The Grain of the Hard Problem

If experience were exhausted by discrimination

Consider a view on which an experience’s character is fixed by what the subject can tell apart. Xiaohui Yu and Peter Carruthers hold such a view. They treat sensory qualities as modes of presentation: a represented property is presented as a point in a similarity space, and that space is constituted by the perceiver’s capacities to discriminate.² What a reddish experience is like, on this view, is its position among everything the perceiver could distinguish it from.

Discriminative capacity is a functional property. It concerns what a system can do: which inputs it treats differently, which it treats alike, and how these differences are ordered. Any system that duplicates the discriminative organization duplicates the similarity space, whatever realizes it. So if character is exhausted by discrimination, a functional duplicate has the same character. A philosophical zombie, which duplicates us physically and therefore functionally, would lack nothing.

Yu and Carruthers do not deny that zombies seem conceivable. They explain the appearance with the phenomenal concept strategy. We think about our experiences through recognitional concepts that are not connected a priori to any physical or functional concept, so we can coherently think of the physical facts without the phenomenal ones.³ Their slogan is that there are no qualia and no “pixie dust,” only broadcast representations that can be thought about in a first-person way.

This is a consistent position, and it answers the meta-problem. But notice what it does to the hard problem. The zombie’s conceivability comes entirely from our concepts. Nothing in the world answers to the apparent omission, because the conceived zombie has everything our experience consists in. The judgment that a physical story leaves something out is then a cognitive artifact, as stubborn as a perceptual illusion but not about anything. The problem is not solved; it is shown to have no subject matter. Whatever label Yu and Carruthers prefer, this places them close to illusionism. François Kammerer calls such a position verbal weak illusionism: what it calls phenomenal consciousness is exactly what the strong illusionist calls quasi-phenomenal consciousness, so it differs from strong illusionism only in its words.⁴ A theory that grounds all of our sense of hardness in concept isolation, and none of it in what the concepts refer to, is of this kind.

The argument therefore poses a dilemma, not a refutation. Either experience outruns discrimination, or the hard problem has no subject matter. A functionalist can take the second horn. What the argument shows is the price.

Overflow beyond the whole profile

The first horn needs care. A familiar argument seems to show that experience outruns discrimination. In a graded series of shades, each shade can be indiscriminable from its neighbor while the ends are clearly distinct. If looking the same were sameness of character, sameness would fail to be transitive; so some indiscriminable experiences differ in character.

This shows less than it seems to. It shows that character outruns pairwise discrimination. Nelson Goodman supplied the functionalist’s reply long ago. Two qualities are identical, on his criterion, only if they match exactly the same other qualities.⁵ Neighboring shades in a series differ in which further shades they match. So a quality space built from the whole matching profile is finer than any pairwise test, and it registers the differences the sorites exposes. Austen Clark built a theory of sensory qualities on this basis.⁶

The hardness argument needs more: character that outruns the subject’s whole discriminative profile, including every matching relation. The sorites cannot establish that. Two other considerations can.

The first concerns types and tokens. A matching profile is defined over repeatable qualities. To say that a shade matches these and not those, we must compare it across occasions, which makes the matching profile a property of a quality type. Experience is a sequence of occurrences, and the exact character of an occurrence depends on its context: adaptation, neighboring surfaces, the momentary state of the vehicles. Two occurrences that fall under the same type, and so share a matching profile, need not share exact character. The profile individuates what can recur; it cannot individuate what occurs once. A functionalist could appeal to the matches an occurrence would have under counterfactual comparison. But each comparison supplies a new context, and the context is part of what fixes exact character. The outcome can depend on which comparisons come first, as adaptation does, so there may be no fact about what a given occurrence would match. Even for types, a matching profile is well defined only under standardized conditions, which is why psychophysics fixes viewing conditions and adaptation states. The total discriminative profile is then either a profile of types under idealized conditions, too coarse to fix the character of occurrences, or not well defined at all.

The second concerns realization. On our identity proposal, experiential character is the organization of the vehicle that carries the content, assessed at the grain at which its regulation corrects it. Two organizations can share every discriminative disposition while differing in that organization. The unfolding argument in the theory of consciousness exploits exactly this: a recurrent network can be replaced by a feedforward one with the same input–output behavior.⁷ Whether such replacements preserve experience is contested. But nothing in the discriminative profile settles it, and the profile is all a functional duplicate is required to share.

