Hawking's Co-Author on Why Reductionism Is Dead

ELI5/TLDR

George Ellis, the cosmologist who co-wrote The Large Scale Structure of Space-Time with Stephen Hawking, has spent the second half of his career arguing that physics is not in charge. Electrons don’t decide what a computer does — the algorithm does. Molecules don’t decide where a pigeon lands — the pigeon does. The physics is the servant, the context is the master. From this one stubborn idea he gets: top-down causation, free will, an “evolving” block universe, a rejection of the multiverse, and a final claim that evil is a fact about the universe, not a fashion.

The Full Story

The claim, stated plainly

Most physicists have a house religion called reductionism. It goes: there are particles, they obey laws, everything else is bookkeeping. Ellis thinks this is obviously wrong, and he spends most of the conversation sharpening why with examples rather than slogans.

His opening move is a computer. Imagine you write some Python. The Python gets translated down into machine code, which flips transistors on and off, which moves electrons around. A reductionist would say: see, it’s just electrons in the end, the Python is shorthand. Ellis says: then hand me Maxwell’s equations and Newton’s laws and show me how the electrons decided to run a browser. You can’t, because the electrons didn’t decide anything. The algorithm decided. The physics enabled the outcome, but the outcome was set from above — from the code — and pushed down. That’s what he means by top-down causation: the higher level constrains what the lower level does.

“The physics is enabling it to happen. We are telling the physics what to do and the physics does what we tell it to do. The physics isn’t deciding anything.”

The thermostat is the cleaner version of the same argument. You turn a dial. That’s a macroscopic thing — a human hand, a number on a label. The moment you turn it, molecules in the room start moving faster. No law of physics reached in and said “be warmer.” You did. The physics obeyed. Same for a light switch — the current doesn’t decide to flow, the switch decides.

What a context is, and why it matters

The word doing all the work in Ellis’s argument is context. Think of it like this: the laws of physics are a grammar, but a grammar by itself doesn’t write a sentence. You need a speaker, a situation, a goal. Maxwell’s equations sitting on a page cause nothing. Maxwell’s equations inside an iPhone, with a battery, a screen, a user, and an operating system, cause TikTok.

He calls this out as a blind spot in how physicists are trained. They work with what he calls “context-free” reasoning — grab the equations, solve them, call it done. The rest of the world — biology, engineering, society — knows the equations don’t tell you what will happen until you specify the situation they’re sitting inside.

Five flavors of causation (because one isn’t enough)

Ellis is not shy about taxonomies. He splits causation into kinds:

Physical — gravity pulling, current flowing. The textbook stuff.
Purposeful — biology. Every living thing has purposes. A heart pumps in order to circulate blood. Physics doesn’t have an “in order to.”
Symbolic — language, law, money. A stop sign stops cars through symbols, not photons.
Abstract — algorithms, mathematics. Computing runs on abstract symbols that then move electrons.
Social — decisions, wars, elections. People cause things by choosing.

Inside social causation he folds in two eerie ones: historical (the Troubles in Northern Ireland, where a grievance from 250 years ago still gets someone killed today) and imaginative (Steve Jobs imagined the iPhone, and now the world has iPhones — a cause that ran from the future backward, from an image in a mind to atoms arranged into a rectangle).

The list matters because the reductionist move is to pretend only the first kind is real. Ellis says all five show up, all the time, in every system more complicated than a rock.

The iPhone test

This is the trick question he likes most. The universe’s baby picture is the “last scattering surface” — the oldest light we can see, from when the universe was 380,000 years old. Cosmologists can tell you everything about that surface: slightly-wrinkled, Gaussian, well-understood.

Now the question: was the iPhone in his hand inevitable given the data on that baby picture? Could a sufficiently smart demon, shown the last scattering surface, have predicted the iPhone?

Ellis’s answer: absolutely not. And the interesting bit is why. If you insist the iPhone was already baked in at the big bang, then somebody had to write “iPhone” into the initial conditions. Which means you’ve just reinvented intelligent design — a nerdy, math-flavored version of it. You’ve smuggled a god into the early universe to explain why a Steve Jobs exists now.

Think of it like this: the universe sets up the conditions for evolution. Evolution produces brains. Brains design iPhones. The early universe didn’t contain the iPhone; it contained the conditions that allow something new to be invented later. That “something new” is the signature of genuine freedom at work.

