Stephen Wolfram on the Wolfram Physics Project: Hypergraphs, Computational Irreducibility and a Theory of Everything

High-Level Overview

In this second conversation with Lex Fridman, recorded outdoors at dusk as an experiment in pandemic-era podcasting, computer scientist, physicist and Wolfram Research founder Stephen Wolfram walks through the Wolfram Physics Project: his attempt to find a single simple rule, acting on a network of "atoms of space," from which space, time, gravity, quantum mechanics and everything else could emerge. Along the way the two cover computational irreducibility and why science can only predict certain things, what the COVID-19 pandemic revealed about people's expectations of science, how special and general relativity and quantum mechanics seem to fall out of the model, the "specs" of the computer running the universe, quantum computing, metamathematics, alien intelligence, warp drives, and the question of why the universe exists at all.

The tone is exuberant and wide-ranging. Wolfram is visibly excited, repeatedly describing results his team had found only days or weeks earlier, and Lex plays the curious computer scientist who keeps asking for intuition, analogies and advice for newcomers. The conversation swings from history and philosophy of science into dense technical physics, then back out to big-picture reflections on meaning, humility and what makes humans special.

The Discussion Explores:

• How scientific revolutions happen through new methods that open up a burst of "low-hanging fruit"
• Computational irreducibility, the principle of computational equivalence, and the "pockets of reducibility" where prediction is possible
• The pandemic as a case study in the limits of what science can deliver
• The Wolfram model: space as a hypergraph, time as the repeated application of a rule, and the causal graph of events
• How causal invariance leads to special relativity, and how curvature and energy in the hypergraph lead to Einstein's equations
• Quantum mechanics as multiway graphs and "branchial space," and the claim that general relativity and quantum mechanics are the same theory
• Quantum computing, rulial space, alien descriptions of physics, and the physics of metamathematics
• Beauty in physics, the physics community's reaction, and what a fundamental theory might mean for technology and for us

Deep Dive into Main Ideas

Revolutions in Science and the Legacy of A New Kind of Science

Lex opens by asking about the key breakthrough moments in physics. Wolfram points to the invention of quantum mechanics in the 1920s and to the 1970s, when quantum field theory and QCD (the theory of quarks and gluons) took off. He was a teenager deeply involved in physics then, and says figures like Richard Feynman, Murray Gell-Mann and the Nobel-winning QCD physicists seem to him like "everyday characters" rather than icons. His general pattern: history looks instantaneous from afar, but there is usually a long buildup, then a methodological advance that unlocks a flurry of easy discoveries lasting five or ten years. He sees the same thing in deep learning, which "started working" around 2011 and 2012, and Lex points to AlexNet and ImageNet as its trigger.

Wolfram describes his own path: in the early 1980s he began asking whether programs, rather than equations, could model the world; around 1984 he realized that very simple programs can produce great complexity; later systematic experiments showed the phenomenon was general, leading to his principle of computational equivalence. He had early ideas about fundamental physics in the 1990s and wrote them up in 2002. Asked whether A New Kind of Science had the revolutionary impact Lex expected, Wolfram says it did, just without the heroism: for 300 years people modeled the world with equations, but in the last 15 years new models are most often made with programs. He is disappointed, though, that the pure study of the abstract "computational universe" has not yet become a big field that lots of people pursue.

Computational Irreducibility and Pockets of Reducibility

The central idea that runs through the whole conversation is computational irreducibility: even if you know the rules of a system exactly, you often cannot jump ahead and predict what it will do; you simply have to run it. Wolfram links this to Gödel's theorem and Turing's halting problem, and explains his principle of computational equivalence: once a rule-following process is doing something not obviously simple, it is as computationally sophisticated as anything else, including our brains. That is why we cannot systematically outrun such systems. As he likes to say, "the computational animals are always smarter than you are."

Prediction is still possible in "pockets of reducibility," which he says must inevitably exist scattered through an irreducible world. Science has concentrated on those places where its methods work. His parable: ancient Babylonians tried to predict where the planets would be, what the weather would do, and who would win a battle, with no idea which was easiest. Planetary motion turned out to be reducible and was essentially cracked about 300 years ago; weather is roughly halfway; battles are not cracked at all. Engineering, he adds, is largely the art of building things that stay inside pockets of reducibility so life can be predictable, and there is an infinite supply of such pockets.

