Consciousness Is a Step Down From Intelligence

Almost twenty years after *A New Kind of Science*, Lex Fridman asks Stephen Wolfram what complexity is. Wolfram declines the definition. It is not the question he cares about, he says, any more than a biologist's most urgent question is what life is. What pulled him in, embarrassingly close to fifty years ago, was how complexity arises: why snowflakes branch, why galaxies take complicated shapes, why living systems get produced at all. Around 1980 he still believed particle physics and fancy mathematical physics should be enough to answer that. They were not.

Engineered objects, he noticed, tend to be circles and lines. The natural world is full of forms that look nothing like that. The puzzle was what secret nature has that ordinary engineering does not. After several years building SMP, the Symbolic Manipulation Program that preceded Mathematica and Wolfram Language, programs felt like the obvious raw material for models. Instead of writing a long purposeful codebase, he went out into the computational universe of the simplest possible programs and asked what they actually do.

Cellular automata looked too trivial to be interesting: a line of black-and-white cells, each updated from itself and its two neighbors. Then he ran the experiment, which originally took a few hours, and the printouts were not what he expected. Even a one-line rule could generate extremely complicated behavior. His favorite example is Rule 30, which, started from a single black cell, produces a pattern that looks essentially random. That, he says, is likely nature's secret: in the space of possible programs, simple programs with complex behavior are easy to find. People assume that if there is a rule you can see the rule, the way early movie robots give themselves away by doing simple things. The computational universe shows that intuition is false. Informally, complexity is just that you cannot easily tell what a system is going to do.

The phenomenon around that discovery is computational irreducibility. Ask Rule 30 what it will look like after a million steps and you can run a million steps; you cannot compress the process and jump to the answer for free. Wolfram has offered a small prize for results about the center column. Even proving that the column never repeats looks doable to him, analogous to the nineteenth-century proof that π is irrational and so its digits never cycle. Whether zeros and ones appear equally often, like whether π's digits are equidistributed, is still far beyond current mathematics. People have been picking away at the challenge. It is hard.

Fridman asks whether complexity might still be filtered thermodynamic randomness rather than generated from simple rules. Wolfram's answer is that once you put randomness in, you no longer know how much of the output was already in the input. Chaos theory's shift map just slides binary digits to the left; if the initial number is random, you only excavate what you put in. If fluid mechanics worked that way, the continuum equations would be wrong, because the millionth digit of the initial conditions would matter far below molecular scales. The clean systems are the ones with obviously simple initial conditions that still look random. If a complex pattern is reproducible, it was not filtered from the outside world.

At the level of the universe, Wolfram says an extra die being thrown is unnecessary. Current fundamental physics can get what we see without it. You could probably add such a feature without breaking the universe, but it is not required for understanding. He has been surprised to find that the question of why the universe exists even looks addressable from the physics project, a metaphysical puzzle that most philosophers have stayed away from. His not-yet-fully-worked suspicion is that a genuine explanation will show there cannot, in any meaningful sense, be randomness underneath, and that if there is, it is irrelevant to our perception of what is going on.

Fridman pulls the conversation back to the Wolfram Physics Project, then about a year and a half old. Wolfram calls the aftermath a Cambrian explosion. The physics model turned out to sit on a more general multicomputational model, applicable also to economics, linguistics, and immunology. Because physics is so successful, the same skeleton would let physicists' intuitions travel into those fields. Computational irreducibility sits in the middle of science's claim to have nailed everything down. We operate, he says, in slices of reducibility that cut through that ocean, and those slices are what look like physics-like laws.

The starting claim is that space is made of something. Since Euclid, space has been a backdrop where you can put points wherever you like. Here, space is like water: it has atoms of space, distinguished only by identity — you could give them UUIDs — and by how they relate. Three atoms associated with each other are a hyperedge; the universe is the giant friend network of those atoms, perhaps 10^400 of them. Electrons and photons are tangles in that structure, like vortices that move through a fluid while remaining themselves. There is a bottom length, perhaps 10^-100 meters, far below the proton at 10^-15 or what accelerators have seen at around 10^-21.

