Appendix B — Scholarly Excurses – Mind, Matter, and Meaning

Deep-dives the general-reader spine deliberately omits. Each chapter states its verdict in brief; the extended treatment lives here.

B.1 Inner Speech and the Vygotskian Story

(Deep-dive from Chapter 17, The Assembled Self.)

Right now, as you read this, the words on the page sound, silently, in something you would call your own voice. Whatever else this thing is, it feels like the deepest interior we have: a private room with the door shut, the last place the world does not get in. This excursus is about why that picture runs upside down.

The bad picture says public speech is the noisy externalization of an already-private inner monologue: the thoughts come first, in the head; the words come later, when the thoughts need company. The picture comes naturally, and it has done more than almost any other to lock the Cartesian theater into modern philosophy of mind. Once you accept that the inner voice is the thing, you have already conceded an interior space with its own contents, accessible only to its owner.

The alternative runs the other direction. Public speech came first, historically in the species and developmentally in each child, and the inner voice consists of that public activity turned inward. The developmental half of that claim is Vygotsky’s story: the child talks with others first, then to herself aloud, then silently, speech migrating inward by stages rather than thought leaking out. What you experience as thinking-in-words is silent speaking in a transformed register: the same lexicon, the same public meanings, the same shared language — transformed, not replayed at low volume. Vygotsky’s own finding was transformation-in-internalization: inner speech abbreviates, drops its subjects, condenses sense, runs in a predicative shorthand no listener could parse. (The mechanism beneath the phenomenology — attenuated motor commands, auditory-forward imagery, some blend — stays contested in the psychology, and nothing philosophical here turns on it.) No separate inner lexicon exists. One lexicon (public, shared, social) does duty in two modes, the inward mode compressed.

Several puzzles relax their grip once you see this. Take meaning. If the inner voice were a soliloquy in a private language, its words would have to be fixed inside the head, by the speaker’s own lights, with no public check — exactly what Wittgenstein attacked in the case of someone trying to name a private sensation, where no public criterion of correct use can settle whether the next use follows the rule or breaks it.¹ Meaning never gets fixed by what goes on inside a single skull; it gets fixed in social practices where words have uses people can correct, share, and inherit. The inner voice borrows those meanings from outside. It does not generate them. Ask the obvious question: where else would it get them? You learned every word you have from speakers around you, in contexts where their uses could be corrected.

Here we owe Sellars and the tradition after him. His Jonesean myth models inner episodes — thoughts — on overt verbal utterances rather than treating them as directly observed.² Crane and Farkas gloss the picture: thoughts are “inner episodes, called ‘thoughts,’ which are conceived on the model of overt verbal utterances, but happen silently in the head of other subjects.”³ The order of explanation runs from public to private. We learn what thinking is by learning to talk; then we learn to call the silent version “thinking.”

Two consequences follow. The first concerns AI. If the inner voice is internalized public speech, then producing fluent inner-voice-like sentences does not, by itself, evidence an inner life behind them. A language model has the surface of inner speech and none of the social, embodied developmental history through which a human speaker learned to use those public meanings; what training can select inside the model’s mechanisms is a further and separate question (note 4, and B.15 below).⁴ Causally internalized public meaning does not exhaust that question: originality of content about actual word-types and their distributions remains unsettled, as B.15 explains. The temptation to see understanding behind the outputs comes precisely from the inverted picture: we assume the words must come from an interior, because that is where our own words seem to come from. They do not. The second consequence concerns us. The most private-feeling activity we have is, at root, a social inheritance — the point Chapter 17 presses at length: every word in the monologue arrived from outside, on loan from a community.

An obvious objection: surely there is more to thinking than silent speech. Mathematicians solve problems without verbalizing them; animals without language clearly think. Take it seriously, because it is mostly right — and the claim survives the objection once narrowed. Not every act of cognition consists in inner monologue; pre-linguistic infants think, non-human animals think, and expert performance outruns any inner narrator. But the experience of thinking-in-words, the voice that feels like the inmost private chamber, is best understood as internalized public speech.⁵ Non-verbal cognition is real; it is simply not what feels private. The room you thought was private always had the door open. You just never noticed the draft.

B.2 Information Talk Is a Metaphor

(Deep-dive from Chapter 24, The Conscious Robot.)

