The Ladder and the Dust

Take a single molecule of water and ask whether it is wet. The question does not really parse. You have two hydrogens, one oxygen, a bond angle of about 104.5 degrees, a small separation of charge. Nothing in that description is wet, and nothing in it could be. Wetness is a property of a crowd. You need enough molecules that hydrogen bonds are forming and breaking millions of times a second, and then the crowd starts doing something no member of it does—clinging to your skin, climbing the inside of a narrow tube, holding a droplet together against gravity.

That is emergence in the version everybody finds agreeable, and its agreeableness is exactly the problem. Wetness is emergent in a thoroughly boring sense: we can derive it, we know the mechanism, nobody thinks a new fact entered the universe when the second water molecule arrived. When people reach for emergence to explain the things that actually trouble them—life, mind, meaning, the fact that there is something it is like to be you—they usually want a much stronger claim, and they do not always say which one.

A Word Coined to Name a Gap

The word is younger than the idea. G. H. Lewes coined emergent in 1875, borrowing a distinction John Stuart Mill had drawn in 1843 between two ways that causes combine. Sometimes they simply add: two forces on a body give you a resultant you could have calculated in advance. Sometimes they do not. Mill called the second kind heteropathic, and his example was chemistry—sodium, a metal that catches fire in water, and chlorine, a gas that was later used as a weapon, combining into the white crystal you put on your food. The product bears no legible relation to the ingredients.

In the 1920s this hardened into a movement. Samuel Alexander’s Space, Time and Deity (1920), C. Lloyd Morgan’s Emergent Evolution (1923) and C. D. Broad’s The Mind and Its Place in Nature (1925) proposed a layered world—physics, then chemistry, then life, then mind—with each layer obeying genuinely new laws that could not be derived from the layer beneath. Alexander thought such facts had to be accepted, in his phrase, with natural piety: taken as found, without expecting a derivation.

Their flagship case was chemistry, and it is worth dwelling on what happened to it. Broad argued in 1925 that you could never, even in principle, derive the properties of a compound from a complete description of its elements. Two years later the argument was dead. In 1927 Walter Heitler and Fritz London applied the new quantum mechanics to the hydrogen molecule and derived the covalent bond from first principles. Chemistry was not a new layer with its own irreducible laws; it was physics, worked out. The British Emergentists lost their best example within the decade and the movement effectively collapsed.

I bring this up not to score a point but because the lesson keeps needing to be relearned. Emergence has been used before to mark a permanent boundary in nature, by careful people, with the best example available to them—and the boundary moved. Any modern claim of the same shape inherits that history whether it wants to or not.

Weak, Strong, and the Difference That Does All the Work

The modern literature owes its clarity to a distinction sharpened by Mark Bedau in 1997 and David Chalmers in 2006. Weak emergence: the higher-level fact really does follow from the lower-level facts, but there is no shortcut to it—the only way to find out what the system does is to let it run. Strong emergence: the higher-level fact does not follow from the complete lower-level description at all, not even in principle, not even for a mind that could compute anything.

Conway’s Game of Life is the clean illustration of the weak kind. Four rules, a square grid, cells that live or die by how many neighbours they have. Out of that come gliders that walk across the board, guns that fire gliders forever, and—this has been built—a universal computer. None of it is visible in the rules. Nobody reading the four rules for the first time would predict a machine. And yet nothing whatsoever has been added: run the rules and the machine is there. That is the signature of weak emergence, unpredictable but not underivable.

Strong emergence is a far bigger claim, and the striking thing is how few candidates there are. Chalmers’s view, which I think is correct as a matter of bookkeeping, is that there is exactly one serious candidate in the known world, and it is consciousness. Everything else we once thought irreducible—the chemical bond, heredity, the supposed vital force that separated living matter from dead—turned out to be weak emergence with the mechanism not yet found.

Why “Merely Weak” Is the Wrong Way to Say It

Popular writing treats weak emergence as a consolation prize—the boring kind, the one that means it is all just particles really. This gets the physics backwards, and the correction is one of the most beautiful results of the twentieth century.

Take a lump of iron and heat it towards the temperature where it stops being a magnet. Separately, take carbon dioxide and push it towards the pressure and temperature where the distinction between liquid and gas disappears. Microscopically these two systems have nothing in common—iron atoms in a crystal lattice with their spins coupled, versus molecules of a gas bouncing around. But as each approaches its critical point, the quantities you can measure blow up as power laws, and the exponents in those power laws are the same numbers. Not similar. The same. Materials with nothing in common fall into a small handful of universality classes, and everything inside a class behaves identically near its critical point.