The usual Planck length is about 10^-34 meters, but the model has an extra parameter: roughly 10^170 simultaneous threads of execution, the independent quantum processes of the universe. That factor also shrinks the oddly large Planck energy, comparable to a lightning strike, down to a tiny elementary energy. Wolfram calls the numerical estimate somewhat rickety. A "gravitational microscope" might be a rapidly rotating black hole near the critical spin, where spacetime is about to fall apart and gravitational waves could betray discreteness. Fluid dynamics already has the analog: at hypersonic Mach 20, it matters that a space plane is hit by individual molecules, not a continuous fluid.

Space is connections. Time, in this story, is the computational process that updates those connections. A rule rewrites little pieces of the hypergraph whenever they match, anywhere, asynchronously. There is no lockstep global clock, so there is no single thread of time in the universe, only many possible orders. Nothing is permanent: if the hypergraph stopped being rewritten, space would fall apart. When Wolfram moves his hand, it is recreated in a different set of atoms of space, the way a wave in the ocean is not the same molecules. Pure motion, the possibility that an object can move and remain itself, is not self-evident. It is part of how we parse the universe.

Trained physicists treat space and time as the same kind of thing, related by the Poincaré group and Lorentz transformations. In this model they are fundamentally different, and relativity still falls out. An observer embedded in the system cannot see which update happened first from a God's-eye view. All they can know is the causal graph of elementary events: a few hyperedges consumed, a few produced. Causal invariance means different update orders produce the same causal graph, which is why we perceive spacetime the way relativity says we should.

Could the universe be a Turing machine with a single head racing around? It sounds absurd — Fridman is updating, Wolfram is updating — but you would never know. Until the head visits you, you are frozen, and you have no idea what happened to anyone else. Unravel that argument and the only accessible structure is the network of causal relationships. Any log you keep is itself part of the universe. You never get to step outside.

High-school physics gives a definite path for a thrown ball. Quantum mechanics replaces that with a bundle of paths and probabilities. In the hypergraph there is no single update sequence, so quantum mechanics is not a plug-in add-on: different orders are different quantum histories. Histories branch, and they also merge, two states sharing a next state, which is nontrivial because of graph isomorphism. Causal invariance lives in that merging. The surprise is not that quantum mechanics appears. The surprise is that you cannot get away from it.

Why don't we notice the universe branching and merging? Because our brains are branching and merging too. Quantum mechanics becomes the story of how a branching brain perceives a branching universe. Insisting that definite things happen is a way of knitting many threads of history together and calling them equivalent. That identification is not obviously consistent. The fact that it mostly holds is causal invariance; the little ways it fails are ordinary quantum effects. Most of the time the world looks like a continuous fluid. Only in the right experiment do you see the molecules that might have gone this way or that.

We parse a universe of perhaps 10^400 busy atoms of space the way we parse air in a room. We do not track 10^20 molecules; we notice density and the gas laws. We are computationally bounded observers. Cameras and pressure sensors extend the brain but are not qualitatively different senses. That boundedness, plus a single thread of experience, is about to become Wolfram's definition of consciousness. The 100 billion neurons in a brain fire in all kinds of ways, yet conscious experience is one thread of time. That is not obvious, and he now thinks it is why we conclude the universe has the laws it has.

For years Wolfram told people he had nothing to say about consciousness. Abstract intelligence, he now thinks, is just computational sophistication: anything that can do sophisticated computation is intelligent in the generalized sense. Consciousness is a step down, not the cherry on top of the cake. It adds two constraints. We take a computationally bounded view of the universe, and we insist on a single thread of time. Brains go to some trouble, neurophysiologically, to present one stream of attention. A clock going tick-tock satisfies both locks. What we add on top is irreducible computational ability.