The evolving block universe

The standard physics picture — especially after Einstein — is the block universe. All of spacetime, past and future, exists at once, frozen. Time flowing is an illusion your brain adds. You’ve probably heard this one in a thousand popular physics books.

Ellis says: pretty, but wrong. His objection has two parts.

First, the block universe argument is made inside special relativity, which is a local, flat approximation. The real universe is described by general relativity plus cosmology. And the moment you put in actual cosmology — the actual Big Bang model physicists use — there are preferred surfaces of time. The standard “13.87 billion years old” age of the universe refers to one of them. The physicists who say “there’s no preferred time” while simultaneously citing “13.87 billion years” haven’t noticed they’re contradicting themselves.

Second, he offers a different picture: the evolving block universe. Imagine the block, but its right edge — the present — is moving. Every moment, spacetime itself grows a little. The past is fixed and frozen. The future doesn’t exist yet. The present is the boundary where “not yet” turns into “now” and then into “always was.” Think of it like a crystal growing — fully formed behind, unformed in front, actively crystallizing at the edge.

This is the picture that lets him have free will without breaking physics. If the future is genuinely open — if it hasn’t happened yet — then there’s room for decisions to actually decide things, rather than just ratifying a script.

Why infinity is not a big number

Ellis has a small, sharp side-argument that a lot of modern physics quietly abuses the symbol ∞. Physicists tend to use “infinity” to mean “some enormous number, pretty much the same thing.” Ellis says no: infinity is bigger than any number that can possibly exist. The universe will be 14 billion years old, then 15, then a trillion. It will never be infinitely old. Not even the first step.

He uses this to dismiss Roger Penrose’s conformal cyclic cosmology (where one universe-cycle hands off to the next across an infinite boundary) and to dismiss AdS/CFT as a description of our universe (it uses a negative cosmological constant; ours is positive). The details matter less than the pattern: “infinity” in physics papers is often smuggling in an assumption that doesn’t hold up if you take infinity seriously.

Laplace’s demon doesn’t work, and here’s why

The old determinist argument goes: if a demon knew the position and momentum of every particle, it could predict the future perfectly. Ellis’s counter is simple and, once you see it, unshakeable.

You are an open system. Stuff comes in from outside all the time — photons, air molecules, phone notifications, a car accident you couldn’t have foreseen. A woman is driving and a car crashes into her; every thought she has after that is different from the thoughts she would have had otherwise. The demon, no matter how much it knew about her brain two minutes before the crash, could not predict the thoughts, because the crash wasn’t in the brain yet.

This is why, he says, evolution built brains as predictive processing systems (the Karl Friston idea) — because the environment keeps surprising you. A closed system can in principle be predicted. A living thing, embedded in a world, cannot.

The mind, dualism, and the hard problem

Ellis describes himself as a dualist, but he’s careful about it. He’s not saying there’s a ghost in the machine; he’s saying consciousness is real, we have no idea how brains produce it, and people who deny there’s a hard problem are being silly.

“We don’t even have the beginning. Now this is of course called the hard problem of consciousness. And the strange thing is some people try to deny there’s a hard problem. Well there is a hard problem. That’s a simple fact.”

He disagrees with Penrose on two things: wave function collapse doesn’t happen inside microtubules (wrong scale, he says — consciousness lives at the level of whole circuits in the whole brain), and Penrose’s conformal cyclic cosmology has an infinity problem.

On the multiverse, he is even blunter. There’s no single “wave function of the universe,” the way Sean Carroll likes to talk about. The reason: the Schrödinger equation is linear, but the real world is not. You can’t get nonlinear stuff out of a linear equation without cheating — and the multiverse is, to Ellis, the cheat. His alternative: local wave functions, stitched together across spacetime like coordinate patches in general relativity, never one global object.

A pigeon, a tower, and what physics misses

The single best image from the whole conversation. Ellis borrows a story from the science writer Philip Ball. Galileo stands on the Leaning Tower of Pisa with Aristotle. Galileo drops a heavy ball and a light ball; they hit the ground together. See, he tells Aristotle — gravity decides everything, nothing can escape. Aristotle nods, pulls out a pigeon, and drops it. The pigeon flies onto a neighboring roof.

