Hassabis Was Right. He Just Didn’t Finish the Sentence.

On a casual remark about information, taking on the wave function, and the architecture nobody has named yet.

Over the weekend I watched a fireside chat on a stage at a venture capital firm in San Francisco, where Nobel laureate and DeepMind founder Demis Hassabis said something that almost no one really picked up on. He was talking about the future of AI. The conversation had moved from neuroscience to AlphaFold to WeatherNext 2, the simulator his lab has built for weather, and on to their breakthrough work in biology. Somewhere in the middle of an answer about world models Hassabis made a metaphysical claim that would have seemed wild coming from anyone with less authority. He said it the way you might mention the weather. Information, he said, is more fundamental than matter or energy. He did not stop to defend it. He moved on to talk about virtual cells. The mainstream coverage focused on his prediction of AGI by 2030 and his views on consciousness as a tool. The information remark dropped into the conversation, made a small ripple, and disappeared.

I want to argue that it was the most important thing he said that day, and that he stopped one sentence too soon.

The standard interpretation of the Hassabis claim is that he was being loose. Of course information matters, the standard interpretation goes; of course modern physics has had an information turn; of course Wheeler said it from bit and Bekenstein wrote down the entropy of black holes and the holographic principle has been with us for thirty years. None of this is news. Hassabis was just expressing, in the casual register of a Q&A, a sentiment any well-read theoretical physicist would nod at. Move on.

I think this misreads what was happening. Hassabis is not a physicist. He is not in the business of restating sentiments from theoretical physics for an audience that already knows them. Hassabis is the person who has spent the last fifteen years building the most successful applied program in AI, the lab whose protein-folding work won a Nobel Prize, the man whose teams have produced more functioning intelligence than any other research organization in the world. When that person says, in passing, that information is more fundamental than matter or energy, he is not citing a slogan. He is reporting from the inside of an experience.

The experience is this. For fifteen years, the systems that work have been the systems that treat the world as a pattern to be read rather than a substance to be modeled. AlphaFold did not start from a theory of what proteins should look like. It learned, from data, what they actually do, and produced, in months, predictions that decades of theory-first work had failed to deliver. The same thing has now happened in dozens of domains. Game-playing systems, language models, weather simulators, materials discovery, the early biology models. Over and over, the lesson has been that the substance-first approach loses to the information-first approach. The pattern is the thing. Find the pattern, and the substance takes care of itself.

Hassabis is the person who has watched this happen most consistently and at the largest scale. When he says information is fundamental, he is not theorizing. He is describing. He is saying: I have built tools that work because they treat the world this way, and the tools tell me something about what the world is.

This is a much stronger claim than the standard interpretation gives him credit for. And it has a consequence he did not, on stage, follow through on. The consequence is about quantum mechanics, of all places, and about an argument that has been running for a hundred years and that almost no one outside theoretical physics has noticed is suddenly relevant.

IS THE WAVE FUNCTION THE MOST SUCCESSFUL WRONG IDEA IN THE HISTORY OF SCIENCE?

There is a famous story physicists tell, most recently by Carlo Rovelli in his Helgoland. In 1925, a twenty-three-year-old Werner Heisenberg, exiled to a small treeless island in the North Sea by his hay fever, came back with the first version of quantum mechanics. It was written in terms of observables, quantities you could in principle measure, arranged in tables that did not commute when you multiplied them. The math was strange. The picture was concrete. Six months later, Erwin Schrödinger published a different version, written in terms of a wave function that evolved smoothly through time. The two versions were soon proven mathematically equivalent. Within a year, almost everyone was using Schrödinger. And his now dead-and-alive cat can be found at every coffee shop, all the way to memes about the Strait of Hormuz being open and closed at the same time.

The reason was simple. Schrödinger’s wave looked like the kind of thing physicists already knew how to think about. Sound waves, light waves, water waves, physicists had been working with waves for two centuries. Heisenberg’s matrices required learning new mathematics for an unfamiliar kind of object. It was no contest. The community chose the wave because the wave was easier to draw on a blackboard, and we have been living inside that choice for a hundred years.

