At a glance
This is a 2 hour 43 minute conversation between Tom Bilyeu and the cognitive scientist Donald Hoffman about a single radical claim: space and time are not fundamental reality, they are a headset. Hoffman argues that what we call the physical world is an interface our senses render on the fly, that what is fundamental instead is consciousness, and that he has the mathematics to back it up. The case rests on two pillars he keeps returning to: the 2022 Nobel Prize in Physics, which confirmed that local realism is false, and his own theorem from evolutionary game theory that the probability our senses evolved to report objective truth is exactly zero. From there the two men build outward into conscious agents, Markov dynamics, Cantor's hierarchy of infinities, the halting problem, dark matter, free will, near death experiences, and what Hoffman thinks happens when you take the headset off. Bilyeu spends the whole interview pressing one alternative of his own: that you do not need consciousness to be fundamental at all, you only need a thirteen year old programmer and a set of rules.
The opening: is AI a threat, and can it be conscious
Bilyeu opens not on simulation but on artificial intelligence, because that is where his own thinking has been running. He asks what Hoffman makes of the AI scientists who signed letters asking to slow AI down, one of whom he had on the show, worried because the technology passed a Turing test faster than expected and is simply moving faster than anyone planned for.
Hoffman is not an alarmist. The thing that would actually alarm him, he says, is a law that criminalized AI for most people while letting a few companies do it. Any rule should be universal, with no one excluded. Beyond that he is unworried, because today's systems are not deeply intelligent. Even ChatGPT, he notes, is easy to stump. It is a very good statistical analyzer that will find patterns in medical searches that humans miss, but only because it can crunch more data than we can, not because it understands. The founders, he says, will tell you the underlying algorithms are fairly straightforward.
On consciousness he is categorical, and this becomes a load bearing claim for the whole conversation. There is no theory on the planet, of any kind, that can start with a physical system (a circuit, a software pattern) and explain even one specific conscious experience. Nothing can take a description of a pattern of activity and tell you that it must be the taste of chocolate and could not possibly be the smell of a rose. There is nothing on the table and nothing close. So if AI can be conscious, there is currently no theory that makes it even plausible. He grants that AI will eventually outperform humans at most everyday tasks, simply through more compute and deeper search, but raw performance is not the same as having an inner life.
He carves out one exception that he will defend all interview: AI could become a window into consciousness, the way a human child or a dog is, without ever creating consciousness. Bilyeu flags an early disagreement here, and Hoffman agrees they disagree, parking it for later.
The setup: space time itself is the simulation
Bilyeu offers a "super brief synopsis" of Hoffman's view and invites correction: this is all a simulation, none of it is real, space time itself is not real, we live inside what Hoffman calls the headset, and everything you have ever experienced is effectively an illusion, a computer video game by analogy. Hoffman does not say no.
Bilyeu confesses he has spent years dismissing the simulation idea out of hand, precisely because he had no evidence for it. He knows the standard argument from probability: if humans can build photorealistic simulations, and if progress continues at any rate, we will eventually build one. AI's pace and Apple Vision Pro have made that future feel concrete, and Bilyeu mentions that the company he is building is itself a simulated world where players will have an identity. Once a simulation gets powerful enough to spin up its own simulation, and that one spins up another, you land at the classic recursive conclusion: mathematically it makes more sense to bet you are in one of the nested simulations than in base reality. What always felt weird to Bilyeu was asserting that we are one of the simulations rather than the original. But the more he researched Hoffman, the more he started to think we really might be.
The hinge that separates Hoffman from the ordinary simulation argument is the role of consciousness. Bilyeu asks him to walk people through how everything they see, touch, taste, and love could be a simulation that arises out of something else, while consciousness is the fundamental foundation. That is the project of the rest of the interview.
Pillar one: the Nobel Prize and the death of local realism
Hoffman gives two independent arguments that what we see is not an objective reality existing prior to and independent of observation.
The first comes from physics. The December 2022 Nobel Prize went to three physicists, John Clauser, Alain Aspect, and Anton Zeilinger, for experimentally testing and confirming a clean prediction of quantum theory: that local realism is false. Hoffman defines the two halves carefully. Realism is the claim that a particle like an electron has a definite value of position, momentum, and spin even when it is not observed. Locality is the claim that those properties have influences that propagate through space time no faster than light. Local realism is the conjunction: properties exist when unmeasured, and their influences are bounded by the speed of light. That conjunction is false.
The proof runs through Bell inequalities. If two entangled particles had definite spin values even when unobserved, the statistics of the correlations between their measured spins would fall in one pattern. If quantum mechanics is right and those values do not exist until measured, the correlations fall in a different pattern. Experiment after experiment, tightening over decades, closed loophole after loophole, and the correlations always came out matching quantum theory, not classical intuition. Clauser did the early version; later experimenters used photons from distant galaxies so the particles could not possibly have shared a hidden causal connection within space time, closing one of the last loopholes. Zeilinger used photons from outer space to rule out that the particles were somehow secretly correlated.
Hoffman anticipates the escape hatch: maybe local realism fails only for tiny subatomic things, while macroscopic objects stay classical. Two problems. First, the effects have been pushed up to groups of roughly 700 atoms, so this is no longer only the very small. Second, and decisively, quantum theory draws no principled line between micro and macro. There is no length scale, no ten to the minus twenty centimeters, where the rules switch over. The boundary does not exist, which is a famous open problem. So local realism is false, full stop.
Pillar two: evolution did not evolve you to see the truth
The second argument comes from evolution by natural selection. The standard story is that sensory systems are shaped to guide adaptive behavior, keeping you alive long enough to reproduce. Most theorists take a further step and assume that guiding adaptive behavior means reporting the truth: when you see an apple, there is a real apple, and its red color and shape exist even when unperceived. Hoffman set out to test exactly that extra step.