Those two locks, he argues, are why we deduce the laws of physics we do. Assume computational boundedness and a single thread, and general relativity is what you parse out of the hypergraph, the pocket of reducibility you slice from the ocean. We treat space as having a definite structure at a moment, then another at the next moment, because light crosses a room much faster than our brains notice. A being the size of a planet would not live in that reference frame. A photon from the cosmic microwave background, for whom time stood still from emission until detection, would tell a still more boring story. Consciousness of our kind is scale-dependent.

Fridman worries this makes consciousness a weird side effect. Wolfram agrees that our kind is special to our scale, and he has been asking science-fiction-writer friends what other factorings of existence would feel like. The weather "has a mind of its own"; a pulsar magnetosphere looks like an agent. If the weather is a mind, he does not know what its physics would be, because it is extremely parallel and it carves up the universe differently than we do. Purpose talk collapses quickly: yes, the weather was trying to move hot air from the equator to the poles. We have no other intelligences to compare ourselves with except ones very much like us.

Rudy Rucker's novel saves the Earth, after nanobots consume it in chapter one, with the Principle of Computational Equivalence, which thrilled Wolfram because it is his core scientific idea. The principle: as soon as a system's behavior is not obviously simple, the computation it performs is as sophisticated as anything. Rule 30, brains, and the rest of physics are ultimately equivalent. That is why we cannot jump ahead of Rule 30. We are running the same grade of computation. In the science-fiction version, you do not need a planet of custom micro-servers. Nature is already computing.

A disappointing corollary is uploading. You might build an elaborate digital environment, then notice the electrons moving around are doing what electrons in a rock do. Computational equivalence says there is, at that level, no mathematical distinction between the rock that is the future of human consciousness and a rock that is just a rock. Consciousness, Wolfram now thinks, is the piece that is more special, though he has not fully teased it apart. Physics, in this framing, is exhibit one of alien intelligence: an alien process we try to map onto a story we can tell ourselves. Wolfram Language is his long attempt at the same bridge, from raw computation to what humans can conceptualize.

The same principle rearranges the search for extraterrestrials. We have assumed they share our physics because we share a universe. They might instead parse the universe differently. For all we know, in the details of the gas molecules in this room there is an astonishing intelligence and a huge culture, invisible because we are slicing the world the wrong way. "If only we could parse the universe in the right way," Wolfram says, that amazing thing "would be obvious to us." He has even been trying to imagine an experiment: what would it feel like to be a piece of a cellular automaton, and can neural nets get us close enough to look from the inside?

Wolfram offers a mechanical story of relativistic time dilation that he finds genuinely exciting. A clock flown at a large fraction of the speed of light ticks slowly for the stay-at-home observer; the twin who visits Alpha Centauri ages less. In the model, to move is to recreate yourself at a different place in space, and to sit still is also to update yourself according to the rules. The universe computes at a certain rate. You can spend that computation on successive ticks, or on rebuilding yourself as you move. The more updating you do because of motion, the less you can update the clock. Time dilation is no longer just a fact about the Lorentz transformation. It is a budget.

Physical space has general relativity as its bulk theory. Branchial space, the space of quantum branches in the multiway graph, has the Feynman path integral, which tells you how things move among quantum phases. The analog of the speed of light there is a maximum entanglement speed: how far an effect can travel through the space of quantum states in a given time. One measured number, he says, would fix the elementary length and leave the theory with no free parameters. Large black-hole mergers, or even many-body atomic physics via the quantum Zeno effect, are places he has been poking for a signal. Chemistry's stubborn definiteness, carbon atoms that "just end up" in a place, is another hint that environments keep making measurements and knitting histories together.

The cosmological prediction he is most eager for is leftover dimension fluctuation. The hypergraph has no built-in dimension; three dimensions are an approximation, as when a ball of radius R contains about R^3 nodes. The very early universe may have been essentially infinite-dimensional and then cooled toward three, leaving patches of dimension 3.01 at the time of the cosmic microwave background. Photons propagating through such a space would not obey the inverse-square law, because the surface of a sphere would no longer grow as R^2. Inflation, which Wolfram and a collaborator had sketched in a 1979 footnote and never quite believed, becomes less necessary: a highly connected early hypergraph already explains uniformity. Seeing dimension fluctuation, he says, would be a huge flag for the model. Someone still has to invent 2.5-variable calculus to compute it.