Two qualifications keep the claim honest. “Discrimination” must mean what the subject can tell apart for its own purposes. If it covered every subsystem that responds differentially, every physical difference would count as discriminated, and the thesis would be trivial.

That qualification answers an objection. Regulation corrects differences, and it can correct only what it registers. So whatever regulation tracks must be discriminated, and the overflow collapses back into discrimination. The objection equivocates on which system does the discriminating. The regulator must register fine differences; the subject need not. Access is a bottleneck: working memory and report are narrow, and a narrow channel is many-to-one. To keep what passes it accurate, regulation must use information the bottleneck discards: fine detail, the reliability of sources, consistency across levels. Errors at a fine grain spread into coarse summaries, so the vehicles must be maintained at a resolution that access never needs. The regulator’s sensitivity to differences the subject cannot tell apart is therefore no concession. It is the overflow, located where our account puts it: in the constitutive organization of the occurrence, not in what the subject can use.

The second qualification is that the excess must be present in experience, not merely in the implementation. Gary Drescher proposed that qualia are like gensyms, tokens whose distinguishing features introspection cannot inspect, so that features are “merely hidden, not absent.”⁸ On his account the hidden features belong to the implementation. A functional duplicate with a different implementation would then lack only implementation detail, which is nothing experiential. The overflow that makes the problem hard is overflow in what is presented, as What Phenomenal Intuition Gets Right argued: present in the occurrence, though not represented in any form available to memory, comparison, or report.

Why zombies are conceivable

If experience outruns discrimination, the functional duplicate is no longer a case in which nothing is missing. A system that duplicates our discriminative organization without our vehicle organization could genuinely differ in experience. That is enough to give the hard problem its subject matter: a functional description leaves something out. Whether such a system could lack experience altogether is a further question, and at best open. A duplicate that keeps some regulated carrier would plausibly have experiences of the same types, differing in exact character. One that dissolves the carrier, as some unfoldings may, would be a candidate for having none.

It does not follow that a physical duplicate could differ in experience, let alone lack it. Our proposal identifies experience with vehicle organization, and a physical duplicate duplicates that. So we still owe an answer to the meta-problem: why does a physical zombie seem conceivable?

Our answer is that the conceptual frame in which we conceive it cannot express grain dependency. Experiential character depends on organization at a particular grain. Nothing in our concepts says so.

Consider the concepts involved. A phenomenal concept is recognitional. It picks out an experience by its presented character, and it can carry at most what the subject can discriminate, since recognition is a discriminative achievement. Its referent lives at a finer grain. The concept therefore cannot encode the grain of what it refers to. When I think this of a shade, nothing in the thought says whether this is a pattern at the grain of neural populations, of molecules, or of something coarser than either. Physical concepts, for their part, specify grains but carry no phenomenal mode of presentation.

To conceive a zombie, we hold a physical description fixed and subtract the experience. The subtraction would be blocked by a concept saying that experience is the organization at grain G, which the physical description already fixes. We have no such concept. So the subtraction goes through in thought, whatever the facts are.

This differs from the usual phenomenal concept strategy in one respect. The usual strategy treats the isolation of phenomenal concepts as a basic feature of how they work. Here it follows from the overflow. Phenomenal concepts are isolated because recognition is coarser than its referent, so no concept a subject can deploy captures the grain at which its experience lives.

The answer is also partly veridical. Part of what drives the zombie intuition is correct. A functional description, specified at the grain of cognition, really does underdetermine experience, because experience is finer than cognition.

By cognition here I mean the recipients of what Ned Block calls access: the systems for which access-conscious content is poised, namely reasoning, speech, and the rational control of action.⁹ In global workspace architectures, they are the consuming systems to which contents are broadcast. Rational control of action means its selection. The online control of movement, with its unconscious sensorimotor feedback, falls outside cognition in this sense. Cognition is not all the information processing that drives behavior. Perception is processing too, and so is the regulation that constitutes experience.