Evolution is a mathematically precise theory, so the question is precise: what is the probability that natural selection would shape any sensory system to perceive the true structure of objective reality? With colleagues Chetan Prakash and Manish Singh and Robert Prentner, and graduate students Justin Mark and Brian Marion, he used evolutionary game theory to answer it. They considered different fitness payoff functions and asked whether those payoffs preserve structures the world might have, a total order, a partial order, a metric, a topology, a measurable structure. The result, case after case, is that the probability is zero. There do exist payoff functions that preserve the true structure, but they form a probability zero subset of all payoff functions. It does not mean truthful perception is impossible, only that if you were a betting man you would bet very long odds against it.
Hoffman treats this as a convergence of two of science's biggest theories. Evolution by natural selection and quantum field theory are both telling us that local realism is false, that we do not perceive a pre existing objective world. The metaphor that fits is a user interface or a video game where you render only what you need, on the fly. He is rendering a Tom face while he looks at Tom, and stops rendering it when he looks away; someone else renders their own Tom face at a different angle. We render on the fly because that is what physics says.
The Grand Theft Auto coordination problem
Bilyeu connects the rendering idea to the quantum uncertainty principle: a property sits as a big question mark until something looks at it, because the system does not need to render what nobody observes. He sketches a hypothesis Hoffman's data seems to support. Big things (the moon, planets, stars) need to be persistent, so there is constant collapsing of their probabilities, because so many things are effectively witnessing or measuring them, if only through gravity. Small things, where almost nothing hinges on the exact value, do not need direct rendering; you can get away with probabilistic rendering. Nobody is effectively looking at the spin of a particle, so it does not need to be rendered.
Hoffman reframes it with a multiplayer game. In Grand Theft Auto, he is playing with someone in Canada, someone in Europe, someone in China. He looks over and sees a red Porsche on his right and asks the others; each of them, when they look, also sees a red Porsche, each rendering their own. So there is some reality coordinating all these separate perceptions. The man in Canada did not see the Porsche until he looked, but when he looked there was a whole hidden world of circuits and software, a supercomputer taking inputs (which way each headset points, body suit arm movements) and feeding a shared model of the game. Crucially, there is no red Porsche inside the computer. The computer knows how to coordinate and send photons to each headset so that everyone shares a persistent reality of a Porsche, even though for each individual local realism is false and the Porsche does not exist until rendered.
Behind space time, Hoffman says, lies a realm at least as much more complicated than space time as the supercomputer is more complicated than the little headset. Science for the last three or four hundred years has only studied the headset. We learned the tools, experiments and clean mathematical theories and the loop between them, thinking we were studying objective reality, but we were studying our interface. Now we have the tools to take the first baby steps beyond space time and find structures behind it and how they project back. Our view of objects in space time as fundamental will look parochial in a few decades, especially to a next generation raised inside virtual reality.
Mapping the analogy back one level
Bilyeu pushes the metaphor deeper. In earlier interviews they spent a lot of time establishing the first level, your life is Grand Theft Auto and you have never existed outside the headset. Now he wants to map that one rung back. Take the idea "your life is GTA, but there is this thing called space time outside it," and shift it one profound level: the universe, space time itself, is an emergent phenomenon from consciousness. Consciousness is the computer and the rules; the simulation is what we all think of as real life. Hoffman accepts the mapping: a model that takes consciousness as fundamental, gives a mathematical description of it, and then shows how space time gets rendered from it. That is exactly his program.
Bilyeu sets a frame for the rest of the talk. He has reverted to the mean after past interviews with Hoffman, but each time his baseline drifts closer. This time, in the research, he genuinely felt "he's right," with consciousness being the one part he may still disagree on. He thinks of the human mind as a prediction engine, and the closer you get to baseline truth, the better you predict. Three things about AI had refused to make sense to him: hallucination, pulling patterns out of noise, and the difficulty of persistence. Researching Hoffman, the pieces fell into place once he assumed it is all already a simulation and AI is simply revealing how the simulation works. He flags stable diffusion as the tell: it dips into noise, finds a pattern, solidifies it, exactly as Hoffman's theory and Gödel's incompleteness theorem (an idea he says he struggled with in earlier interviews) would predict, an image pulled from an infinite possibility space.
Consciousness as fundamental, and Leibniz's monads
Now the hardest inversion. Everybody intuits that consciousness comes from stacking neurons: ant, mouse, cat, dog, dolphin, gorilla, human, just add more neurons and you get richer consciousness. Hoffman says no, and notes the idea is not new. Gottfried Leibniz, the genius contemporary and rival of Newton (they invented calculus around the same time and feuded over priority), held the same view. Leibniz had a famous argument, the mill, in which he dismisses in a single paragraph the idea that objects inside space and time, like neurons, could create consciousness; it was so obvious to him he moved on. In his book the Monadology he proposed that conscious, perceiving entities are the fundamental reality. A monad was his technical term for these perceiving entities. His dynamics was a strange one he called pre established harmony, with God brought in to coordinate all the perceptions. Hoffman raises Leibniz only to show this lineage: centuries ago, a thinker held that experiencing entities could be more fundamental than the physical space time world.
Hoffman's own version is conscious agents, a mathematically precise statement of what he means by consciousness. As a scientist it is not enough to say consciousness is fundamental; he has to write it down. He pauses to define consciousness for newcomers: the ability to have experiences, the taste of chocolate, a headache, emotions, the fact that there is something it is like to be a conscious entity. Philosophers call this qualia, and qualia become central to the paper.
For the mathematical definition he applies Occam's razor and takes the bare minimum. There are experiences, like the taste of chocolate, and those experiences affect the probabilities of other experiences occurring. That is it. No self, no learning, memory, problem solving, intelligence, or free will baked in; all of that has to be proven to arise from just experiences and their probabilistic relationships. The experiences are formalized as probability spaces, and the relationships among them as Markovian kernels, giving Markov chains. Few miracles, simple dynamics.
The rule set, and why every headset is a constraint
Bilyeu seizes on Hoffman's phrase that experiences do not lead to other experiences "in a vacuum." That is his whole thesis. Space time, the simulation everyone lives in, is a required constraint that gives context so that something can be like anything. For consciousness to explore the possibility space of qualia, you need a rule set. The rule set we are in (maybe one of a gazillion headsets) creates the possibility for the subset of qualia that humans or lizards experience. Without the limitations that space time imposes, there would not be enough context to feel a certain way. Hoffman calls it a very good way to put it.