Physical space is the extent of the hypergraph. Branchial space is motion between histories. Rulial space is a level up: the space of rules. The puzzle that had confused Wolfram was, if we hold the rule for the universe in our hand, why that rule and not another? Copernicus taught that we are not special. Why universe 312 rather than universe quadrillion-quadrillion-quadrillion? His resolution is that the universe is applying all possible rules at all possible places, not just one rule everywhere.

The resulting object he wants to call the ruliad: the limit of all possible rules applied in all possible ways. It is not the case that everything happens, because different rules can land on the same state; the structure is entangled. It need not be finite. Time can keep spewing structure forever. It has three limits at once: how long you run, how many rules you apply, how many initial states you start from. Mathematically it is a horrendous object. It appears related to Grothendieck's ∞-groupoid, though Wolfram says he does not yet fully understand the correspondence.

We perceive a particular rule the way we perceive a particular place. We do not live on Alpha Centauri; we also occupy one patch of rulial space, defined by our sensory apparatus and our way of parsing. Travel in rulial space would mean successively interpreting the universe differently. An alien intelligence might live here on Earth at a different rulial coordinate, with a perception so different it is "bizarrely unimaginably completely different." That, he says, has given him more respect for traditions of interpretation he does not himself understand. Even the immortal soul, mocked in his youth with "how much does it weigh?", looks less crazy once what matters in a brain is an abstract computational pattern rather than the matter.

The ruliad is the working-out of all possible formal systems, specifically all computational rules, not hypercomputation. Turing knew you could imagine an oracle box that solves the halting problem in one press. A "hyper-ruliad" containing those oracles would sit behind a cosmological event horizon relative to ours; the two could not communicate. Wolfram adds, as a footnote to the footnote, that he has not thought the hyper-ruliad through properly. The point of the ordinary ruliad is that no choice of universe is being made. Every formal system is already in it.

Why does 2+2=4 exist? It is necessary given the meaning of the terms. The ruliad, he argues, is the same kind of object: not made of something, just the necessary collection of formal relations. Embedded in that structure, an observer perceives tables and chairs and physics. Elephants are contingent products of biological evolution; plus is not. Fridman keeps reaching for an outside. Wolfram's reply is that once you have all possible formal systems, there is nothing left to imagine that still counts as a computation. Talking it through on the podcast, he admits he may still be missing pieces of the argument. People have been less interested in this question than he expected.

A weird consequence: if the universe exists, mathematics must exist too. Wolfram had long thought mathematics was arbitrary axioms inherited from the Babylonians. His wife, a mathematician, told him for twenty-five years he was wrong, and he now thinks she was more right. Axioms are the molecular dynamics of mathematics. Gödel's theorem is computational irreducibility, the explosion of Hilbert's dream of a machine that grinds out every theorem. Mathematicians rarely hit undecidability because they are computationally bounded observers with a single thread of time, reading fluid-like laws off metamathematical space the way we read fluid mechanics off molecular chaos. Pythagoras's theorem can be proved from different axiom systems, the way air and water both have fluid dynamics.

A proof is a path in metamathematical space. A bundle of proofs of the same theorem is the analog of a quantum path integral, an idea Wolfram says is only a couple of days old. Two proofs that are incompatible at an intermediate stage would be like destructive interference. Jonathan Gorard already has a heuristic for automated theorem provers inspired by the physics project: energy in the models is activity in the hypergraph, and that can guide which branch to explore. A black hole, in the discrete models, is simply a place where no rule applies anymore and time stops, the analog of a decidable theory such as Boolean algebra, where proofs have bounded length. Arithmetic and set theory, with proofs of arbitrary length, are the theories in which you can wander forever.

