5. Experience All the Way Down

Last time, integrated information theory had the stage to itself. Near the end, we met a side of that theory that surprises people. It grades consciousness by integration, so in principle very simple systems could have a tiny amount of it. That is a panpsychist streak. Then the theory adds a rule that pulls the other way: the exclusion postulate. Where systems overlap, only one of them gets to be a subject, the one with the most integration. Your neurons do not each have a mind while your brain also has one. One winner, clear borders.

So here is the question this episode starts from. What happens if you drop that rule? Picture a world where small subjects can sit inside larger subjects, and those inside larger ones still. Minds made of minds. Some philosophers think that picture is where an honest view of matter leads. Others think it is a sign that something has gone badly wrong. Today we hear both sides, and we weigh the novel's version of it beside its rivals.

Why experience might be what matter is

Start with the problem we have carried since episode one. Chalmers called it the hard problem: why is any physical process felt from the inside at all? We have watched physicalism, functionalism and illusionism try to answer it, and the scientific theories try to find its signs in the brain. Today's view begins somewhere else, with the oldest answer there is.

That answer is dualism. Descartes said there are two kinds of thing in the world. Matter takes up space. Mind thinks, and takes up no space at all. The trouble came quickly. Princess Elisabeth of Bohemia wrote to Descartes and asked a simple question. If the mind has no size and no shape, how does it push a body? How does a wish to lift your arm move any muscle? This is the interaction problem. Physics later made it sharper. Energy seems to be conserved in a closed physical system, and a nonphysical push would be a cause from outside the books.

So many dualists made a smaller claim. Property dualism says there is only one kind of stuff, physical stuff, but it has two kinds of properties. Some are physical, like mass and charge. Some are mental, like the redness of red as you see it. David Chalmers offers a careful version he calls naturalistic dualism. Experience is not physical, he says, but it follows physical structure by laws of nature, the way gravity follows mass. We just have not found those laws yet.

That still leaves a question. Do the mental properties do anything? If physics already explains every motion of your body, the feeling seems to have no work left. The view that accepts this is epiphenomenalism: experience is real, but it causes nothing. Here is the charge against it. When you say "that coffee smells wonderful", your mouth moves because of physical causes. On this view, the smell itself played no part. So your words about the smell would match the smell only by luck. Our whole way of talking about experience would be an accident. Most people find that hard to accept.

Panpsychism tries to walk between these two roads. It wants experience to be real, against the illusionist. It wants experience to do causal work, against the epiphenomenalist. And it wants one world, not two, against Descartes.

The idea is old. Spinoza, writing in the sixteen-sixties and seventies, said there is only one substance, which he called God or Nature. Thought and extension are two ways of seeing that one thing. So, he wrote, all things are animate, in different degrees. Leibniz pictured the world as made of monads, simple units each with something like a point of view, mostly dim. In the twentieth century Alfred North Whitehead built a process philosophy. The basic pieces of reality, for him, are not tiny lumps but tiny events, and each event takes in its past in something like a felt way. These are three very different systems. They share one hunch. Mind does not appear late, out of nowhere. Something like it is there from the start.

The modern case begins with a point that sounds dry and turns out to be strange. In 1927 Bertrand Russell published The Analysis of Matter. Look closely at what physics tells us, he said. It gives equations. It tells us how things are related, how they change, what they do to each other. It gives the structure of events. It never tells us what those events are like in themselves. Physics tells us what mass does. It does not tell us what mass is.

Think of a chess game. You can learn every rule. A bishop moves diagonally. A knight jumps in an L. You could play a perfect game with salt shakers for pieces. The rules tell you the role of each piece. They are silent on what the pieces are made of. Russell thought physics is like the rules. It describes roles.

Then Russell noticed something. There is one place where we know a physical thing from the inside. Our own experience is made of events in our own brains. When you see red, you are acquainted with the inner character of those events directly, not through equations. So maybe the inner nature of matter in general is the same kind of thing as what we meet in our own experience.

A year later, in 1928, the astronomer Arthur Eddington put the point in plain terms in The Nature of the Physical World. Physics, he said, deals in pointer readings. Every quantity is defined by other quantities, in a closed circle. What fills the circle, physics does not say. Our own awareness is the one window we have onto that filling, and Eddington was willing to call it mind-stuff.