He even reframes the "heat death" of the universe: a gas in thermal equilibrium is full of elaborate computation among bouncing molecules. It only looks boring because our science has not found a way to tell an interesting story about it. The principle of computational equivalence, he says, carries both hope and humility: what we do is not fundamentally more sophisticated than what gas molecules or the weather do; our computations are special to us only because they connect to our purposes.

The Pandemic and What Science Can Deliver

Wolfram says that when the pandemic began he was about to release the physics project, but he felt he should do the public-service thing of looking at data and modeling. He concluded it is "just really hard": you need to know a great deal about human interactions, and it has become clear there was also much we did not know about viruses and immunity. Whenever a clean narrative emerges (for example about T cell immunity or asymptomatic spread), that is a sign of reducibility; working from first principles, you may simply be stuck simulating every step. He notes that ideas his team had worked out about foliations of causal graphs turned out to be relevant to contact tracing, and that the standard epidemic model was, he recently learned, invented by the grandfather of a high school friend. One of his children studied disease spread on different kinds of graphs and found only a few robust results, such as that a small number of big gatherings is worse than a large number of small gatherings; for more detailed questions, it "just depends."

What frustrates him most is the mismatch between what people expect science to deliver and what it actually can. He calls it "kind of scary actually how much people believe in science," since people want to be told definitively what to do even when the science cannot say. He also sees an upside: if science could always tell us the answer, living would be pointless. Irreducibility means something is actually being achieved as we live our lives. He calls himself an optimist, observing that organizations usually come out of a shock better unless it kills them, and mentions that as a remote tech CEO for 30 years, coming to meet Lex was the first time in six months he had been in a building other than his house. There is also a playful moment when Wolfram checks a Wolfram Alpha sunburn computation during the outdoor recording, calls it "a QA moment," and promises angry feedback if he gets burned.

What a Theory of Everything Means, and a Short History of Physics

For Wolfram, a fundamental theory of physics would be a rule, possibly quite simple, that if run long enough (perhaps 10 to the 500 steps) would reproduce everything in the universe, down to this very conversation. That reduces physics to mathematics, much like a definite procedure for generating the digits of pi. Irreducibility means such a rule would not let you shortcut to predicting everyday events, and he originally expected that the best he could do in his lifetime was to describe the first 10 to the minus 100 seconds of the universe. To his huge surprise, it turned out that certain pockets of reducibility occur generically across large classes of these rules, and those pockets are "basically the pillars of 20th century physics": general relativity and quantum mechanics. Even more striking to him, the two theories turn out to be, at some level, the same theory.

At Lex's request he gives a quick history. Ancient Greek philosophers tried to reason out how the world works; Galileo and Newton (whose Principia is about mathematical principles of natural philosophy) established that we can write down equations and check them against observation. Einstein's special relativity in 1905 was unusual because it was logically derived from the constancy of the speed of light, which forces time dilation and length contraction. General relativity in 1915 describes gravity as the curvature of space by mass, so that "straight lines" become curved, like great circles on Earth. Everything measured about gravity since then, including black hole collisions seen through gravitational waves, agrees with it. His lesson for theory builders: Einstein tried to patch his equations to avoid an expanding universe, and he should have trusted them.

Quantum mechanics grew out of the old question of whether the world is continuous or discrete. It gives extraordinarily accurate calculations, but what it means has always been hard; Wolfram recalls his friend Dick Feynman saying nobody understands quantum mechanics. Quantum field theory allows particles to be created and destroyed and is mathematically difficult; the Standard Model works very well but has unexplained parameters, such as why the muon is 206 times the mass of the electron. The problem with unifying the two is that the methods of quantum field theory break down when applied to gravity and gravitons. Wolfram now sees approaches like string theory and twistor theory as something like domain-specific languages describing the "machine code" his model provides, and he is struck that highly abstract mathematics such as higher category theory and infinity groupoids turns out to be anchored by his framework.