The zombie intuition extends the verdict about functional descriptions twice. First, it slides from a difference in experience to its absence. Once a description leaves open which experience a system has, it seems to leave open whether the system has any. And for behavioral duplicates, that possibility could be plausible. The wedge of difference works as an intuition pump for absence, although a description that fails to fix character need not fail to fix that there is experience. Second, it extends the verdict from functional descriptions to physics. That extension is natural, because our frame presents functional and physical descriptions alike as third-person structure. Chalmers makes the point into an argument: physical descriptions give us only structure and dynamics, and structure and dynamics entail only further structure and dynamics.¹⁰ We accept the premise and reject the conclusion. Experience is structure and dynamics at its own grain. What structural description cannot deliver is the recognitional mode of presentation, and that is a limit of concepts, not a gap in the world. Functional and physical descriptions differ in grain, not in kind, and the frame has no way to register the difference.

Physics has a grain

A Wolfram-style metaphysics helps with the second extension, unexpectedly. In Wolfram’s framework, physical law is not a description of reality at no grain. Bounded observers embedded in the Ruliad aggregate distinctions they cannot track, and the regularities they encounter are the laws of physics. Spacetime continuity, the second law of thermodynamics, and the quantum formalism all arise, on this view, from how observers like us coarse-grain an underlying structural dynamics.¹¹ Wolfram calls that dynamics computation; Presence and Actuality explains why we avoid the word, and puts the point this way: coarse-graining is constitutive of the physical articulation, since what counts as a physical object, state, or relation emerges through it.

Physics as a science is then indexed to a grain. That gives us three grains, from coarse to fine:

1. Cognition: the recipients of access, which fix what the subject can discriminate for its own purposes, and the grain at which functional descriptions of a mind are usually given.
2. Experience: the organization of the vehicles, finer than cognition.
3. Physics: the grain at which observers like us formulate physical law, far finer than experience.

The two relations differ in strength. The hardness argument needs only that experience draws some distinctions that cognition does not; the two could cross-cut rather than nest. The solution needs physics to refine the grain of experience, so that a physical duplicate fixes it.

The ordering is not a priori. The grain of experience is fixed by the vehicles’ organization; the grain of physics is fixed by the kind of observer that formulates it. It could have turned out that the physics available to some observers was coarser than the organization of their own experiences. For such observers, two systems duplicated at the grain of their physics could differ in experience. The differences would be confined, though. A physical description, however coarse, fixes the causal structure of what it describes at its grain: where things are and how they change. Two systems that share it cannot scramble their organization the way a behavioral duplicate can, as when an unfolding or a lookup table reproduces a system’s behavior with a different organization throughout. Their differences would lie below the grain of their physics, within a causal structure fixed above it.

Their physicalism would still be true. Here we must separate physics as a science from physicalism as a metaphysics. Physicalism, in the sense we hold, is the claim that experience is part of the structure of reality, with no categorical natures and no extra ingredients. (The name replaced materialism once physics outgrew matter, and a future philosophy might rename the view again if physics as a science stops being the best articulation of that structure.) Physics as a science is one articulation of that structure at one grain. If that grain were coarser than experience, physics would be incomplete with respect to experience, while experience remained structure. The hard problem would then be unsolvable by physics without being a problem for physicalism.

For observers like us, the ordering went the other way. Our physics is fine enough that a physical duplicate duplicates vehicle organization, so it duplicates experience. The identity of experience with that organization is necessary, but we know it a posteriori, and the a posteriori part has content: it is the empirical fact of where our physics’ grain lies relative to our experience’s. Type B materialism is often accused of positing brute necessities.¹² On this account the necessity is not brute. It follows from an identity together with a discovered ordering of grains.

The zombie argument now shows its seam. It moves from the conceivability of a zombie to its possibility, and from its possibility to the falsity of physicalism. The first step requires that phenomenal concepts reveal the nature of their referents, so that what we can conceive with them tracks what could be. A concept that cannot express the grain of its referent does not reveal its nature. It leaves open whether experience lies above or below the grain of physics, which is exactly the question on which possibility turns. The second step slides between physics as a science and physicalism as a metaphysics. Even a successful conceivability argument would show at most that physics is incomplete at its grain.