They riff on the diversity of headsets across animals. A lizard sees very differently than we do. Pigeons have four color receptors to our three, so they see more color; the mantis shrimp has more than ten photoreceptor types. We cannot perceive the polarization of light, but birds and baby bees can. We cannot directly sense electric fields, but some aquatic animals can. Some animals see infrared, some ultraviolet. We have a very small window, and other animals are not restricted to it. Hoffman likes the framing: there is an infinite space of conscious experiences, and different animals are different explorations with different headsets and different constraints. Consciousness is exploring all its possibilities, and this ride never ends.
What is consciousness "made of"? Awareness and the probability space
Bilyeu admits he keeps imagining consciousness as a blob that takes the shape of a human or a lizard or an avocado, and asks what image Hoffman holds. The closest Hoffman can communicate it: go into a quiet dark room, close your eyes, get very still, and try not to think. If you manage even a few seconds with no thought, you are just aware, aware without being aware of anything in particular. You realize you are fundamentally awareness, and into that awareness come a cup, a microphone, a table; close your eyes and they vanish. There is a field of awareness that is, in some sense, deeply and fundamentally who you really are.
Bilyeu asks whether the theory would call that awareness false. Hoffman explains why he even brings it up. To do the math, he had to first write down a probability space, the space of all potential qualia a given conscious agent could experience, sitting there as mathematical structure prior to any particular experience. For years he just wrote it down and moved on to the fun dynamics. Then he came back and asked what that space actually is, a space prior to any specific conscious happening, and the best answer he has is that it may be the mathematical counterpart of that bare awareness, the capacity to experience prior to any specific experience arising.
Bilyeu restates it carefully, invoking the warning Annaka Harris gave Hoffman about letting people drag their sense of self into this: consciousness is the awareness that animates Bilyeu, or that needs his constraints in order to experience qualia. In meditation he can reach the state of being simply aware of being aware, not aware that his foot hurts or his stomach is churning, just the potential to point awareness at something. That, Hoffman agrees, is who we really are. Bilyeu's framework clicks: he is one instantiation, a set of biological limitations, and once those constraints exist the fundamental element of consciousness can explore the qualia available to this human form.
Why does consciousness explore at all? Cantor's infinities
Bilyeu's fundamental question: why is consciousness compelled to explore these qualia states? Hoffman calls it the sixty four thousand dollar question and does not claim to know, but offers two mathematical hints. What we are, he says, are avatars of the one; the one awareness is exploring all its possibilities through different avatars.
First hint: no system can completely know itself. If a computer tries to model itself, the very act of writing facts about itself into memory changes it, making it more complicated, so to truly understand itself it must now describe what it just did, and you get an infinite loop. In many cases it is provably impossible to have a complete self understanding.
Second hint: the hierarchy of infinities, from Georg Cantor. The integers are an infinite set, called countable infinity or aleph null, the smallest infinity. But take the set of all subsets of the integers, the power set, and you get a strictly bigger infinity. Bilyeu's reaction, that his speaker goes to 11 instead of just making 10 louder, captures the absurdity, but Cantor's diagonal argument proves on pen and paper that you cannot capture the power set with the smaller infinity. And it never stops: take the power set of that, and the power set of that, forever, an unending hierarchy of ever larger infinities. Hoffman's theory of consciousness has to take this in: all these infinities are valid directions for the projection of the one deeper consciousness. So the answer to "why explore" may simply be that Cantor's hierarchy never ends, and so neither does the exploration of qualia.
In or out of the headset? Einstein, GPS, and the first baby step
Bilyeu asks the operator's question: when Hoffman says he wants to get out of the headset, does he want out, or does he want to manipulate the headset? Einstein's breakthroughs were useful precisely because, inside the headset, they let us do things: bending space time, GPS, nuclear energy, the atom bomb. Are you after a headset breakthrough or an exit breakthrough?
Both, says Hoffman. He wants a theory of what lies beyond, at least a baby step past the headset, and there is infinite job security in that because Cantor's hierarchy never ends. But as a scientist he must make predictions back inside the headset, because that is the only place experiments happen. The theory should recover quantum field theory as a special case, general relativity as a special case, evolution by natural selection as a special case, or sensible generalizations of them; otherwise it is wrong. You cannot go outside space time and do anything you like; you must tie it back to what we can perceive. He notes that in the last decade physics has already stepped past space time with structures like the amplituhedron and decorated permutations, first baby steps that will be refined and superseded.
Particle scattering, quarks, and probing space time itself
Bilyeu presents one of those baby steps he thinks he has grasped well enough to relay. Physicists smash particles together to reveal smaller building blocks, and in the flood of collision data they find patterns that replicate endlessly. Out of an overwhelming amount of data, only so many patterns appear, like a mirror that always shatters into the same shape. Hoffman confirms it is statistical: there are statistical commonalities to the interactions.
He explains why this matters. Eugene Wigner taught that particles are irreducible unitary representations of the group of symmetries of space time, the Poincaré group. Particles are the simplest entities allowed by the symmetries of space time, so studying them is really studying the structure of space time itself. In the Large Hadron Collider, smashing protons (or an electron into a proton) at high enough energy destroys the proton, and out spray quarks, gluons, mesons. From the angles, spins, magnetic charges, and masses, patterns emerge.
He details quark confinement as a famous surprise. A proton holds three quarks (two up, one down); a neutron two down and one up. But a quark cannot fly out alone. As you try to pull two quarks apart, the force between them stays constant while the potential energy keeps growing with distance, until at some point the bond snaps and that energy creates a brand new quark, so they pair off again. We have no analytic proof of quark confinement for non abelian gauge theories; lattice gauge models and experiment convince physicists it is true, but the analytic proof is one of the big open questions (indeed a Millennium Prize problem). He recalls the early atom: the plum pudding model predicted particles shot at atoms would mostly pass through, and most did, but a tiny fraction bounced back, revealing point like protons and neutrons in mostly empty space. Look inside the protons and you find quarks and gluons, which at our best resolution (a thousandth to ten thousandth the diameter of a proton) still look point like, and might themselves be composite, we just cannot resolve it.
