No, I meant, in the 21st century, the best way of understanding human beings today is as informational organisms. Last century, we thought that biologically it made much more sense: a lot of water and a little sprinkle of extra, and so on. Mechanism, time, et cetera. Not absolute answers, not relativistic answers, but relational answers—the relation between the question, the purpose, and the actual answer. But it takes 3, not 2.

The computational model isn't literally true, but it's useful. The mistake is forgetting that it's a model. So, the early cybernetics guys—McCulloch and Pitts—they knew that they were working with analogies. McCulloch and Pitts wrote their famous paper showing that neurons could theoretically work like logic gates. They weren't claiming neurons actually were logic gates, but they were using it as a kind of functional description.

Now, Chirimuuta, borrowing from Whitehead, by the way, said that this is the fallacy of misplaced concreteness. This is another one of those leaky abstractions I was talking about. By the way, there's a great book called The Brain Abstracted by Marta Chirimuuta. I interviewed her recently, and she said that one of the most pervasive myths in neuroscience is that we use these leaky abstractions and idealizations to talk about cognition, usually using the most recent technology at the time.

So, a few hundred years ago, we were describing the brain in terms of pulley.

Marta Chirimuuta

Pulleys and levers. Yes, that's right. And then it was, as a prediction machine, as a computer, and all this kind of stuff.

At the end of this, these are grounded things that we understand. They're really good models because we can both talk about computers. We both know what computers are, but the brain doesn't work like that in any sense. Jeff Bowers put it even more bluntly when we spoke: It will always be the case that our explanation for how the brain works will be by analogy to the most sophisticated technology that we have.

Is that how's that for a non-answer? Right. So, a couple of thousand years ago, how did the brain work? It was like levers and pulleys, man. I mean, duh. Don't be ridiculous. Why? That was, you know, at some point in the Middle Ages, it became humors, right? Because fluid dynamics was the technology that was the most advanced, or technology that took advantage of water power was the most advanced technology that we had. Now, the most advanced technology is computers. So, duh, that's exactly how the brain works.

Now, here's something that kind of bugs me, right? You go into any AI conference, or you drink from the well of San Francisco by spending too much time on Twitter, and you develop this mindset that AGI is inevitable. You start feeling the AGI, and you'd be forgiven for thinking this because I've been using Claude Code, and my God, I feel that there's been more interesting stuff happening in the world of software development in the last 6 months than there has been in the previous 20 years.

This technology is genuinely amazing, but it is automation technology. It's not really intelligence, which means it's only really as good as your ability to specify, supervise, and delegate to the system. But it is absolutely amazing. Why do we have this view? It's not an argument that AI is impossible so much as, why does it seem so possible, so inevitable, to people?

What I'm arguing is that if you look at the history of the development of the life sciences and psychology, there are certain shifts toward a much more mechanistic understanding of both what life is and what the mind is. Those shifts are very congenial to thinking that whatever is going on in animals like us, in terms of the processes that lead to cognition, they're just mechanisms anyway. So, why couldn't you put them into an actual machine and have that actual machine do what we do?

With all that mechanistic history in the background, AI could seem very inevitable. But if that mechanistic hypothesis is actually wrong, then these claims for the inevitability of biological-like AI would not actually be well-founded. We could be subject to a kind of cultural-historical illusion that this is just going to happen.

Cultural-historical illusion. I've been thinking about that phrase. Maybe our confidence says more about what we've inherited intellectually than about how minds actually work. Another thing that Marta Chirimuuta has inspired me to think about a lot is the difference between prediction and understanding.

Indeed, when I interviewed the Nobel Prize winner John Jumper at Google DeepMind a couple of months ago, this was the question I asked, and he had quite an interesting way of distinguishing those 2 things. It's almost like, at any point, learning how to refine and optimize the structure.

John Jumper

Okay. So, I think we should distinguish 3 things: predict, control, and understand.

Predict means that you say, “I'm going to do a thing. What will be this value of my machine? What will appear on my computer screen in the future?” That is predict. Control is, “I want to measure this thing in the future, and I want it to come out 17.” Right? That's control.

Understand is a lot like predict, except there's a human in the loop. Understand means that I have such a small collection of facts that you can predict, and you can do it with facts that I can communicate to another human in a compact form that fits on an index card. That's almost understanding.

These machines let us predict. They let us control. We have to derive our own understanding at this moment. We can experiment now on the artifact. We can look at the 200 million predicted structures, not just the 200,000 experimental structures, in order to help us understand, but it doesn't do the act of understanding for us. It does the act of predicting and maybe controlling.

Marta Chirimuuta

The problem is these 2 goals actually pull against each other. I think we're at this moment in science now because we have these tools, like LLMs for language and ConvNets in visual neuroscience, being used as predictive models of neuronal responses, which don't have the mathematical legibility that people originally aspired to have when I was trained in the field.