The literature holds about three million published theorems, like cities on the Earth; the question is the underlying continuum. Time dilation in metamathematics would be the cost of translating a theorem as you move from algebra to geometry. You keep spending steps mapping back to where you started, so the proof runs slower. Rulial space is messier than Hilbert-like branchial space, closer to that ∞-groupoid. Whether proof space has holes, whether two proofs of the same result cannot be continuously deformed into each other, is something Wolfram hoped to test in toy theories within days of the conversation. Category theory's higher morphisms, mappings between mappings between proofs, are the natural language for that limit.

He frames the whole turn as a fourth epoch of models. Antiquity asked what the universe is made of, with no real dynamics. The 1600s wrote equations and treated time as a slider. The 1980s computational paradigm gave a single thread of irreducible rule-running. Multicomputation adds distributed asynchronous threads of time. After a thousand steps it is not even clear what a thousand steps means, because there is no one state. An embedded, bounded, sequential observer sampling that mess, he argues, inevitably sees laws structured like general relativity and quantum mechanics, which in the physics model are the same law. The 1980s computational story felt like a downer, because so much could not be said. The observer's blindness is what makes simple laws possible again.

Chemistry, viewed abstractly, is graph rewriting: molecules as graphs of bonds, reactions as events that consume tokens and produce tokens. Synthesis is theorem proving. Wolfram suspects biology may store information not only in DNA sequences but in the dynamics of reaction networks, the analog of a chemical observer who cares about the causal graph, not just concentrations. The big news of 1953 was that a molecule can store information. The next qualitative leap might be that a dance of reactions can store it too. Molecular computation, on this view, is not mainly the old nanotech dream of a constructor that builds any shape. The one working example we have is us, and what may matter is the dynamic process, not which chemicals you make. His personal test is almost comic: compute the primes with molecules in an automated lab, and read them off as stripes in a gel.

Immune memory is the present-day test case. The standard clonal-selection story says we randomly produce antibodies, on the order of ten billion kinds, plus a trillion kinds of T-cell receptors, and amplify whichever binds an antigen. In the 1960s Niels Jerne, whom Wolfram also credits with monoclonal antibodies, had a network theory: antibodies to antibodies to antibodies, a dynamic web. Wolfram has been disappointed to find that theoretical immunology seems to have shrunk even as the experimental books got thicker, and that the most advanced theory he hit was a cellular automaton he at least understands. If memory lives in the network, shape space would have a speed of light: how fast a vaccine's poke spreads to nearby antibodies. General relativity might be in there and still be irrelevant to whether the human gets sick. Biology, he notes, has almost no global theories besides Darwinian evolution and digital DNA.

The same picture, applied to ledgers, is what he calls a quantum analog of money. Instead of one sequential blockchain, many ledgers become consistent the way causal invariance knits space, perhaps after a delay. Insist on knowing your bank balance right now and there is uncertainty; wait and it settles. Transactions are events, agents are atoms of economics, and dollar values are how an "economic observer" coarse-grains an intractable web. Wolfram is not yet happy with the prototypes, because he thinks he still needs a foundational theory of economics. The Cardano collaboration with Charles Hoskinson is more pragmatic: Wolfram|Alpha as an oracle for smart contracts, computational language for blockchain analytics, cellular-automaton NFTs as a bet on permanence, and a long-term hope for a symbolic discourse language that could write computational contracts and even an AI constitution. Proving a blockchain correct, he warns, collides with computational irreducibility. If you want serious computation, you cannot box it in.

He closes on the sociology of his own ideas. Complexity as a field, which he helped seed at the first gathering of what became the Santa Fe Institute, now has on the order of a thousand institutes and forty journals. He rolls his eyes at the mistake of treating "my model makes complexity" as evidence, because complexity is generic. What he wants instead are two bedrock activities: metamodeling, asking what Turing machine or automaton sits under the hedgehog model, and ruliology, the natural history of simple rules in the wild. Rule 184 is a minimal model of traffic; the catalog of such rules is timeless in the way an ancient Egyptian icosahedron is still an icosahedron. Applications, he now thinks, sucked the pure science away in the 1980s. He wants a home for it, perhaps a ruliological society, before the next generation of applications in biology, economics, or elsewhere does it again.