A theory of space time vs a theory of everything, and Turing's limit
Bilyeu asks why we have failed to get a theory of everything for space time, given that even in GTA the rules are knowable, cause and effect, and a bug is just the program doing what it was told even if not what you intended. Hoffman partly agrees: within that framework we could get a complete theory of space time, though not a theory of everything, because for him space time would be a trivial aspect of everything. We will also see space time's limits: it falls apart at ten to the minus 33 centimeters and ten to the minus 43 seconds.
Then he throws in a wrinkle from computation. Alan Turing gave us the universal Turing machine, our model of a universal computer, but it has a hard limit. Consider all functions from the integers to the integers, like squaring. How many are there? A bigger infinity than the integers, uncountable. But Turing proved that the set of computable functions is only countable. So when we program, we use computable functions, a probability zero subset of all functions. Almost every function is non computable, like the function that solves the halting problem, yet almost every function we can think of is computable, because we are stuck inside the headset's limitations. Thinking outside the box in the simulation will be mind numbing, because to really do it you have to learn to think about non computable functions, which is not trivial. If someone claimed computable functions are all we need, Hoffman says the burden of proof is on them.
Dark matter, dark energy, and Markovian dynamics
Bilyeu turns to dark matter and dark energy, the stuff we cannot identify but know must be there or the universe would not hold together or would not be flying apart as it is. What does consciousness as fundamental say about it? Specifically, nothing yet, Hoffman admits; it is a big open question, and one of his collaborators, a student named Ben Knepper, is doing dark energy experiments, mostly to learn the ropes of real experimental work, not because Hoffman expects it to yield results tied to consciousness.
But from a view where space time is not fundamental, it is no surprise that there could be influences on the headset that are not explicitly represented inside it, seen only as influences. To explain how, he defines Markovian dynamics: what you do next depends only on where you are now, not on your whole history. Step right with probability two thirds, left with one third; the key Markov property is that only the current state carries all the information about the probabilities for what happens next. Bilyeu offers the airport analogy he found helpful: if you are at an isolated airport with one connection, no history is needed to know where you fly; at a hub with five options it is a probability curve. This matters computationally, because tracking every element interacting with everything would require a computer the size of the universe; Markov dynamics eliminates the need for infinite memory by demanding only finite memory, which you can make as large as you like.
Hoffman adds a deeper reason to care: Markovian kernels are computationally universal (anything a neural net or Turing machine can compute, they can compute), and yet they are not limited to computable functions, because the sets on which the probabilities are defined need not be computable. So they offer a window toward going beyond computation, a window he is not stepping through yet but knows is there.
Now the payoff for dark matter. Space time is just a projection of the conscious agent dynamics. Most of the states of that dynamics are not represented in space time; they are dark. When you count all the matter and energy visible inside space time, you are missing everything that did not project in. So dark energy and dark matter are probably much more than we have discovered, and they are influences from states outside the headset. Bilyeu, reminding himself it is all qualia, gropes for an image: dark energy as the energy of a quale he will never experience, "an alien drinking blood wine." Hoffman says it is even more complicated, not one quale but countless infinities of interacting qualia affecting the dynamics we perceive.
Position is a quale, and the moon trick
Hoffman makes a point that reframes the whole thing: among the qualia is position. It is very different to experience Tom four feet away versus four inches away; depth and space are themselves qualia. And our quale of depth is heavily compressed. A distant mountain might be twenty miles away while the moon rising over it is a quarter million miles, yet the moon looks only a little farther. Our qualia space of depth is compressed compared to the measured world, exactly like a GTA world where you only render a little of the road around you even though you could drive thousands of simulated miles. There is a supercomputer with a much bigger world than your headset shows at any moment. Going to the moon in a rocket is like walking through GTA to a place too far to see before, getting it into your headset, which again points to a world outside your headset. Galaxies far away are headset stuff you can eventually render; dark matter and dark energy are the deeper states in the computer you will never directly see but that subtly influence what you do.
Bilyeu presses a sharp objection: if consciousness is cycling through different qualia, why would something that can never interact with us be in the same simulation? Hoffman's answer is stark: almost everything the real consciousness is doing is not in our headset. Of all the things experienced in consciousness, the percentage we experience is, effectively, zero percent.
Bilyeu's counter theory: the thirteen year old programmer
Here Bilyeu lays out his own alternative, which he returns to repeatedly. He does not need the conclusion that consciousness is fundamental; that just becomes a debate about whether consciousness can emerge. What feels more right to him, while he fully admits it only kicks the can down the road and pushes God one level down, is that he exists in someone else's simulation that exists in their real world, and that person still needs God or something. The rules feel a little too perfect, too finely tuned; you have the Fermi paradox; the fact that most of the universe's energy is "dark stuff" feels like a thirteen year old programmer hand waving away the thing he needed to make it all work. As a game developer, Bilyeu notes the hardest thing is creating the art assets, so you need something that renders on the fly, which is exactly what he sees AI doing now, pulling things out of the possibility space. You could go to the Unreal Engine store, grab Einstein's physics, plunk it in, and it works without the programmer even understanding it.
So what gives Hoffman the confidence the thing giving birth to all this is consciousness itself? "I'm not confident at all," he says; it is just his leading theory. He agrees you could stay agnostic, write down dynamical entities outside space time with no claim about their nature, derive the projection to space time, and be fine. He goes after consciousness for two personal reasons. We all have headaches and conscious experiences, and we want to understand what consciousness is. And the standard view, that consciousness is created by brain activity while physics is fundamental, collapses if space time is not fundamental. If space time is doomed, that whole story of consciousness goes out the window.
If space time is doomed, is physicality doomed too?
Bilyeu answers his own question with Hoffman's words: if space time is doomed and local realism is proven false, then physicality (at least the physicality we experience) cannot be true, because it is all rendered in real time as you engage with it. Hoffman puts it bluntly. Right now he does not have neurons. If you looked inside his skull you would render neurons, but there are no neurons there when unperceived. So neurons cannot create consciousness, because they are not even there to do it; and likewise particles do not exist when unperceived.