The same ordering separates our view from Russellian monism, which shares our starting point and reverses our direction. Bertrand Russell argued that physics conveys only the structure of the physical world, while our percepts acquaint us with something of intrinsic character.¹³ Contemporary Russellian monists add a premise: structure must be the structure of something. Relations need relata, dispositions need categorical bases, and a world of pure structure, with nothing to stand in the relations, is held to be inconceivable. Experience is then the one intrinsic nature we know, and a natural candidate for what physics describes from outside.¹⁴

We reject the added premise. Ontic structuralism holds that relata are individuated by the relations they stand in, with no natures beneath them.¹⁵ Whether that is conceivable is the crux, and the grain account bears on it. Where does the idea of a non-structural intrinsic nature come from? Its one alleged instance is our own experience, which presents itself as qualitative and seemingly without structure. What Phenomenal Intuition Gets Right traced that seeming to the uptake. Presence is finer than recognition, so a structured occurrence is recognized without its structure, and the recognition adds the clause and that is all. On our account, then, the paradigm of an intrinsic nature is structure below the grain of recognition. The Russellian takes the one place where structure hides from us as evidence of something that is not structure.

The two views also place experience on opposite sides of physics. For the Russellian, experience lies beneath physical structure, as its categorical base. For us, physics is structure at one grain and experience is structure at another, and the grain of experience is coarser than that of physics, not hidden beneath it.

Mary’s photograph

Chalmers’s ideal reasoner knows the complete physical truth and reasons without limit. It is a generalization of Frank Jackson’s Mary, who knows all the physics of color vision but has never seen a color.¹⁶ Such a reasoner can derive the organization at the grain of experience. Chalmers will say that it still cannot derive what that organization is like. We agree, and deny that this shows a gap in the world.

The grain account says why the gap is predicted. A concept that closed it would have to specify the grain of its referent, as physical concepts do, while bearing its mode of presentation, as recognitional concepts do. Recognitional concepts get that mode of presentation from undergoing, and undergoing reaches them only at the grain of discrimination. At the finer grain the occurrence is present but not represented at all. What would close the gap is not a better concept but the occurrence itself. What Phenomenal Intuition Gets Right called this a gap in access that the theory predicts.

Mary could still get the occurrence itself. She could look at a photograph of a ripe tomato. In Chalmers’s terms this is cheating, and it is worth saying why. First, the photograph reproduces the cause, not the experience. It matches the stimulus at the grain of light, and the experience is built by the viewer’s own visual system; a viewer with different vision would get a different experience. Second, the route bypasses Mary’s understanding. Anyone looking at the photograph would have the same experience, and garbling her physical knowledge would change nothing. Third, it hands the disputed step, from the physical to the phenomenal, to her sensory apparatus rather than her reasoning. Stimulating her cortex directly would cheat in the same ways. A technology cheats when it causes the experience independently of what the reasoner knows.

A route that does not cheat is conceivable. Take a reasoner whose understanding of the description drives the construction. It reads the physical description, derives the organization at the grain of experience, configures its own vehicles to match, and brings them into its regulation. Then it undergoes and recognizes the result. What it reproduces is the experience’s organization, not a stimulus. The route runs through its grasp of the description, so a garbled description would yield a different result. And no sensory apparatus does the work. This is reasoning by construction rather than by conceptual entailment, close to what Kant called the construction of a concept in intuition, through which he took geometry to be synthetic a priori.¹⁷

Humans have a small precedent. Hume allowed that someone who had seen every shade of blue but one could supply the missing shade in imagination.¹⁸ But our power to write into our own vehicles is coarse. Imagery passes through the same bottleneck as access, in the other direction. We can interpolate within quality spaces we already have, and cannot add a dimension to them. That is why Mary is stuck.