The Model: Atoms of Space, Hypergraphs and Time as Computation

Wolfram says the "meta idea of using simple computational systems is the right mountain; the specific model details are clean but not unique, since the same thing can be described with hypergraphs or higher category theory. The first key idea is that space is not continuous, as Euclid assumed when he defined a point as that which has no part. Instead space is made of discrete atoms of space, with no coordinates, only connections: all we have is the friend network, as Wolfram puts it, which Lex likens to a Facebook where we know friendships but not addresses. With perhaps 10 to the 100 such points, possibly around 10 to the minus 100 meters in size (compared with 10 to the minus 15 meters for a proton), large-scale behavior can look like continuous space, just as discrete pixels form a smooth image. The connections are best thought of as a hypergraph, where a single hyperedge" can link three or more points; realizing this, he says, removed arbitrariness he felt in earlier versions.

The second key idea is that there is nothing except space. Particles are particular structures in the hypergraph, perhaps "twisty knotted" features, and energy, mass and momentum are also features of the same hypergraph. By one rough estimate, everything we care about in the universe is only one part in 10 to the 120 of what is going on; the rest simply maintains the structure of space, and matter is "a tiny little piece of froth on the top of all this activity."

Time, in this picture, is the progression of computation. A rule says that whenever a small pattern of connections appears, it is replaced with another pattern. Each application is an "event," and events can happen anywhere, asynchronously, as long as their inputs exist. Those dependencies define the causal graph, the network of which events depend on which. Unlike relativity, where space and time are the same kind of thing, in Wolfram's model space is the extent of the hypergraph and time is the inexorable progress of the computation, and the familiar unity of spacetime only emerges at large scales.

Causal Invariance, Special Relativity and Einstein's Equations

The key property is causal invariance: in some cases it does not matter in which order you apply the updates, because you always get the same causal graph. Wolfram illustrates with multiplying out a series of polynomials in any order, or sorting a string of As and Bs by repeatedly flipping BA to AB. He also imagines a universe with a single update head running around like a Turing machine: since neither he nor Lex could know whether the other had been updated until they themselves were updated, the causal graph would be the same. Different ways of slicing the causal graph into moments of simultaneity correspond to Einstein's reference frames, and because the causal graph is invariant, special relativity follows. He worked this out in the 1990s.

To get general relativity he needs dimension and curvature. On a square grid, the number of points reachable in r steps grows like r squared; in general it grows like r to the power of the dimension, which lets you measure a hypergraph's effective dimension (possibly fractional, like 2.6). Curvature is a correction term, like how the area of a circle drawn on a sphere is slightly less than pi r squared. Rules with causal invariance, computational irreducibility and finite dimension give large-scale structure that obeys Einstein's equations. He even speculates the early universe may have been infinite dimensional and that there may be "dimension fluctuations" today, places that are 3.01 or 2.99 dimensional, which experiments might detect.

Energy has a precise definition in the model: formally the flux of causal edges through spacelike hypersurfaces, informally the amount of activity in the hypergraph. Activity that stays in place relates to energy and rest mass, activity that spreads relates to momentum, and E equals mc squared falls out. Adding how energy produces curvature yields the full Einstein equations. Because the formal limits are hard to prove (he notes that deriving fluid equations from molecules has never been done rigorously either), his team uses "proof by compilation": running the discrete model directly, as numerical relativity does, and checking that it matches what the Einstein equations predict for systems like black hole mergers.

The Specs of the Universe and Lessons for Parallel Computing

Answering a Reddit question about the specs of the computer running the universe, Wolfram explains that elementary time is arbitrary, but the ratio of elementary distance to elementary time is the speed of light. There are two other maximum speeds: a maximum entanglement speed in the space of quantum states, which might be measurable in black hole systems, and a maximum speed in rulial space. His rough, self-described rickety estimate is on the order of 10 to the 500 Wolfram Language operations per second, with perhaps 10 to the 400 atoms of space. He adds that reducibility goes deeper than he expected: studying quantum interference turns out not to require knowing what an electron is.

Lex, recalling his own struggles with concurrent programming in Java, sees causal invariance as a potential mathematical framework for massively distributed computation. Wolfram agrees, noting that eventual consistency in distributed databases is essentially the same idea, and predicts that one day programmers will talk about programming "in a certain reference frame," just as physicists choose reference frames. He admits he tried designing languages for parallel computation and graph rewriting in the 1980s but could not make them useful then.