Let me be blunt about what this means. The wave function, the central object of every undergraduate quantum mechanics textbook, the thing on every cover, the symbol the field has organized its imagination around for a century, is not what the textbooks suggest it is. It is a representation, and the representation has been mistaken for the thing represented. Some physicists, including Deutsch, argue that the underlying quantum reality is more naturally described in the Heisenberg picture, in terms of local observables; others read the wave function as epistemic, a tool for tracking what we know rather than a piece of furniture in the universe. What these positions share, and what most working physicists quietly know but rarely say out loud, is that taking the wave function as a literal classical-feeling object is the source of much of the conceptual difficulty in the foundations: the measurement problem, the collapse problem, the question of how a wave “over here” can change instantaneously when you measure something “over there.” Many of these problems are artifacts of reifying a piece of mathematical bookkeeping into a piece of furniture in the universe.

Notice that AI researchers, almost without exception, do not make this mistake with their own systems. Nobody at DeepMind thinks the weights of AlphaFold are proteins. The weights are how the model encodes the structure of proteins. The structure is real; the encoding is bookkeeping. Physicists, alone among scientists, made the opposite move. They let their bookkeeping become their cosmology. They mistook the map for the territory and then spent a century arguing about the strange properties of the map. The convergence with AI is partly a chance to un-make that mistake.

David Deutsch, one of the founding figures of quantum computing and author of one of my favorites (The Beginning of Infinity) was the person who in the 1980s asked what would happen if you took quantum mechanics literally and tried to compute with it. He has been arguing for forty years that the choice of the wave function picture was a mistake. Not a calculational mistake. The calculations work either way. An ontological mistake. A mistake about what kind of theory quantum mechanics is, and therefore a mistake about what kind of world it describes.

Almost ten years ago I had a Sunday evening chat with David about his work as he was enjoying a bowl of cereal, and I have been a follower ever since. In a recent conversation with Brian Greene, Deutsch put it sharply. The wave function picture, he said, expresses the theory correctly but misleadingly. It gives you a wave that evolves in some abstract space, and then at the moment of measurement it tells you to square the amplitude and read off probabilities, and the question of what actually happened, what mechanism in the world produced this outcome rather than that one, is left as a kind of mystical residue. The Heisenberg picture, the one we abandoned in 1926, does not invite this problem. In the Heisenberg picture, the action lives in the observables, in what can be asked of the world and what can be answered. There is no wave doing mysterious things. There is a local algebra of askable questions, evolving according to definite rules.

Deutsch and his collaborators have argued for decades that the Heisenberg picture lets us understand quantum information locally and clearly in ways the wave function picture does not. They have been mostly ignored, because the Schrödinger picture is easier to compute with and the calculations are what get you tenure.

THE RETURN OF THE OBSERVER

Here is what I want to suggest. The information turn that Hassabis was reporting on is, structurally, a vindication of Deutsch’s argument. They are the same point, made from opposite directions, by people who have probably never had a serious conversation with each other. And neither of them has quite said it.

I think this is about to change. A hundred years after the famous Solvay Conference, the most intellectual picture ever taken, and Heisenberg’s move to Leipzig, I believe 2027 will mark a centennial switch in how we look at reality, as theory finally unites with practical implication, namely a deeper understanding of what reality is.

To see why, you have to ask what it means to say that information is fundamental.

Information, in any operational sense, is the answer to a question. A bit is the answer to a yes-or-no question. A pattern is the answer to a structure question. The “amount of information” in a system is, formally, a measure of how many distinct questions you would have to ask to fully characterize it. There is no information without questions. There is no information without the structure of asking-and-answering.

Now ask: which formalism of quantum mechanics is naturally a formalism of asking-and-answering? Not the Schrödinger picture. The Schrödinger picture has a wave, and the wave is a thing, and the thing does things. You are still in the world of stuff. The wave is a piece of metaphysical furniture sitting in some abstract space, evolving on its own according to its own equation, and the act of measurement is something that happens to it from outside. The whole picture is substance-first, with information as an afterthought you compute by squaring amplitudes.