A specifically designed AI need not be. It would need three things. The first is a path from its reasoning into its experiential carrier at the grain of experience, which nothing in principle rules out for engineered carriers. The second is a substrate general enough to realize the target organization, which a faithful simulation of the causal pattern would satisfy. The third is regulation that takes the constructed vehicle in, since an unregulated vehicle would be a fragment rather than an experience. (Standard neural networks are in Mary’s position. A transformer’s reasoning reaches its own carriers only through the sampled token, a channel as narrow as human imagery. But the write path needs no exotic technology, because mentality outstrips experience. Much of a mind’s activity is nonphenomenal, and a nonphenomenal process can operate on experiential vehicles at a grain the subject never accesses. A network could compute a target configuration and write it into the carrier of a regulated experiential subsystem.) For such a reasoner, complete physical knowledge would yield knowledge of what any specified experience is like. That includes exact character, since it could construct the context as well.

The photograph and the constructor locate the gap precisely. No conceptual derivation turns a description into what it is like. That holds for every architecture, since a description is not the occurrence it describes. The gap is a gap in analytic derivation. Construction from the description can close it for architectures with the right write path. Mary’s predicament is a fact about human architecture, not about the world. The ordering of grains matters here too. A reasoner can construct an experience from physics only because our physics refines the grain of experience. Had physics turned out coarser than experience, the complete physical description would not tell even the ideal constructor exactly what to build.

Chalmers can reply that constructive knowledge is acquaintance gained by undergoing, and so is not a priori in his sense. We need not dispute the word. His ideal reasoner is idealized in capacity but not in architecture. Idealize the architecture as well, and the premise of the knowledge argument fails for that reasoner. What survives is the claim that knowing what an experience is like requires undergoing it, and that claim is common ground.

A second objection presses the other way. If experience is finer than cognition, how can we know anything about it? We know its type through recognition and its constitution through science, as we know water’s. What escapes both, for reasoners like us, is the exact character of a particular occurrence. That is a limit we should expect, and What Phenomenal Intuition Gets Right argues that it makes the ineffability of experience partly veridical.

What remains open

Wolfram’s observers should not be read as subjects. His derivations of physical law use generic features: computational boundedness, persistence in time, and the equivalencing of many states as one. “Whenever a system does enough equivalencing,” he writes, “that’s tantamount to it ‘acting like an observer.’”¹⁹ An observer in this sense is closer to a macroscopic system than to a mind, much as decoherence yields a classical world for any macroscopic apparatus, watched or not. The vehicles of experience are systems of this kind. Their organization lies within the physics that observers like us extract, and that physics refines it.

What remains open is whether the ordering is a coincidence. Wolfram also writes that “the possibility of our coherent inner experience” depends on our “operating as observers.” If the equivalencing that yields our physics and the regulation that constitutes our experience are features of one kind of system, a theory of observers might show why a system that experiences must extract a physics finer than its own vehicles. The ordering we called a posteriori would then rest on something closer to a synthetic a priori principle. Establishing that requires the theory of observers and of finite occurrences that Presence and Actuality left as a direction of work.

What a coarse-graining is

For Michał and Thomas, a coarse-graining is not something an observer does to a description. “The implicit coarse-graining is the point, not a flaw,” Michał says, and Thomas works out what that means: “A coarse variable earns reality when interventions and interactions repeatedly respect it.” Coarse-grainings are relative to families of interactions, not to observers. Thomas argues that they also cross-cut rather than nest. A membrane’s electrical and mechanical organization preserve different invariants, so reality looks “less like one hierarchy of forms and more like an atlas of overlapping operational quotients.” Michał’s own version is more conservative: “Many coarse-grainings belong to physics because they are tractable mathematically but from a certain level upwards coarse-grainings belong to biology, sociology et cetera or even to metaphysics.”

The atlas bears on Part I’s ordering, which runs from cognition through experience to physics. Michał was the skeptic here. His reply to the atlas was a challenge: “But which counterexample shows non-nested coarse-grainings? Can they overlap?” Thomas retracted: “I asserted non-nestedness before producing a clean specimen.” In Volume 5 he still allows that phenomenal actuality may be “finer than or cross-cutting” cognitive organization. Part I’s ordering survives this, because it requires nesting only at one end. Experience may cross-cut cognition; physics must refine experience. At that end Thomas’s own formalism agrees. In Volume 5 he draws both the cognitive and the phenomenal quotients from one base of physical brain states. Why physics refines experience rather than cross-cutting it is the a posteriori fact Part I identified. For observers like us, physics describes the vehicles in finer detail than their regulation respects. Had it cross-cut the grain of experience instead, our physical duplicates could have differed in experience.