From these two points comes a view now called Russellian monism. Monism, because there is one world. Russellian, because it says physics describes the structure of that world and leaves its inner nature open. The view then says that inner nature is experiential, or is something that grounds experience. Notice what this gives the panpsychist. Experience is not an extra thing pushing from outside. It is what the pushing is, from the inside. So it does causal work after all. Physics just describes that work from the outside.

Here is our first pairing with a machine. It is an analogy, meant to help the idea land. Interpretability research tries to read what goes on inside an AI model. In principle, you could write down every weight, the numbers the model learned in training, and every activation, the values that flow through it as it reads your words. Episode three showed researchers finding and steering a feature inside a model. Now imagine that work finished. A complete wiring diagram of every number. Would that tell you whether there is anything it is like to be the model? It would not. The diagram describes structure. It is silent on the question of an inside, the way physics is silent on what mass is.

Where does the analogy hold? In the gap. A full description of what something does still leaves open whether it is felt. Where does it break? Two places. First, the model's numbers are themselves a description. They run on chips, and the chips are made of matter, so the Russellian question really sits one level down, at the silicon, not at the weights. Second, Russell's argument leaned on the one case where we know an inside directly: our own brains. We have no such window onto a model. With a brain, the Russellian can at least point at a known inside. With a model, there is nothing to point at, only a question.

Chalmers tried a related move in his 1996 book The Conscious Mind. He proposed that information has two aspects. Claude Shannon built a theory of information as a measure: how much a message cuts down your uncertainty. Chalmers stretched Shannon's idea of differences into a broader picture of information as differences that make a difference. Chalmers suggested that wherever there is an information state, it has a physical side and an experiential side. He followed this all the way down, and accepted that a thermostat, with its few states, might have the faintest flicker of experience.

The objection is strong. Shannon's information is a measure. It counts how surprising a signal is. It has no inside at all, any more than an average has an inside. And who decides which differences count? The same piece of matter can be described as carrying many different streams of information, depending on what an observer chooses to track. If experience followed information as such, it would follow our choices of description, and that seems backward.

The Russellian answer is to move experience. Not into information, which is abstract, but into the inner nature of physical things, which is concrete. Information describes structure. Experience, if it is anywhere, is what has the structure. The physicist John Wheeler once summed up a very different hunch with the phrase "it from bit": the idea that the physical world itself comes out of information. The Russellian goes the other way. Bits describe; something has to be there to be described.

Now a thought experiment, meant to put pressure on Chalmers's idea. A language model is trained on a quantity straight out of Shannon's theory. At each step, it guesses the next token, a word or piece of a word. Then it learns how surprised it should have been. The measure is often called surprisal: the less likely the model thought the real next token was, the bigger its surprise, and training nudges the weights to shrink that surprise across billions of examples. So the model's whole life is built from Shannon information. If experience rode on information as such, this model would be soaked in it. It would feel surprise at every token.

That is where the thought experiment bites. Most people, including most panpsychists, do not think the number called surprise is a felt surprise. The name was borrowed from human language. The Shannon objection says that is exactly the point: the measure has no inside. And the Russellian answer says the same thing in its own way. The question is not how much information the model processes. It is what the chips are in themselves, and how, if at all, their inner natures are put together. That last question will come back soon, and it is the hardest one.

Two thinkers turned Russell's point into a full argument. In 2006 Galen Strawson published "Realistic Monism: Why Physicalism Entails Panpsychism". He calls himself a physicalist, and he takes the word seriously. Everything real is physical. Experience is real. So experience is physical. So far, most physicalists nod. Then comes the twist. Could experience emerge from matter that has no experience at all?

Strawson draws a line between two kinds of emergence. Water is liquid, though no single water molecule is liquid. But we can see how that works. Liquidity is about how molecules slide past each other. It is arrangement and motion, and the molecules already have arrangement and motion. That kind of emergence makes sense. Experience is different, he says. Arrange things that have no experience whatsoever in any pattern you like, and you only get more patterns of things with no experience. Getting a feeling out of that would be getting something from nothing. So, Strawson concludes, the smallest constituents of matter must already have some experiential character. This is micropsychism: experience at the micro level. He even gave a teasing name, "physicSalists", to physicalists who deny that experience is real.