⤢ enlarge

The explanation is the renormalisation group, for which Kenneth Wilson took the 1982 Nobel Prize. Look at the system at coarser and coarser scales, averaging away the fine structure at each step, and watch what survives. Almost all of the microscopic detail washes out. What is left is a short list: the dimensionality of the space, the symmetry of the order parameter, how far the interactions reach. Everything else is irrelevant, in the precise technical sense that it makes no difference to the answer.

Now turn that around, because the consequence is the whole argument. If a large class of different microscopic worlds all produce the same macroscopic law, then that law cannot be an approximation to any one of them. It is a fact about the class. The higher level is not a blurry summary of the lower level—it is where the law actually lives, and the microscopic details are the part that turns out not to matter.

“More is different.”Philip W. Anderson, Science, 1972

That is Anderson’s title and his entire thesis. His target was what he called the constructionist hypothesis: the assumption that if you know the fundamental laws you can, given enough computing, reconstruct the universe. He argued it fails, and fails for a structural reason—symmetry breaking. At each level of complexity the system settles into a state with less symmetry than the laws governing it, and once it has, new organising principles apply that are simply not statements about the level below. Robert Laughlin and David Pines pushed this further in 2000, describing higher-level laws as protectorates: regimes so insulated from microscopic detail that you could get the microscopic physics wrong and still predict the macroscopic behaviour exactly.

So the reductionist rejoinder—it is all just particles—is not so much false as useless. True in the sense that a chess game is just electrons in a wooden board. False in every sense that would let you say anything about the game.

The Bill Arrives

Emergence has a price, and Jaegwon Kim presented the invoice in 1998. Suppose your decision to raise your arm causes your arm to rise. That decision has a physical realiser—some state of your brain. Now add the principle that the physical world is causally closed: every physical event that has a cause has a sufficient physical cause. But then the brain state already suffices to raise the arm. So either your decision just is that brain state, or the arm is being caused to rise twice over by two independent causes, or the decision is doing no work at all.

This is the causal exclusion argument, and its force is that all three exits cost something. Deny causal closure and you are proposing that physics is incomplete in the brain, which is a large empirical claim with no evidence for it. Accept systematic overdetermination and you have every action caused twice, which no one really believes. Or retreat: say that higher-level causation is a matter of explanation rather than of pushing—that the mental level tells you why in a way the physical level cannot, without adding any shove. Most working scientists take the third exit and sleep soundly. Philosophers tend to notice that it quietly hands back the very thing strong emergence was supposed to buy.

The Other Road: If It Was Not Added, It Was Always There

Panpsychism gets treated as the mystical option. It is not. Its motivation is not wonder but exhaustion, and the argument that drives it is a process of elimination.

Set out what a science of consciousness has to explain. Chalmers’s 1995 framing splits it in two. The easy problems—easy in the sense that a century might do it—are all about function: how the brain discriminates a red surface from a green one, how it integrates information from separate senses, how it monitors its own states, how it produces a report. These are hard engineering questions with no philosophical mystery. Explain the mechanism and you have explained the function completely.

The hard problem is what is left over. Suppose every function is explained, down to the last synapse. You can still ask why any of that processing is accompanied by experience—why there is something it is like to see the red rather than merely to register it, sort it, and act on it. A complete functional account seems to leave the question untouched, because it was never a functional question.

Then the panpsychist argument runs like this. Experience cannot be derived from parts that have none. You do not want it appearing from nothing at some arbitrary threshold of neural complexity. So it did not appear. It was in the parts.

A Very Long Genealogy

The idea is about as old as philosophy. Thales is reported as saying all things are full of gods, and the early Greek hylozoists took life and matter to be inseparable. Spinoza’s one substance has thought and extension as two attributes of the same thing, so mind is not added to matter but is another face of it. Leibniz built the world out of monads, each with some degree of perception. Schopenhauer said the world known from the outside is representation and known from the inside is will—and we each know exactly one piece of it from the inside. Gustav Fechner, the founder of psychophysics and a serious experimentalist, thought plants had inner lives.

But the strongest modern version comes from an unexpected quarter, and it is not mystical at all. In 1927 Bertrand Russell published The Analysis of Matter, and Arthur Eddington made a related argument the same year. Their observation was this. Physics describes structure and relation, and never once tells you what anything intrinsically is. Mass is characterised by how a thing resists acceleration. Charge is characterised by how it interacts. Every term in the theory is defined by its role in a web of other terms, all the way down. Physics hands you a complete relational skeleton of the world with the intrinsic natures left as blanks.

⤢ enlarge

Eddington’s addition is the sharp bit. There is exactly one region of the universe whose intrinsic nature we are not guessing at, and it is the small patch of it that each of us is. We know what it is like to be that patch from the inside. Russellian monism proposes filling in physics’ blanks with the only kind of intrinsic nature anyone has ever met. On that reading panpsychism is not the claim that rocks have feelings; it is the claim that the categorical basis of the physical world is of the same general kind as the one instance of it we happen to be.