The Heisenberg picture is the other way around. The fundamental objects are observables, the things you can ask the world. The state of the system is a fixed reference; the dynamics live in how the observables evolve. Physics, in the Heisenberg picture, is the dynamics of the questions you can ask and the answers you can receive. Information is not something you compute from the theory. Information is what the theory is of.

If Hassabis is right that information is more fundamental than matter or energy, then the formalism of physics that takes this most naturally is the formalism Deutsch has been defending for forty years and that the field has been quietly ignoring because Schrödinger’s wave is easier to draw. The information turn in AI is not a separate development from the Heisenberg-picture argument in foundations. They are the same convergence, arriving by different roads.

This is the sentence Hassabis did not finish. Information is more fundamental than matter or energy. Full stop, on the venture capital stage. The next sentence, the one he did not say but that follows from his own claim, is: and therefore the formalism of physics that has been treated as a kind of stylistic preference for the last hundred years is, in fact, the more revealing one.

That is a much bigger claim, and it has consequences.

THE SINGULARITY PARADOX & THE QUANTUM MEMORY MATRIX

My co-author, Dr. Florian Neukart, has developed a theory called the Quantum Memory Matrix. The theory says, in plain language, that spacetime at its smallest scale is not a smooth fabric but a lattice of tiny cells, each of which is a small quantum register that records the interactions passing through it. Every photon scattering, every electron exchange, every gravitational ripple leaves an imprint in the cells where it occurs. Geometry is what you see when you stand back far enough that the imprints blur into a surface. Time is the rate at which the lattice fills up.

This sounds wild. It is, on closer inspection, conservative. It does not require new fundamental forces or hidden variables. It builds on two things physicists already accept, the holographic limit, which says every region of space has a finite information capacity, and the unitarity of quantum mechanics, which says information is never destroyed, and asks what kind of architecture would naturally realize them. The holographic bound says: every region has a finite memory budget. The QMM proposes one architecture in which that budget is used, Planck-scale cells doing the storing. It is not the only possible architecture, but it is one whose imprint-and-retrieve primitive has now been tested in hardware.

Florian’s group has also reported the first hardware test. They implemented small versions of the QMM cells on IBM quantum processors, ran a sequence of imprint-and-retrieve operations, and measured the fidelity. In the cleanest baseline configuration, fidelities reached approximately 77 percent, meaning that most of the information written into the memory cell came back faithfully. The remaining loss was consistent with ordinary gate noise and decoherence, the kind of error that improves with each generation of hardware. The result is a proof of principle, not a confirmation of cosmology, and the authors are careful to frame it that way. But it does establish, for the first time, that the imprint-and-retrieve cycle the QMM hypothesis requires can be built on present-day hardware, and behaves the way the hypothesis predicts. The slogan has become a research program.

I do not want to overstate this. The experiment was small. A handful of qubits in a refrigerator, not the lattice of the universe. There is a long road from a working primitive to a confirmation of cosmology. But I do not want to understate it either. After a century in which “information is fundamental” was a slogan that could be neither tested nor refuted, there is a number on a screen. The number says the architecture is at least possible to build. And this is why I believe 2027 will be a breakthrough year.

The architecture, formally and obviously, is a Heisenberg-picture architecture. The QMM is a theory of local operators acting on local cells. The dynamics live in the operators. There is no global wave function doing mysterious things across cosmic distances. There is a local algebra of imprints, accumulating cell by cell as time passes. This is exactly what Heisenberg would have written down if he had had a hundred years and modern quantum field theory and a small experimental quantum computer. The QMM is the kind of theory the Heisenberg picture has been waiting for. The IBM result is experimental support, a working proof of principle, that this kind of architecture is buildable. Hassabis’s claim about information is the methodological vindication, from the AI side, that this is the right way to think about reality.