Quantum Mechanics in Branchial Space

Because updates can happen in many orders, the model naturally produces a multiway graph of all possible histories, which branch and merge, much like a graph of every possible tic-tac-toe game where different move orders reach the same board. Wolfram says quantum mechanics is really two theories glued together: one for how amplitudes combine into probabilities, and one for measurement, which has always been mysterious. In his model, causal invariance means every branching is eventually matched by a merging, which relates to why definite things seem to happen. Slicing the multiway graph at a moment gives "branchial space," where the distance between states (how far back you must go to find a common ancestor) is essentially an entanglement distance.

Measurement corresponds to observers choosing "quantum observation frames," the analog of reference frames in branchial space, and because observers are computationally bounded, they are limited in the frames they can construct. He compares this to thermodynamics: we see entropy increase because we cannot track every molecule. The result he is most excited about is that energy (more precisely the Lagrangian density) deflects paths in branchial space in exactly the way that mass and energy deflect paths in physical space. The Feynman path integral describes the first and Einstein's equations the second, and mathematically they are the same. General relativity is motion in physical space, quantum mechanics is motion in branchial space, and "they are the same." He also mentions that the uncertainty principle comes out as a consequence of curvature in branchial space.

On the double-slit experiment, which his team had been working through in recent weeks, the destructive interference arises because the two paths through the slits wind up at opposite ends of branchial space and cannot merge into something measurable in physical space. He cautions there is some sleight of hand, since his team is still in a "vigorous battle" over how to coordinatize branchial space, which involves space-filling curves and Riemann surfaces. He adds that the project calls for mathematics that does not exist yet, such as a calculus for fractional and dynamically changing dimensions.

Quantum Computing, Spin and the Frontier

Wolfram worked with Feynman around the early 1980s on an attempt to build a quantum randomness chip, and they concluded that the quantum measurement process was a serious damper on what quantum mechanics could offer computing. In his model, the parallelism behind algorithms like Shor's seems to be lost again in the work needed to pull all the branches back into a classical answer, which may relate to decoherence. His team has a compiler that turns standard quantum gates into multiway systems and has "unimpressively" factored about two times three, roughly where physical quantum computers are. He still thinks quantum computing research is worthwhile for practical speedups; his one bizarre remaining hope for exponential speedups is mining the expansion of the universe in branchial space, analogous to connecting a spring between two receding black holes to harvest energy from cosmic expansion.

He describes the team as being in "harvesting mode," picking off known effects one by one. Near-term targets include why particle spin is quantized (a story involving homotopy groups in twistor space) and the difference between fermions and bosons. His current, subject-to-revision view is that bosons are associated with merging and fermions with branching in multiway graphs, so exclusion is things being forced apart in branchial space. He speculates that spinors, which take 720 degrees of rotation to return to their original state, may come down to the difference between directed and undirected hypergraphs. Lex mentions playing with Möbius strips, and Wolfram mentions the belt trick.

Reflecting on the project's history, Wolfram says he had graphs, multiway systems and general relativity by the late 1990s but was bothered by awkward constraints like requiring trivalent graphs. Hypergraphs, which he found less than two years before this conversation, made it elegant. The "real embarrassment," he says, is that the final structure is an idealized version of the same structure he has used to build computational languages for more than 40 years.

Getting Started with the Physics Project

For newcomers, Wolfram recommends starting with the project's visual summary picture, his original announcement blog post, and his long technical introduction, which he says has been accessible to people who understand computation but are not physicists. Jonathan Gorard, a key contributor whom Wolfram calls a strong mathematical physicist, has also written papers. Understanding why the model connects to physics (for example, what an energy-momentum tensor is) requires roughly graduate-level physics, achievable in a year or two of study but not a month.

Exploring the computational universe, though, is "very close to the surface": high school students can run experiments and discover things, all the code is on the project website, and you can start experiments "day one basically." The real skill is judgment about which experiment to do and how to interpret it. He hopes future physics textbooks will be simpler, because discrete models let you show a picture of what happens instead of arguing about limits.

Rulial Space, Aliens and the Physics of Mathematics

Why would our universe run on one particular simple rule? Wolfram's answer: imagine applying not one rule but all possible rules in all possible ways. This rulial multiway graph is itself causal invariant, so different ways of slicing it are equivalent. The rule we attribute to the universe reflects our own reference frame in rulial space. The one immovable fact is that the universe is computational, not a hypercomputer, and exactly at the level the principle of computational equivalence describes. He connects this to non-deterministic Turing machines and suggests a geometric view of the P versus NP problem. One striking consequence: aliens might not experience the same physics as us, but could have a description of the universe utterly incoherent with ours. Intelligence, he says, is everywhere; he is fond of saying "the weather has a mind of its own," and the difference is only whether we have a story connecting it to our way of thinking.