That’s the whole argument in one gesture. Yes, gravity is acting on the pigeon. Yes, it’s the same physics as the cannonballs. But the pigeon has wings, and a brain, and a goal — “land on that roof” — and those higher-level structures channel the physics into a completely different outcome. The physics didn’t disappear. It was used.

Richard Dawkins is wrong about genes

A fun aside. Ellis, who has worked with biologist Dennis Noble, thinks Dawkins’s claim that “genes are replicators” is just false as biology. Put DNA in a petri dish with all the molecular ingredients. Nothing happens. The replicator is the cell, which replicates DNA along with mitochondria, membranes, metabolic machinery, and a bunch of other things nobody puts in the “selfish gene” story. Focus only on genes and you’re missing most of biology.

Moral realism, the punchline

He ends somewhere his physicist colleagues mostly wouldn’t. He thinks some things are good and some things are evil as a matter of fact, not a matter of opinion. He draws a line between ethics (what a given culture or person happens to believe at a given time — varies, relative) and morality (what’s actually good or evil — absolute, built into the deep structure of the universe’s possibility space).

He can’t prove it. He says so himself. But after decades spent fighting apartheid in South Africa, he’s not willing to entertain that the Holocaust was merely a matter of taste. And he thinks our tendency to ignore everyday moral reality when we do “serious” science — Steven Weinberg’s famous line that “the more the universe seems comprehensible, the more it also seems pointless” — is tunnel vision. Your wife, your home, your moral intuitions: those are data about the universe too. Telescopes don’t get the last word.

Key Takeaways

Reductionism describes, but does not decide. Physics tells you what’s allowed; context tells you what happens.
Top-down causation is real and everywhere — thermostats, algorithms, pigeons, hearts, organizations. Higher levels constrain, create, modify, and destroy the lower levels they rest on.
The block universe is the wrong picture. Spacetime grows. The past is fixed, the present is where growth happens, the future doesn’t exist yet.
The iPhone argument against determinism: if the last scattering surface contained the iPhone, someone had to write it in. Determinism secretly needs a designer.
“Infinity” is not a stand-in for “a really big number.” A lot of modern physics — cyclic cosmology, AdS/CFT, the multiverse — cheats on this.
You are an open system. Laplace’s demon fails because new information is always arriving from outside. This is why brains evolved as prediction machines.
The multiverse is metaphysics, not physics — you can’t see past the horizon, nobody can prove anyone right or wrong, so stop calling it science.
Moral realism: some things are evil in fact, not opinion. Everyday moral intuitions count as data.

Claude’s Take

Ellis is the kind of physicist who makes you want to put down whichever popular-physics book you’re reading and think harder. He’s not crank-adjacent — he genuinely wrote The Large Scale Structure of Space-Time with Hawking, a book that sits at the technical core of general relativity. So when he says the multiverse is nonsense and Sean Carroll is wrong, that carries more weight than the same sentence from a random contrarian on Twitter.

The strongest move in the interview is the computer/thermostat duet. Once you see it, you can’t unsee it: the reductionist claim “it’s all just physics” quietly depends on context, goals, and structure that physics itself doesn’t provide. The iPhone-at-last-scattering argument is also genuinely tight — either you accept open futures, or you accept a designer smuggled in by the back door.

Where I’d push back: Ellis slides a bit fast between “physics alone doesn’t determine outcomes without context” (which is obviously true) and “therefore the higher level causes the lower level” (which is a stronger metaphysical claim, and what a serious reductionist like Carroll would actually dispute). The philosopher’s version of the debate is about supervenience — whether the macro supervenes on the micro — and Ellis mostly waves that aside with “Carroll is just wrong, that’s simply wrong.” There’s a real argument there and he doesn’t engage it as carefully as he could.

His dualism is also more asserted than argued — “we have no idea how the brain makes consciousness, therefore dualism” doesn’t quite follow. And the moral realism bit, while moving, is him saying I believe this rather than showing you how the universe would contain it.

Still: 8/10. This is a serious physicist saying out loud what many suspect but most are too polite or too trained to say — that physics imperialism has limits, that the world is actually made of levels, and that the “just particles” story is a philosophical stance, not a scientific finding. The evolving block universe alone is worth the price of admission.