THREE THREADS. ONE CLOTH. WILL IT NOW BE WOVEN?

Let me say what this means, because the implications are larger than they look.

If the substrate of physics is informational, and the substrate of mind has, since Turing, also been understood as informational, then the category distinction we have been drawing for centuries between mind and matter is not a fundamental distinction. It is a level-of-description distinction. Both are made of the same kind of stuff. Both are patterns in the same underlying ledger. This does not mean current AI systems are conscious. It does not mean they will be conscious soon. It means that the question of how mind fits into the physical world, the question that has tied philosophers in knots for four hundred years, has a different shape than we have been assuming. It is not a question about how to bridge two different kinds of substance. It is a question about which patterns in a single informational substrate count as the kind of thing we recognize as mind. This is a much more tractable question. We do not have an answer yet. But we know what kind of answer it would be.

If the substrate of physics is informational, then the deepest-running argument in AI right now, the argument about whether large language models are on the path to genuine intelligence, looks different. Hassabis says yes, AGI by 2030. Deutsch says no, current systems do Edison’s ninety-nine percent perspiration but not the one percent inspiration, the genuine creative leap that introduces something new into the world. They are both partly right, and the way to see they are both partly right is to notice that the underlying question, what kind of pattern can host genuine novelty, what kind of process can generate real explanation rather than sophisticated recombination, is a question about the structure of the informational substrate, not a question about whether silicon is the right kind of stuff. The substrate is the same. The question is whether the patterns we are currently building have the right structure. That is an empirical question, and it is the right kind of question to be having.

If the substrate of physics is informational, then the project of understanding AI and the project of understanding physics are, at the deepest level, the same project. Both are asking what the world is doing when it is doing the things we recognize as intelligent, or alive, or real. The traditional academic division, physicists in one building, computer scientists in another, has hidden this from us. The QMM and the success of modern AI suggest that the division is artificial. Both fields are converging on the same answer.

There is a deeper point I want to land on, because it is the point that makes the whole convergence feel less like a clever observation and more like the beginning of something.

For a hundred years, the question of what is fundamental, what the world is made of, has been treated as a question for theoretical physicists, to be settled by clever experiments at increasingly extreme energies and scales. We built the LHC. We are building bigger telescopes. We are looking for new particles, new symmetries, new fields. The implicit assumption has been that the answer to “what is the world made of” will be found by going deeper into the substance, by dividing matter into smaller and smaller pieces until we hit bedrock.

What the convergence I am describing suggests is that we have been looking in the wrong direction. The bedrock is not deeper into the substance. The bedrock is in the structure, in the patterns, in the algebra of askable questions, in the informational architecture that the substance is, in some sense, an expression of. We will not find what is fundamental by smashing protons harder. We will find it by understanding what kind of pattern the world is, and by building tools that read the pattern back to us.

The QMM is one such tool. It reads the pattern at the substrate level, in the structure of spacetime cells. AI, increasingly, is another. It reads the pattern at every other level, proteins, languages, games, weather, the long correlations that make a system behave the way it does. Both are doing the same thing. Both are pointing, in their respective domains, toward the same conclusion: that the pattern is what is real.

Hassabis was right. He was righter than he said. The sentence he did not finish is the one that connects his casual remark to the deepest-running fight in foundational physics, to the experimental confirmation that has just arrived from a quantum computer in upstate New York, to the formalism of quantum mechanics that we have been quietly ignoring for a century, and to the question of what the AI we are building is actually for. What it is for, if any of this is right, is to read the ledger.

So the story goes: the universe has been keeping it for thirteen point eight billion years. We are only now learning the language. Will we now unify mythos and enlightenment? Science and philosophy?

If you want the longer argument, the one that lays out the Heisenberg-picture case in full, walks through the QMM in detail, and traces the convergence chapter by chapter, Florian Neukart and I are working on our third book, titled Anamnesis: Why the Universe Was Conscious Before We Were. This essay is a postcard from inside it.