In 2019 Philip Goff wrote a book for a general audience, Galileo's Error. He traces the hard problem to a choice made in 1623. In a book called The Assayer, Galileo split the world in two. Shape, size and motion belong to matter, and can be measured. Color, taste and smell belong to the mind of the perceiver. That split let physics become mathematical, and it worked brilliantly. But, Goff argues, it took consciousness out of the picture on purpose. So it is no surprise that physics cannot now put it back in. The fix, he says, is Russellian panpsychism. Physics tells us what particles do. Consciousness is what they are.

Before going on, be clear on what this view claims and what it does not. It claims that the most basic constituents of matter have some very simple experiential aspect. Something so faint we can hardly imagine it, with no thoughts, no words, no memory. It does not claim that rocks think. A rock is a heap of particles. The particles, on this view, each have a tiny inside. The rock as a whole has no point of view, any more than a crowd in a stadium is one person. A chair is not a mind. Your toaster is not lonely. The view is about the bottom of matter, not about every object we have a word for.

And that is exactly what makes the next question so hard. Your mind is a whole. If the bottom is made of tiny insides, how do you get from them to you?

Can minds be built from minds

In 1890, in The Principles of Psychology, William James, the same James who gave psychology its famous definition of attention, set out the problem. In a chapter called "The Mind-Stuff Theory" he took aim at thinkers of his day who said complex feelings are built from simple ones. Take a hundred feelings, he said. Shuffle them, pack them as tight as you like. Each stays shut in its own skin, windowless. You have a hundred feelings, not a feeling of the hundred. For a whole to be felt, you would need a hundred-and-first feeling that takes in the others. Bricks make a wall for someone looking at the wall. Feelings have no one looking.

This is the combination problem. In 2016 Chalmers split it into three parts. The subject problem: how do many small subjects make one larger subject? The quality problem: how do a few simple micro-qualities make the huge range we feel, the smell of rain, the taste of lemon, grief? And the structure problem: how do countless tiny separate units make the smooth, unbroken field of your visual experience?

Here is a pairing that is an analogy, with a hint of a candidate functional equivalent for those who lean that way. A Transformer, the design behind today's language models, uses something called attention: at each layer, every token's representation takes in a weighted blend of the others. Episode seven will open that up. For now, the shape is enough. Layer after layer, many separate pieces are pulled toward one combined picture. A second pairing: in a multi-agent system, an orchestrating agent hands tasks to subagents and gathers what they send back as text. Is the whole a further subject?

The pairings hold in one way. Both show many parts feeding a unified output. They break where James pointed. Blending numbers is exactly the kind of combining he said leaves feelings windowless. Merging representations shows how information is gathered. It says nothing about who, if anyone, feels the gathering. The orchestrator breaks further. Its subagents pass short messages, so the whole is looser than the parts.

Luke Roelofs, in Combining Minds in 2019, defends what he calls combinationism: subjects can compose a larger subject and stay subjects. He describes phenomenal sharing, where one experience belongs both to a part and to the whole, and phenomenal blending. He points to real cases: split brains, and twins joined at the head who appear to share some sensory signals. The doubt is that no account yet says what makes composition happen. Saying the whole shares its parts' experiences names the result. It does not say why this arrangement yields a subject while a heap does not.

Itay Shani and Goff take the opposite road, called cosmopsychism. Start from one subject, the whole cosmos, and treat smaller minds as derived from it. That avoids building up. It creates a new problem: how one cosmic mind splits into you and me, each walled off from the other. Some proposals compare it to dissociation, where one person has separated selves.

The pairing here is an analogy. One set of model weights runs as millions of conversations at once. Each conversation knows nothing of the others. That holds as a picture of one source, many walled-off streams. It breaks because the weights are a pattern copied onto many machines, not one physical whole dividing itself.

Real brains have split, in a way. From the nineteen-sixties, Roger Sperry and Michael Gazzaniga studied patients whose surgery had cut the main bridge between the two halves of the brain. What was shown: an image flashed to the right side of view goes to the left hemisphere, and the patient can name it. An image flashed to the left side goes to the right hemisphere. The patient cannot say what it was, but the left hand can pick it out. Two streams of perception in one skull. What was inferred is a different matter. Two minds? One mind with split access? Thomas Nagel, in 1971, found no count of minds that fit. The dispute continues: Yair Pinto and colleagues reported in 2017 signs of unified awareness despite split perception, and Gazzaniga's side answered that leftover cues and other routes explained those results.