He then turns to metamathematics. The standard axioms of mathematics fit on about a page and a half; theorems are expressions reachable from the axioms, and a proof is a path through a multiway graph. Gödel's incompleteness corresponds to there being no upper bound on path length. Mapping Euclid's 10 axioms and 465 theorems, he found the hardest theorem is that there are five Platonic solids, at the end of a 33-step path. Math is hard because of computational irreducibility, but doable because human mathematicians build out along constructed paths, like Lewis and Clark, rather than parachuting in at random, where you quickly get lost. He believes the univalence axiom of homotopy type theory is equivalent to causal invariance, that the limit of "paths between paths" is Grothendieck's infinity groupoid, that non-constructive proofs may relate to event horizons, and that critical pair lemmas in automated theorem proving are precisely branch pairs. He also sees a connection to blockchains, where not every transaction needs immediate global reconciliation, much like light cones.

Beauty, the Physics Community and Why It Matters

Lex brings up Sabine Hossenfelder's argument that the pursuit of beauty leads physics astray. Wolfram says he cannot speak to exactly what she wrote, but that before this project he would have said we do not know how complicated the universe's rule is; the one thing we know is that there is order in the universe. If the rule is simple, it will almost inevitably seem beautiful because so much must fit into so little. He recounts that physics was only about 100 pages of his 1,200-page A New Kind of Science, meant as a use case, yet fundamental physicists responded with "heavy pitchforking," and he stopped working on it around 2004 because he did not want to build "a product for a target market that doesn't want the product." Since then, he says, the hope that string theory would deliver has faded and quantum information has made computational ideas more welcome. On Eric Weinstein, he says his work is a nice specific piece of mathematical physics in the tradition of gauge theories, and that his own team is close to understanding local gauge invariance, possibly including why the Standard Model's symmetry groups sit inside E8. Historically, he observes, people have repeatedly thought they were almost done and then given up; the fundamental theory became "almost a footnote," like machine code beneath the high-level languages everyday physics uses.

Why should anyone care? His analogy is Copernicus: his technical impact on astronomy touched few people, but the philosophical shift that math and science can overrule common sense touched everyone. Grounding the world as computational would make ideas like computational irreducibility (a term he was proud to see enter the Congressional Record) foundational rather than optional. On engineering, he had agreed to speak at a NASA workshop on faster-than-light travel and floats ideas like "space tunnels" and a "vacuum cleaner" that clears out the activity maintaining space, creating negative energy density and repulsive gravity. His immediate instinct is that it cannot be done, but he recalls people once saying data could never travel fast over copper wire.

Finally, on why the universe exists at all, Wolfram guesses the question is like Gödel's second incompleteness theorem: from within the universe, its existence is essentially undecidable. He admits it is strange to think "it's all hypergraphs down there," but argues that because all intelligence is computationally equivalent, any superintelligence will not be fundamentally more meaningful than us; what makes us special is simply the details of us. The conversation ends as darkness falls and mosquitoes arrive, with Lex closing on Feynman's words: "physics isn't the most important thing love is."

Conclusion

This conversation is a tour of Stephen Wolfram's bold claim that the universe is a simple computational rule rewriting a vast hypergraph, and that the pillars of 20th century physics are exactly the parts of that computation we can predict. Computational irreducibility is the thread tying everything together: it explains why science can predict planets but not pandemics, why mathematics is hard yet doable, why life has meaning, and why the universe's existence may be undecidable from within. Wolfram presents relativity and quantum mechanics as motion in physical space and in branchial space governed by the same mathematics, while frankly flagging open problems, rough estimates, and results only days old. Whether or not the model proves right, the episode rewards anyone curious about physics, computation or the philosophy of science, and Wolfram's message to newcomers is encouraging: you can start running experiments on these systems on day one.

This is a summary of "#124 – Stephen Wolfram: Fundamental Theory of Physics, Life, and the Universe" from Lex Fridman Podcast, giving you the GIST so you can decide if you want the REST. Listen to the full episode.