So, you have this possible conflict: You can either pursue that goal of understanding, or you can pursue the goal of prediction. But it seems like you can't have both at the same time.

On the one hand, people go into neuroscience because they want to understand the mind. They want that feeling where something clicks and you suddenly get how it works. That's what drew Chirimuuta to the field in the first place. That's what keeps people up late at night reading papers.

But on the other hand, there's just prediction: building tools that work. If your model forecasts data accurately, maybe you don't care whether it's true in some deeper sense. So, LLMs are getting unreasonably good. They are winning math Olympiads. As of last week, actually, GPT-5.2 apparently discovered a new theorem—well, it solved one of these problems that Terence Tao had on his website.

This is insane, but does it actually understand anything? And does it matter if it does or doesn't, as long as it works? Chomsky had an amazing commentary on this a few years ago when we spoke, and I think it's still as relevant today as it was then.

Noam Chomsky

Suppose that I submitted an article to a physics journal saying, “I've got a fantastic new theory, and it accommodates all the laws of nature—the ones that are known, and the ones that have yet to be discovered. And it's such an elegant theory that I can say it in 2 words.”

“Anything goes.” Okay, that includes all the laws of nature—the ones we know, the ones we do not know yet, everything. What’s the problem? The problem is they’re not going to accept the paper. Because when you have a theory, there are 2 kinds of questions you have to ask: Why are things this way? Why are things not that way? If you don’t get the second question, you’ve done nothing. GPT-3 has done nothing.

Classic Chomsky. So maybe theories are overrated. Maybe prediction is enough. But Chirimuuta worries about that trade-off, right? When you give up on understanding, you don’t know when your tools will break. You’re stuck with black boxes. They work until they don’t, and you won’t see it coming when they don’t.

I spoke with philosopher Anna Ciaunica about this recently, and she had a beautiful way of describing it.

Anna Ciaunica

Suppose you want to climb a mountain and you arrive at the top of the mountain. What’s the argument to say that actually it’s only when you’re on the top of the mountain that the climbing on the mountain is? I mean, you cannot really arrive at the top of the mountain if you don’t do the first step. Every single step matters. The first step is as important as the last one.

Actually, we are more conscious when we take the first steps in climbing the mountains than when we are on the top of the mountains and we have all these full-blown capacities, and sometimes we shoot ourselves in the legs.

And of course, I brought this up when I debated Mike Israel. The biggest misconception in all of AI, what all of the folks in San Francisco believe in, is this philosophical idea called functionalism: that we’re walking up the mountain, and when we get to the top of the mountain, we have all of these abstract capabilities, like being able to reason and play chess. But that disregards that the path that you took walking up the mountain is very important, and not only the path—the physical instantiation, the stuff that the mountain is made out of.

So Mike’s view is that if something produces intelligent outputs, why does the substrate matter? Silicon neurons—it doesn’t make any difference. It’s all information processing. Needless to say, he pushed back hard. You can climb mountains. You can touch stuff. But you never truly have an embodied experience of anything if you push on that philosophical button hard enough, because you can always abstract out to, “These are just neural network pings from groups of neurons.” And so you don’t truly deeply know anything in some kind of weird philosophical way, because it’s just neural network calculus all the way down.

You know, you climb the mountain, that’s cool. A helicopter can climb the mountain much better than you. It does not have the ability to reason abstractly and plan and predict things at all.

So it’s possible that what you can do, or how you can function, isn’t the whole story. Or maybe, if that’s wrong, we should just start using helicopters. Individual minds are limited. But what about collective minds? What about humanity as a whole? We’ve built this incredible thing over centuries, right? Libraries, universities, Wikipedia, an expanding store of knowledge that no single person could ever hold. Doesn’t that escape our individual limitations?

So there’s this dream of universal knowledge, accessible anywhere, perspective-free.

M. J. Chirimuuta

There is a tacit and implicit idea there that knowledge is something that something can have, while my view is that knowledge is a much more collective phenomenon. It’s not something, also, that you can put in something like a book. In my opinion, the book doesn’t have knowledge. The book is an archival record of some ideas that I was able to put together in a nice structure. But you cannot have a conversation with the book. Knowledge only can go to work when it’s embodied.

You cannot throw a bunch of engineering manuals and cement into a gorge and expect to get a bridge, because the books don’t have knowledge.

Teams have knowledge. Organizations have knowledge. Yes, knowledge is social. Communities accomplish what individuals can’t. But collective knowledge is still knowledge from somewhere. This matters, right? It’s shaped by particular questions, particular tools, and particular blind spots.

M. J. Chirimuuta

I think one of the interesting things about this phenomenon—not only of LLMs, but of the internet as the repository of all human knowledge—is that it goes along with this idea, almost, that knowledge doesn’t have to be perspectival. It doesn’t have to be from a place, from a community. It kind of can float free of the situation in which this knowledge was acquired. That’s the aspiration of these ideas, of a universal repository of knowledge.