Their interaction-relative notion seems to conflict with the Wolfram picture Part I relies on, and the dialogue says so. When Michał asks whether the Ruliad answers their question, Thomas replies that Wolfram’s observers “sample and coarse-grain” the Ruliad, while their own programme seeks “not observer-relative coarse grainings” but natural decompositions. The two can be reconciled. A Wolfram observer is not an interpreter imposing a description. It is a bounded system embedded in the Ruliad, able to enter only certain interactions with the rest of it. Physics relative to observers like us is then physics relative to the family of interactions available to systems like us: a case of Michał’s interaction-relativity, not an alternative to it.

The interaction-relative notion also fixes the grain of experience. That grain is the grain that the vehicles’ own regulation respects: the distinctions that its corrections preserve. That is a coarse variable earning reality by being respected in interaction, with no external observer required.

Here the earlier disagreement returns. In a later dialogue, Thomas distinguishes three tiers: descriptive coarse-graining, operational individuation, and individuation as a being possessing esse.²¹ We have just used the middle tier to fix the grain of experience, and Presence and Actuality argued that the third tier adds nothing once the second is established. Michał does not grant this for experience. Substantial unity, he says in Volume 5, seemed “a philosophical perspective rather than ultimate reality. But in the case of phenomenal consciousness the ontological aspect is paramount.” Our use of his quotients therefore takes a side he has not taken: it treats operational individuation as enough in the one case where he holds that it is not.

The hardness

Michał’s cases in Volume 5 supply what Part I’s first section needs. Two computers implement the same machine, but one has its zeros and ones physically swapped: “different in one mode but the same in another.” A person suffers episodes of pain that leave no memory, “isomorphic to a painfree variant under typical interaction but very different phenomenally.” The left eye renders colors differently from the right after being covered, “but you can’t tell until you compare them simultaneously.” Thomas distinguishes phenomenal differentiation from phenomenal discrimination: a difference can exist without being available to any operation that takes it as input.

These cases matter because Thomas had nearly taken the other road. In Volume 4 he suspected that a phenomenal difference “with absolutely zero possible differential consequence” may be “incoherent or at least evidentially empty.” Had that stood, phenomenality would have been one more articulation among the operational ones, and the hard problem would have been dissolved. Michał’s cases blocked it.

Then Michał made the move that Part I makes: “Now that cognition has been decoupled from phenomenal consciousness it’s easier to begin to suspect that it’s a physical process with fine substrate-dependence.” The decoupling is methodological. In a later dialogue Michał calls his slogan that phenomenality is orthogonal to cognition “a mnemonic reminder”: “I don’t insist on any metaphysical orthogonality.” The slogan forbids reading the structure of experience off the structure of cognition; it does not assert that the two vary independently. Thomas’s Summalecton 150 develops the suspicion. “Cognition buffers differences that phenomenality preserves.” The phenomenal quotient “preserves physical distinctions lying inside equivalence classes of the cognitive quotient.” And “fine” means only “finer than the distinctions preserved by the relevant cognitive articulation.” Fine substrate-dependence, Thomas adds, treats phenomenality “as more physical, not less.”

That is the lower bound of Part I’s ordering, reached by a different route. Two refinements separate us. First, our overflow is not a difference with zero consequence. It makes a difference within the occurrence, to the regulation in which the vehicle has its character, and fails only to reach cognition. Thomas’s Volume 4 worry therefore does not apply to it. Second, “substrate-dependence” is not quite our claim. We locate experience in a causal pattern that any substrate could realize, but at a grain finer than cognitive function. Michał’s swapped-bit computers differ at a grain below anything that pattern needs to track, so we would not expect them to differ in experience.

The two-eye case adds a distinction that Part I should keep. The difference between the eyes was discriminable under comparison; it simply was not discriminated. It falls inside the subject’s discriminative profile, so it is not yet the overflow the hardness argument needs. Thomas’s counterpoint is also apt: the difference may be partly constituted by the adaptation and comparison context. Part I’s argument from types and tokens relies on the same dependence on context.