Bilyeu hits the wall every limited mind hits: for this to be a simulation, something has to be running it, and somewhere that something has to be physical. Hoffman confesses this is hard for him too; he has all the same knee jerk emotional reactions, even stronger. His emotions do not believe any of it; only the mathematics pushes him, kicking and screaming at each step. He does not find it intuitive. He grants consciousness could be part of the simulation, and that there might be that bare awareness prior to any particular experience, but admits he has nothing intelligent to say about that thing and feels no confidence there at all. Bilyeu floats a guess he likes: maybe awareness is just the quale of your process being run by the central computer, of being rendered. Hoffman calls it as good an idea as he has had, but does not feel confident.
A crucial clarification lands here. Bilyeu assumes the headset is computational. Hoffman corrects him: we have not proven that the failure of local realism means reality must be computational. It does not entail that at all. The processing that moves you from one quale to another need not be a physical process, and need not even be a computational process; it could be functions that are not computable. The standard non computable example is the halting problem, which Hoffman walks through in detail: there is no Turing machine that can decide, for every possible program and input, whether a given Turing machine will halt. The machine that tried to decide it would itself never halt. (He notes that being "Turing complete" is a different concept entirely.)
Persistence: the hard problem for AI and the rule of the headset
Bilyeu names what he calls a key part of his thesis: persistence is the hard problem for AI right now. AI can give amazing output, but not the same amazing output over and over from different angles and setups, which is why it is hard to drop into real workflows. He suspects persistence is part of the rule set you need inside the headset to let qualia be explored.
Hoffman grounds the term. Persistence here means: he looks at the moon, looks away, looks back, and the moon is still where he expects it; and if a photo were taken, both would agree it is accurate. He does not think you strictly need that kind of persistence to have a headset, and raises E. coli as a minimal case. The bacterium swims up a gradient of food; when the gradient goes bad, its flagellum reverses and tumbles it into a random new direction, a stupid but effective search, a random orientation generator. Does E. coli need persistence? From its own headset, maybe it only needs to know "I'm happy, I'm eating" or "bad, turn." Bilyeu insists persistence is still there from the outside: if he discovers E. coli and someone else independently discovers it, it looks the same to both. Hoffman agrees on that external sense.
This is where Eric Weinstein, whom Bilyeu had introduced to Hoffman right before recording, comes up. When Hoffman said "I don't have neurons," Weinstein balked; he had argued that Hoffman could not render him on first meeting because he remained consistent to himself, which is precisely Bilyeu's persistence. Bilyeu drives it home: you build a simulation with one persistent moon that looks the same to everyone who discovers it, rather than coding a thousand private versions and a lookup table for whose red is whose. Smash a skull a hundred thousand years ago and the neurons look the same as smashing one today, even if you do not know what neurons "are," because the simulation has a level of persistence built into its assets and rules. Hoffman fully agrees on persistence as uniformity of assets, while distinguishing it cleanly from behavior, which is a separate matter (the E. coli's "good or bad" response).
| The claim | Hoffman's case for it | The friction / counterpoint |
|---|---|---|
| Local realism is false | Settled. 2022 Nobel, Bell tests by Clauser, Aspect, Zeilinger; no micro/macro boundary in QM. | Establishes that unmeasured properties lack definite values; it does not by itself prove consciousness is fundamental. |
| Senses hide the truth | Theorem. In evolutionary game theory the probability of truth preserving perception is 0. | Shows perception is an interface, but an interface still needs something on the other side; nature of that something is open. |
| Consciousness is fundamental | His leading hypothesis; the simplest model points to a single universal consciousness. | Hoffman: "I'm not confident at all." Bilyeu: you could stay agnostic about the entities outside space time. |
| The headset is computational | Markov kernels are computationally universal and model the dynamics. | Not proven. The dynamics may use non computable functions; processing need not be physical or computational. |
| What runs the simulation | Conscious agents outside space time, rule governed, existing unperceived. | Bilyeu: something somewhere must be physical; this just pushes God down a level. |
The paper: free will, action, and mapping qualia to particles
Bilyeu moves to Hoffman's then forthcoming paper, which felt to him like a step up in mathematical certainty from the popular book. He raises an idea he calls "action potential" (admitting he may be misremembering the term) and the discomfort that, if consciousness is fundamental and all it does is run something like Gödel's incompleteness against every possible quale, then every computation must be about qualia and all math is just the representation of the elements of qualia. He feels he has free will, thinks it is an illusion, and cannot reconcile living a life with the picture of consciousness simply cycling through qualia. It feels like "I'm tired, so I go to bed, so I feel the softness of the sheets," not "consciousness decides to cycle through the softness of sheets now."
Hoffman says "not quite." There is a complicated dynamics of consciousness, and the coin of the realm is experiences, shared and triggered. He reaches for Twitter as the analogy: experiences are like tweets being broadcast and received across a vast social network of conscious agents. Your individual headset is a tiny projection of that huge dynamics, the way Hoffman's own Twitter feed (a small set he follows and who follow him) gives a wildly unrepresentative picture of the whole twitterverse. So all the headsets are projections of the one bigger conscious dynamics onto consistent subsets, and to keep persistence you might want a whole group of agents that see the same thing and can all agree there is a red Porsche.
On free will, Hoffman explores his own math live. He could interpret the Markov kernel for a given agent as that agent's choice. But that agent's probability space is impacted by all the other conscious agents. The trick is that groups of conscious agents satisfying certain conditions are themselves a conscious agent. So if the small agents have free will, the larger composite agent needs a new notion of free will too, a scale free notion that holds no matter how many agents you combine. His guess is this can be made to work: it is not "only the one big consciousness has free will and we are puppets," nor "only the little ones," but possibly both, with the free will of the one composed of the genuine free choices of the individuals. Not an autocratic God commanding, but a God freely exploring all its own possibilities through components that are themselves freely interacting. He has not written this down yet; these are deep waters, and he admits that ultimately he cannot give a final description of the one agent, because the mathematics says it transcends description. He can only describe projections, and the projections point toward the one.