The Combination Problem

And then the whole thing hits a wall, and the man who put it there was William James, in 1890, while considering and rejecting the view himself. Take a hundred feelings and combine them. What you get is a hundred feelings. There is no arithmetic by which small experiences add up into one larger experience that has them as parts. James called the doctrine mind-dust, and he did not think the dust could be made to cohere.

⤢ enlarge

Sam Coleman sharpened this into the subject-summing problem. Experiences are not free-floating; they are always somebody’s. So to build a big subject out of small subjects, the small ones would have to stop being subjects—which is not combination, it is replacement. And replacement is exactly the emergence of a new subject out of things that were not it, which is the thing panpsychism was adopted to avoid.

Panpsychism was adopted to avoid one act of emergence. It turns out to require another. The mystery moved; it did not shrink.

This is, to my mind, the most interesting fact in the whole dispute and the one least often stated plainly. Both roads end at the same wall. The emergentist has to explain how experience arises from things that have none. The panpsychist has to explain how one experience arises from many that are not it. Neither has a mechanism. They have relocated the same hole.

What the Science Is Actually Doing

The most developed scientific theory in this neighbourhood is Giulio Tononi’s Integrated Information Theory, which starts from properties experience seems to have and asks what physical system would have them. Its measure, Φ, is meant to quantify how much a system’s causal structure is more than the sum of its parts. The theory’s structure carries a panpsychist implication it does not hide: anything with the right kind of integrated causal structure has some experience, in proportion.

It is also contested to an unusual degree. In 2023 a large group of researchers signed a public letter calling IIT pseudoscience—an aggressive move that split the field, but whose central complaint was testability rather than implausibility. More productively, IIT and its main rival, Global Neuronal Workspace theory, agreed to an adversarial collaboration: predictions registered in advance, experiments designed jointly by people who disagree, results published whatever they show. The first major results came in 2025 and did what good adversarial work does—neither theory came out clean, and both had to give something up. That is what progress looks like in a field this young.

Meanwhile the emergence literature got a cold shower from a direction nobody expected. Large language models were widely reported to show emergent abilities: capabilities entirely absent at small scale that appear suddenly once the model passes some size. It looked like a textbook phase transition, and it was cited as such. Then in 2023 Rylan Schaeffer, Brando Miranda and Sanmi Koyejo pointed out that much of it was an artefact of the ruler. If you score a task with an all-or-nothing metric—exact string match, say—then a capability that is improving perfectly smoothly underneath will cross the scoring threshold abruptly and look like a jump. Re-score the identical model runs with a continuous metric and the discontinuity disappears.

⤢ enlarge

This settles nothing about consciousness. But it is a real methodological warning, and it applies well beyond machine learning. A discontinuity in your measurements is not automatically a discontinuity in the world. Claims about emergence have always been partly claims about what is visible from where the observer happens to stand, and it is worth asking, each time, which part is the world and which part is the ruler.

Where the Bet Is Really Being Placed

Emergence and panpsychism get presented as rivals. They are better understood as two placements of the same bet. Everyone in the argument agrees there is a gap between the physical description of a system and the fact of its experience. Emergence puts the mystery at the top: pile up enough complexity, cross some line, and something genuinely new is there. Panpsychism puts it at the bottom: it was always there in some minimal form, and complexity merely organises it into the kind of thing that can write essays about itself. Neither has produced a mechanism for the step it needs.

What has changed is that both positions are respectable again after most of a century in the wilderness. That is not because either found a proof. It is because the confident reductionism of the middle of the twentieth century made a promissory note it has not redeemed. Fifty years of extraordinary neuroscience have told us in astonishing detail what the brain does, and have not laid a finger on why doing it should be like anything at all.

For what it is worth, my own weight sits closer to weak emergence than to either extreme, and for a thoroughly unglamorous reason: it is the only option with a track record. Every previous case where a gap in nature looked permanent—the chemical bond, the mechanism of heredity, the vital force that was supposed to separate living matter from dead—the gap closed. And it closed by somebody finding the mechanism, never by anyone declaring a new fundamental principle. That is evidence rather than proof, and it does not touch Chalmers’s argument, which is precisely that consciousness is not a functional question and therefore is not the kind of thing that a mechanism would answer even if we had one.

So the honest position is that we do not know. And it is worth being clear that ”we do not know” here is not a placeholder for a fact we expect to arrive next year. It is a description of the shape of the problem: we cannot presently say what a satisfying answer would even look like, which is a different and more serious kind of ignorance. Alexander’s phrase has outlived his theory. Some things may simply have to be taken as they are found, with natural piety, without the comfort of a derivation. Whether the fact of experience is one of them is, still, the most interesting unanswered question in philosophy—and nobody honest is close to closing it.