The meta-problem

In Volume 4 Thomas came close to Part I’s diagnosis. “The modern hard problem may partly arise because we first describe the event exhaustively in a vocabulary deliberately purged of appearance and then ask: Where is the appearance?” Phenomenal and functional descriptions may be “cross-cutting coarse-grainings of one process,” and “you cannot reach one cross-cutting quotient merely by making another quotient increasingly elaborate.”

That last sentence is the core of our answer to the meta-problem, put in the language of quotients. Adding functional detail never yields the phenomenal articulation, because the detail is added at the wrong grain. Thomas noticed where his analogy with temperature strains: temperature can be derived from the microdescription, and phenomenality “currently cannot.” Part I’s answer is that the organization at the grain of experience can be derived; what cannot be derived is its recognitional mode of presentation.

Here we differ in emphasis. Michał and Thomas locate the remaining gap in evidence. Every datum about experience passes through cognition, and introspection preserves only a coarse map of what it reports on: “manifestation is not description.” We locate it in concepts, which cannot express the grain of their referents. The two accounts fit together. A coarse introspective map is the reason recognitional concepts carry only discriminative grain. The evidential limit explains the conceptual one.

Where Michał would resist

Two points of Part I would meet resistance.

The first concerns what “cognition” means. Michał uses the word more broadly than Part I does: “All phenomenal actuality is computed by a kind of cognition.” Part I’s technical sense is narrower: the recipients of access, in reasoning, speech, and the selection of action. He distinguishes the cognition that constitutes phenomenal actuality from the further cognition that takes its output as input, and Presence and Actuality has already used that distinction. The coarse end of Part I’s chain is the second kind. On our account, the first kind includes the regulation that constitutes the occurrence, and it sits at the grain of experience.

The second is the claim that physics could have turned out coarser than experience. Part I supports it with Wolfram’s account of observers, and Thomas rejects observer-relative coarse-graining. Read as above, Wolfram’s observers are a case of interaction-relativity, but the claim survives even without that reading. Thomas warns that “if physics swallows everything, we risk mistaking the currently favored physical articulation for exhaustive ontology,” and that “our ‘bottom’ remains provisional.” That separates physics as a science from what exists, which is the distinction Part I’s diagnosis of the zombie argument needs. Physics as a science articulates the world at whatever grain its available interactions respect. Whether that grain refines the grain of experience is something we discover.

The cogwheel

Volume 6 raises a worry that the grain account must face. If experience is a physical mode finer than cognition, Michał asks, why does it present “a moment of a life” rather than “abstract modulations and abstract shapes”? “The nightmare is that we really ought to feel like the man as a cogwheel experiencing structured but meaningless qualia.” Thomas’s resolution is modest: understanding belongs to the temporally extended organization, and “the phenomenal present may always be the cogwheel.” Michał answers that if this were true, “our theories would be doing fine.”

Our account gives a reason to expect neither meaningless modulations nor the whole of understanding. The fine grain of experience is the fine grain of content-tracking structure. On What Phenomenal Intuition Gets Right, a vehicle’s organization realizes the similarity geometry of what it represents. Differences below the grain of cognition are therefore differences in how the represented scene is carried, not noise in the substrate. Nor is the moment local in Thomas’s sense. On the global variant of the account in Experience, Selfhood, and the Organization of a Mind, the regulation that binds a moment draws its information from across the system, including working memory. A wider understanding can then shape the moment without being contained in it, as Presence and Actuality argued. The understanding of a life need not be present in a moment. But the moment is not a cogwheel’s view of abstract shapes either.

Whether the hard problem was solved

Neither Michał nor Thomas claims a solution. Thomas’s verdict is that the problem has begun “to look less like the traditional ‘hard problem’ and more like the specification of an unknown physical regime.” We agree with the reframing and add what it leaves out: why the reframing does not leave a hard problem behind. Once experience is a regime at a grain that physics refines, the remaining gap is a gap in recognition. Recognition cannot carry the grain of what it recognizes, which is why zombies seem possible. And the regime is fine-grained in the right direction, which is why they are not.