Bilyeu asks how the projections point to a single God consciousness. Hoffman is careful: the data do not uniquely determine the theory. He is a conscious being interacting with others he believes are conscious, his physicalist framework cannot explain them, so he proposes conscious experiences giving rise to space time as an interface, and when he writes the simplest mathematical model, the mathematics, unbidden, points to a single major consciousness he can never describe. He guesses other conscious agents with entirely different interfaces, hitting the limits of their own interfaces and realizing they were just interfaces, might get pointed by their own mathematics to the same universal consciousness.
Does consciousness have a form? The neurons question
Bilyeu drags him "to the bottom of the Marianas Trench": does consciousness have a form? Awareness has no form, Hoffman says, but assumes all sorts of conscious forms. Bilyeu's worry is that for the headset to be so specific, consciousness must be a certain way, and if there is no physicality, what sets consciousness moving in a specific direction? If it must pull from a probability space, then math comes before consciousness; if not, what gives it direction?
Hoffman's answer is that there are countless ways consciousness can create headsets, and it does them all. To answer Bilyeu properly he would have to write down an actual dynamical system of conscious agents that constructs our space time framework. He sketches the recipe. Start with a large dynamical system of conscious agents (say a trillion), arrange the dynamics of a subset (say a billion) so they create a particular space time headset (possible because Markov kernels are computationally universal). Then turn that headset around and have it look back at the whole agent system, in particular the agents involved in creating the headset. What does that set of agents look like? They look like neurons and brains.
Bilyeu balks: he thought neurons did not map like that, that they were just artifacts of the interface. Hoffman keeps it consistent. Neurons are interface symbols. There is a systematic relationship between the symbols in your headset and the underlying software. The moon does not need to exist to simulate gravity; gravity and the moon can be program rules. But there will still be a systematic relationship between headset symbols and the software. So consciousness will not look like neurons, but you could build an interface such that, when it looks back at the conscious agent network that created it, it sees that network as if it were neurons and brains. In Hoffman's clean formulation: neurons are our interface looking at the conscious agents that are constructing our interface. They are the headset symbols our headset gets when it looks at the agents building the headset. A simpler creature might see only a single neuron, or some other structure, depending on its interface.
Death: take off the headset
Bilyeu pushes to the existential edge. Two buttons. Button one: completely exit the headset, your avatar falls over and appears dead in the game world, but you are now out with the consciousness, aware of your oneness. Button two: stay in the matrix knowing it is fake, but with special powers, like Neo. Which do you press?
Hoffman first reflects on death itself. His guess is that at death we take off the headset; we may lose a lot of what was in it, but we are still aware, just no longer tacked into the headset. He is completely open to being wrong, deeply wrong, and notes that near death experiences may or may not point to this. If you are a physicalist, it is clear: brain dead means consciousness gone. But the view that consciousness creates space time and brains are just headsets leaves open that the consciousness looking through this avatar does not perish when the avatar perishes. That is not what motivated the view, but it is compatible with it. Intellectually he is open to it; emotionally he fears death, the Darwinian fear wired into him as part of the game.
Pressed on the two buttons, Hoffman picks the new stuff. Three dimensions of space and one of time feel confining; we got a cheap headset, a cheap simulation. He would love to see what else is on offer. Solving mathematical problems, he can imagine a three dimensional shape but not a four dimensional one, and the geometries he needs (six, nine, more dimensions) can only be reached by crawling up through theorem and proof, "like blind men feeling the elephant with theorems." He would love a headset where he could see nine dimensional space at a glance, or thirty, or a thousand, or a billion dimensions.
Bilyeu counters with the Freaky Friday problem. He often wishes he could swap bodies with his wife for 24 hours to understand her, but the moment you swap bodies you simply become her; there is no "him as her." So even if Hoffman is right, taking off the headset removes everything you think of as you. If all consciousness lives to do is cycle through other qualia, you would either be reincarnated (popping back into a new headset) or return to the hive, the Borg, the ant colony, reinstantiated as pure awareness, and all the loving, clinging, hating, attachment, precious moments, all of it gone. Bilyeu's hard won view is that people exploring these ideas forget how mired they are in the gruesome reality of human experience; to transcend it and feel no pain would make you so different you would not recognize anyone, and he has heard no theory that cheats death except regrowing your biological organs. Everything else means you die and everything you love goes away, maybe traded for something better, but gone.
Hoffman offers another take: if you and I are just the one looking at itself through avatars, then whatever the one learns through these avatars is not lost; it becomes part of the one, which is in some sense eternal. He would still pick the new headset, partly out of inquisitiveness (what is it like to live in five dimensions, to have twenty dimensions of color, a thousand dimensions of emotion), partly because what we have now is the training wheels version. Nothing is lost by moving on, and the potential is mind bogglingly infinite, this whole Cantor hierarchy. He stresses these are deep waters and he is now speaking intuitively, not by theorem and proof, and that all the ideas of this first generation stepping outside the interface will look naive a century or two from now.
The arrow of time is an artifact, and Einstein's lesson
Bilyeu asks whether there is grounded mathematics that keeps the intuitive exploration from spinning into la la land. Hoffman points to the paper, due out June 24th and to be tweeted, which shows how specific properties of the Markov dynamics of conscious agents map to specific properties of particles, mass, spin, momentum, energy. One precise proposal: mass is the entropy rate of recurrent communicating classes of conscious agents. He goes after particle scattering not from a fetish for high energy physics but because particles are the simplest things the interface offers, the place to make the first connections; brains are quadrillions of particles and the wrong place to start. A real particle physicist put his name on the paper, not because he is convinced it is right, but because if it is wrong it is not obviously wrong and is worth pushing on. Bilyeu praises the stance: be fearless in predictions, hold yourself accountable to learning rather than to being right, and actively want people to find the edges of your theory.