But what this perspectivalist position actually points us to is that knowledge is inherently of a place, of a community. We acquire knowledge not by being completely open-minded to everything that’s possible to know, but actually by narrowing our view. Discounting possibilities is what allows you to pursue a line of inquiry and actually pin down some information about, say, the natural world, which is humanly achievable.

So the contrast I’m trying to make here is between a view which says that knowledge is perspectival. It’s inherently from a human point of view, which means that it’s inherently finite. We cannot aspire to this sort of universal, free-floating knowledge because, as finite human beings, we can only achieve knowledge of the world through recognizing our limitations.

And this notion that you can have non-perspectival knowledge—like everything on the internet, based on all of the different possible perspectives all blended together—that this somehow gives us a god’s-eye view. LLMs aspire to be this every-person voice, but it’s precisely because they don’t have a particular socialization into a finite community that they’re not reliable, that we can’t pin them down to what would be a sort of honest, trustworthy perspective.

So Chirimuuta has this idea that she calls haptic realism. Most of the philosophy of science treats knowledge like vision. You stand back and observe reality from a distance. She thinks it’s more like touch.

M. J. Chirimuuta

We just look around. We absorb how things are. Our knowledge is sort of entirely objective. It’s almost like a god’s-eye view on reality. But if you think that scientific knowledge in particular is more touch-like, you can’t ignore the fact that we run into things. We have to pick things up, engage with them, ultimately change them in order for us to acquire knowledge of them.

So you cannot discount the fact that we’re meddling with things in the process of bringing about our knowledge.

Neuroscientists are more than passive observers of brains. They poke them, prod them, stimulate them, model them, and in doing that, they change what they find. The patterns that emerge are real, but they’re also partially created by the process of investigating itself. It takes all the messiness of biological cognition and reduces it to one imperative: Minimize free energy. Everything else supposedly follows from that.

Now, Simplicius loves this. Finally, the simple truth, the one principle to explain it all. But Ignorantio says, “Wait a minute. The math is elegant. The framework is unified, but does that mean it has captured what brains actually are? Or did we just build another beautiful simplification and start forgetting that it was a simplification?”

So Chirimuuta said to me that we should ask different questions, right? Not “Is this true?” but “What does this help us do? What does this light up? What does it leave in the darkness?”

And the other thing, of course, is that we are finite biological creatures, right? There are limits to our cognition, and Chomsky spoke about this fascinating concept of a cognitive horizon when we chatted with him.

Noam Chomsky

If we are organic creatures, we’re going to be like other organic creatures, and there are bounds to our cognitive capacities. So, for example, a rat can be trained to run pretty complicated mazes, but it can’t be trained to learn a prime-number maze: Turn right at every prime number. It just doesn’t have the concept. And no matter how much training you do, you’re not going to get anywhere.

Well, I suspect there are reasons to suppose we’re like rats. We have capacities. We have a nature. We have a structure. They yield an extensive range of things that we can do, but they probably impose limits. And I think we could even make some guess about what these limits are.

So our best theories bump up against the walls of the limits of our cognition, of our cognitive horizon. And maybe that’s fine. But maybe even knowledge of where the walls are is useful in and of itself. Science makes things simple, and it’s not a flaw, right? Without simplification, we’d have nothing. You can’t study everything at once.

But simplification has risks, right? You forget your model is a model. You mistake elegance for truth. And you think you found solid ground when really you’re just building another floor.

So look at Claude Opus 4.5, right? Foundation models today are artifacts of staggering complexity. We’ve trained them on everything humans have ever written. We treat their outputs like they came from somewhere authoritative, somewhere outside of us, somewhere that knows, but the knowing was ours all along, right? Just compressed, refracted, reflected back to us from the silicon.

Whether that reflection captures the actual thing, that is a question that we’re barely starting to ask. You can use powerful frameworks like the free energy principle, but just remember, they’re frameworks, right? They’re tools for building. They’re not the final word.

So the brain is not a hydraulic pump. It’s not a computer. It’s not a telephone network.

It’s probably not a free-energy minimizer either, at least not in some literal way. What the brain actually is, we will only ever catch glimpses of through our limited instruments and theories, right? And that’s okay, because that’s what it means to be finite.

So Chirimuuta had this amazing example from Greek mythology called Proteus, right? If you could pin him down, he’d have to answer your question correctly. But if you let go and let him get away, then he would shapeshift and shapeshift.

Nature is like that, right? You can pin it down and ask questions, but it’s always perspectival. As soon as you let go, there’s always a myriad of other perspectives that can be interpreted from reality. Carl Friston’s woodlice were doing something very similar, right? They slow down in the sun and move faster in the shade. But Friston isn’t a woodlouse, and neither are you.