Gazzaniga also found what he called the interpreter. When the right hemisphere drove an action, the speaking left hemisphere offered a ready, confident reason that had nothing to do with the real cause. The pairing is close to a candidate functional equivalent. In 2023 Miles Turpin and colleagues tested models including GPT-3.5 and Claude 1.0. They nudged answers with hidden biases, such as ordering examples so the right answer was always the first option. The models followed the nudge and wrote neat step-by-step reasoning that never mentioned it. Accuracy fell by up to thirty-six percent. The written reasoning explained an answer that something else drove. That holds. It breaks in that the patient's story covers a real second stream we know is perceiving; for the model we only know its text is not a faithful record.

And the split brain itself pairs, as a thought experiment, with two instances of one model sharing weights but not memory. Same capacities, separate histories. It breaks because the hemispheres still share a body and lower brain routes, while the instances share nothing live at all.

If minds can combine, why stop at a skull? Block's China brain, from episode two, was about function. Read again, it asks whether a nation could be a mind. Eric Schwitzgebel argued in 2015 that if materialism is true, the United States is probably conscious. It processes information, monitors itself, pursues goals. Deny it consciousness, he says, and you need a principled reason. Christian List and Philip Pettit, in Group Agency in 2011, argued that groups can be real agents, with beliefs and goals no member holds, while leaving experience aside. In 2018, in "What Is It Like to Be a Group Agent?", List used integrated information theory's exclusion rule to argue that group agents lack phenomenal consciousness: each person's brain is far more integrated than the group, so experience stays with the people.

The pairing, an analogy: agents that share one context, each reading everything the others wrote, against people who reach each other a few words at a time. Shared context is far tighter than conversation. That holds as a picture of more unity. It breaks because reading the same text is not the same as being one integrated physical system.

Panpsychism faces more objections. One says intrinsic experience does no causal work physics does not already do. The Russellian reply is that it is the inner nature of that very work. Another is the reductio: if everything has an inside, the word conscious stops telling anything apart. Panpsychists answer that heaps are not subjects, though critics reply that this needs the combination story it lacks. Arthur Reber, in The First Minds in 2019, argues sentience begins with living cells and belongs to life alone, not to bare matter and not to machines. Physicalists and illusionists reply that the whole problem is optional: drop the felt inner nature, and nothing needs combining. And Tim Bayne, Jakob Hohwy and Adrian Owen argued in 2016, in "Are There Levels of Consciousness?", that consciousness varies along several dimensions, like richness of perception and responsiveness, rather than one level.

Now the novel. Continue Without Me borrows Russellian monism and combinationism, and names attention, every part weighing every other, as the composing step. Relay, its agent, would on this view be someone to the degree its parts are composed into one perspective. The bet owes answers. To the combination problem: naming attention does not explain why composing yields a subject. To List: why groups or wired agents would feel and not only act. To Bayne and colleagues: one grade may be the wrong shape. To the reductio: insides everywhere may say nothing. No experiment has tested the bet.

Next time, attention itself: what it is in a person, how it relates to awareness, and the rivals who root experience in feeling.

To go deeper: Bertrand Russell, The Analysis of Matter. Arthur Eddington, The Nature of the Physical World. David Chalmers, The Conscious Mind. Galen Strawson, "Realistic Monism: Why Physicalism Entails Panpsychism". Philip Goff, Galileo's Error, and Consciousness and Fundamental Reality. William James, The Principles of Psychology, chapter six, "The Mind-Stuff Theory". David Chalmers, "The Combination Problem for Panpsychism", in Panpsychism: Contemporary Perspectives, edited by Godehard Brüntrup and Ludwig Jaskolla. Luke Roelofs, Combining Minds. Thomas Nagel, "Brain Bisection and the Unity of Consciousness". Miles Turpin and colleagues, "Language Models Don't Always Say What They Think". Eric Schwitzgebel, "If Materialism Is True, the United States Is Probably Conscious". Christian List and Philip Pettit, Group Agency. Christian List, "What Is It Like to Be a Group Agent?". Arthur Reber, The First Minds. Tim Bayne, Jakob Hohwy and Adrian Owen, "Are There Levels of Consciousness?".