Hoffman then drops one of the most counterintuitive claims in the talk: the dynamics of conscious agents need not have an arrow of time. You can write down a Markov dynamics in which entropy does not grow. And there is a three line theorem proving that any projection of such a dynamics that loses information (say by conditional probability) yields a new dynamics that does have an arrow of time, precisely because of the lost information. So our experience of an arrow of time, the big bang, the eventual crunch or heat death, is not an insight into what lies beyond space time; it is an artifact of the projection. From an evolutionary view, time is the fundamental limited resource: run out of time before your next meal or drink and it is over. But that limited resource is also an artifact of projection from a timeless conscious agent dynamics, which suggests all the limited resources are artifacts. Evolution by natural selection is a beautiful theory, but it is the theory of the artifacts you see when you project from a realm with no limited resources and no competition.
He underlines why precise mathematics matters, with Einstein as the parable. Einstein had the falling elevator intuition around 1906 or 1907 and worked for years, through sleepless nights and state of the art new math, to make it precise by 1915. A year later Karl Schwarzschild wrote back with a solution that predicted what we now call black holes. Einstein did not foresee them, did not like them, disbelieved in them, and wanted to get rid of them. His own theory came back and surprised him. That is why we do science: make your best ideas mathematically precise, and the mathematics slaps you in the face with implications you could not think deeply enough to reach on your own, and tells you the limits of your concepts (Einstein's gravity plus quantum field theory tell us space time has no operational meaning below ten to the minus 33 centimeters, something not even Einstein could have guessed). An amateur with binoculars might be smarter than the astronomer at the James Webb, but will never beat the better tool. So our intuition about exploration happening within an arrow of time will have to be re jigged, because in the realm beyond, there need be no entropy arrow at all.
A new color you cannot imagine
Bilyeu notes a subtle disagreement in assumptions: he had resisted Hoffman's claim that the dynamics need no arrow of time, because for at least the qualia we can access, the arrow of time is required. Hoffman agrees, and reframes it: the only qualia we have direct access to require the arrow of time, but we presume infinite headsets offering unimaginable qualia that may not. We have been shaped by our headset to think in time. Try to imagine a new color you have never seen; you cannot, not because none exist (pigeons with four receptors presumably see colors no human could imagine, and the mantis shrimp sees still more) but because your interface forbids it. What is it like to see the polarization of light, to have infrared vision like a pit viper, to sense an electric field, to do echolocation like a bat? He has no idea. These are pointers, inside the headset, to realms of qualia utterly outside anything he can imagine.
Near death experiences, and where to follow Hoffman
Finally, near death experiences. From a physicalist framework, they are obviously just a brain malfunctioning in the final throes of oxygen deprivation. But if space time is doomed and not fundamental, that leaves open (without dictating) that near death experiences are genuine insights into conscious experiences transcending the space time interface. Hoffman would treat the reports like any eyewitness testimony: with a grain of salt, seeking corroboration, discounting, but not ignoring the data either. Our preconceptions are usually deeply wrong (flat Earth, geocentrism, absolute space and time), so even if consciousness survives death, the way we picture it is probably wrong; the move is to make bold but precise proposals so we can find where they fail.
He notes common features catalogued by researchers like Raymond Moody, a light tunnel, a life review, though not all reports are pleasant, some are horrific. Someone he knows personally had a very pleasurable near death experience and came back claiming no fear of death. He mentions being part of a film on near death experiences connected to cardiologists (he references NYU Langone) who, with modern techniques, can keep a body from deteriorating long enough to bring patients back and document what they report. If that data is real, the prediction is not only that a sense of consciousness remains but that sensory perception persists for a while, since people retain peaceful experiences from the period. Hoffman keeps insisting on the scientific posture: this is data, ask what it entails about letting go of the headset, and stay careful about interpretation.
The interview closes with Hoffman pointing people to his work: he posts on Twitter at @DonaldDHoffman, links every paper to the actual journal article along with talks and podcasts, and scientists can find his roughly 120 publications on Google Scholar. Bilyeu signs off with his usual "be legendary," and notes a follow up compilation of his best moments with Hoffman, teasing the line that with this mathematics, "for the first time ever we might be able to formulate a precise hypothesis about what we mean by the word God."
Key takeaways
- Hoffman's case for "space time is a headset" rests on two pillars: the 2022 Nobel confirming local realism is false, and his theorem that the probability evolved senses report objective truth is exactly zero. Both say we perceive an interface, not reality.
- Conscious realism is his proposed alternative: consciousness is fundamental, formalized as conscious agents whose only ingredients are experiences (qualia) and the probabilities (Markov kernels) linking them. Space time, brains, and particles are projections of this dynamics.
- Hoffman is explicit that consciousness being fundamental is his leading hypothesis, not a confident conclusion. The two pillars are strong; the metaphysics on top is a bold proposal offered for falsification.
- Bilyeu's running counter is that you can accept the interface idea without consciousness being fundamental; a "thirteen year old programmer" with a rule set explains the data and just pushes the deeper question down one level.
- The failure of local realism does not prove reality is computational. The underlying dynamics may use non computable functions; the halting problem and Cantor's unending hierarchy of infinities recur as reasons our headset cannot capture everything.
- Provocative implications follow: dark matter and dark energy as states outside the headset, neurons as interface symbols of the agents building the interface, the arrow of time as an artifact of an information losing projection, and free will as a possibly scale free notion true at every level of agent.
- The paper's concrete bet is that properties of the agent dynamics map to particle properties (e.g. mass as the entropy rate of recurrent communicating classes), starting with particle scattering because particles are the simplest interface symbols.
- Both men insist on the scientific method throughout: make precise, falsifiable predictions, invite people to find the edges, and let the mathematics surprise you, the way Schwarzschild's black holes surprised Einstein.
Chapters
0:00:00 AI, consciousness, and why Hoffman is not an alarmist 0:01:30 The synopsis: we live in a simulation, space time is the headset 0:04:00 Pillar one: the 2022 Nobel and the death of local realism 0:08:00 Pillar two: evolution did not evolve us to see the truth 0:11:00 Rendering on the fly and the Grand Theft Auto coordination problem 0:16:00 Mapping the analogy back: consciousness as the computer 0:22:00 Consciousness as fundamental, Leibniz, and the monads 0:27:00 Conscious agents, qualia, and the minimal definition 0:30:00 The rule set: every headset is a constraint on qualia 0:35:00 Awareness, the probability space, and the quiet room 0:42:00 Why explore? Cantor's hierarchy of infinities 0:48:00 In or out of the headset? Einstein, GPS, the first baby step 0:52:00 Particle scattering, quark confinement, probing space time 1:02:00 A theory of space time vs a theory of everything; Turing's limit 1:10:00 Dark matter, dark energy, and Markovian dynamics 1:20:00 Position is a quale; the moon and the compressed depth 1:28:00 Bilyeu's counter theory: the thirteen year old programmer 1:36:00 If space time is doomed, is physicality doomed? No neurons 1:46:00 Persistence: the hard problem for AI and the rule of the headset 1:58:00 The paper: free will, action, and the Twitter analogy 2:10:00 Does consciousness have a form? Neurons as interface symbols 2:20:00 The two buttons: death, reincarnation, and taking off the headset 2:30:00 The arrow of time is an artifact; Einstein and Schwarzschild 2:38:00 A new color you cannot imagine; near death experiences 2:42:00 Where to follow Hoffman
Notable quotes
"There's no theory on the planet today that can start with an artificial intelligence and a description of some kind of circuit or some kind of software pattern of activity and can give you a specific conscious experience like the taste of chocolate or the smell of garlic." Donald Hoffman, 0:01:00
"The Nobel Prize last December was given to three physicists for the experimental testing of a clean prediction of quantum theory, that something called local realism is false." Donald Hoffman, 0:04:00
"In case after case the answer is the probability is zero. There are payoff functions that would preserve the structure, but those payoff functions have probability zero in the set of all payoff functions." Donald Hoffman, 0:09:00
"The Porsche doesn't even exist until I render it, and there's no red Porsche inside the super computer." Donald Hoffman, 0:13:00
"What we are are avatars of the one. The one awareness is exploring all of its possibilities through different avatars." Donald Hoffman, 0:18:00
"My emotions don't believe any of this, they don't believe it at all. It's literally only the mathematics pushing me kicking and screaming at each step." Donald Hoffman, 1:42:00
"I don't have neurons right now. If you looked inside my skull you would see neurons, you would render them, but there are no neurons. Neurons do not exist when they're not perceived." Donald Hoffman, 1:40:00
"Neurons are our interface looking at the conscious agents that are constructing our interface." Donald Hoffman, 2:15:00
"Three dimensions of space, one dimension of time feels quite confining to me. I feel like we got a cheap headset." Donald Hoffman, 2:24:00
"Einstein's theory came back and surprised him. That's why it's so important for us to do science, because the mathematics can take you where your own consciousness wouldn't necessarily go." Donald Hoffman, 2:33:00
"You should hold yourself accountable to always learning and getting a little bit better." Tom Bilyeu, 2:08:00
Resources mentioned
- Donald Hoffman, cognitive scientist at UC Irvine, on Twitter and Google Scholar
- Tom Bilyeu and Impact Theory
- The Case Against Reality, Hoffman's book on the interface theory of perception
- Conscious agent theory, the foundational paper with Chetan Prakash
- 2022 Nobel Prize in Physics, awarded to John Clauser, Alain Aspect, and Anton Zeilinger
- Bell's theorem and the principle of locality
- Gottfried Leibniz and the Monadology
- Qualia, probability spaces, Markov kernels, and Markov chains
- Georg Cantor, aleph numbers, the power set, and Cantor's diagonal argument
- Alan Turing, the universal Turing machine, and the halting problem
- Gödel's incompleteness theorems
- Eugene Wigner and the Poincaré group
- Large Hadron Collider, quark confinement, Yang Mills theory, and the Yang Mills Millennium Prize problem
- The amplituhedron
- Stable diffusion and ChatGPT
- Apple Vision Pro, Unreal Engine, and Grand Theft Auto
- Karl Schwarzschild and black holes
- Eric Weinstein and Annaka Harris
- E. coli, the mantis shrimp, and pit vipers
- Raymond Moody on near death experiences, and NYU Langone
- The Fermi paradox
Where it stands
It is worth separating, honestly, the parts of this conversation that rest on settled science from the parts that are speculative architecture built on top of it.
The two pillars are real. The 2022 Nobel did confirm, through decades of tightening Bell tests, that local realism is false; physicists genuinely accept that unmeasured quantum properties do not hold definite pre existing values in the classical sense. And Hoffman's fitness beats truth theorem is a published result in evolutionary game theory: under his models, perception is tuned for fitness, not for veridical reporting of structure. Where the mainstream parts ways is on the leap from "perception is an interface" to "consciousness is the fundamental substrate of reality." That second move is a philosophical interpretation, idealism in modern mathematical dress, and it is not entailed by either pillar. Most physicists who accept the failure of local realism do not conclude that space time emerges from consciousness; competing readings (relational quantum mechanics, QBism, many worlds, superdeterminism) keep a mind independent world. Hoffman knows this, which is why he repeatedly says "I'm not confident at all" and calls conscious realism his leading hypothesis rather than a finding.
The honest strength of Hoffman's program is that, unlike most consciousness talk, it is trying to be falsifiable: write down precise dynamics, derive particle properties, get slapped by the math. The honest weakness is that the central object, the one universal consciousness, is by his own admission beyond mathematical description, and the bridge from agent dynamics to our specific physics is, as of this conversation, a promissory note. Bilyeu's "thirteen year old programmer" objection is the sharp one and lands cleanly: every simulation style argument, including Hoffman's, explains the within world rules while pushing the question of what grounds it all down one more level, and conscious realism does the same, relabeling the floor "consciousness" instead of "base reality." That is not a refutation, it is a reminder that the deepest claim here is metaphysics, chosen for its elegance and its dissolution of the hard problem of consciousness, not compelled by data. The right posture is the one both men keep returning to: take the two pillars seriously, hold the grand inversion loosely, and watch whether the promised predictions about mass, spin, and scattering survive contact with the collider.