Open any popular book on the brain and somewhere in the first chapter you read that the brain processes information — said with the same equanimity as the kidneys filter blood. Information goes in through the senses, the brain does things to it, behavior comes out. The picture seems so plainly right that the only response is to nod and turn the page. Look at the sentence again.

What exactly enters the brain? Not the photons; those stop at the retina. Not the air-pressure waves; those die in the cochlea. What travels into the skull does so as ion flux across neural membranes — sodium in, potassium out, at rates no ledger could follow. The neurons carry no envelopes marked information any more than a river carries envelopes marked fluid dynamics. Somewhere between the neurons fire and the brain processes information, something has been added, and it pays to ask what.

The deflationary point begins where Chapter 22 left off, but it must begin on the right side of that chapter’s distinction. Unconstrained computational interpretation is observer-relative; disciplined implementation can be an observer-independent fact about a mechanism’s causal organization. The same care applies to information. Shannon information measures real statistical dependence. Semantic information adds a further claim about what a state is about. A neuron fires; another fires in response; the causal and statistical relations belong to the tissue. Calling the second state information about an edge or about food requires more: a history, selected function, and producer-consumer use that privilege one correctness condition over its rivals. Searle puts the warning at its most disarming: “‘Information’ does not name a real physical feature of the real world in the way that neuron firings, and for that matter consciousness, are real physical features of the world. Except for the information that is already in the mind of some conscious agent, information is relative to an observer. […] Information is anything that we can count or use as information.”⁶ His target remains the slide from a useful reading to intrinsic semantics. The book grants more causal structure than his warning acknowledges: selection and training can produce observer-independent, content-bearing mechanisms. What it refuses is the inference from causal traffic, computation, or measured bits to semantic content without a privileged explanatory mapping.

Notice what has not been said. Nothing against describing the visual system as extracting edges or the hippocampus as encoding locations, and nothing against those descriptions being literally correct when selected functions fix the content. The point cuts between a relation and a substance. Information does not name stuff the traffic contains. The grammatical mistake runs ahead of the metaphysical one: processes information suggests handling, handling suggests cargo, and cargo lives somewhere — little packets of meaning piling up in the visual cortex like luggage in a baggage claim, waiting for someone to collect them. Cue the inner theater.

Now the objection that wants it all back: Shannon. Surely there we have information in a precise, mind-independent sense — bits, channels, probability distributions, all perfectly real. The objection deserves a careful answer, because most of it is correct. Shannon information exists and gets measured; Dretske built an epistemology on it.⁷ But two points. First, Shannon information is not semantic information: it measures statistical dependence between source and channel, the same dependence that holds between a room’s temperature and the height of mercury in a thermometer. (Dretske’s natural information — factive correlation, what a state’s obtaining guarantees about its source — sits between the two: stronger than Shannon’s averaged quantities, still short of a correctness condition a state could fail to meet.) Statistical covariance is real; content needs something further to select one condition as what the state is correct about. Second, the slide from computability or simulability to a substantive theory of mind repeats the Church–Turing fallacy Chapter 22 diagnoses.⁸ A brain may genuinely implement computations; that empirical result still does not show that computing alone constitutes thinking. Shannon attributions come nearly free. Semantic ones cost the right selected function, causal embedding, consumer role, and environmental history. Conflating the two gives the illusion of having explained meaning by measuring bits.

The bad picture is not that we talk about information in the brain; we can hardly do without the talk. The bad picture is that the talk names a stuff. It does not. Shannon information names a statistical relation; semantic information names a representational relation whose correctness conditions are fixed by function, history, and producer-consumer organization. The organ stays biophysical through and through. No packets required.

B.3 The Conceivability Trap

(Deep-dive from Chapter 9, Why Dualism Keeps Winning Arguments It Should Lose; the strong-necessity exchange deferred from Chapter 11 completes it.)

Imagine a being physically identical to you in every respect. Same neurons, same synapses, same electrochemical cascades firing in the same sequence. It winces when burned, says ouch with conviction, and passes every behavioral test. The only stipulated difference: nothing it is like to be this creature. No red when it sees the tomato, no warmth when it holds the coffee. The machinery runs in darkness.

This philosophical zombie gives David Chalmers his sharpest modal argument against physicalism.⁹ The popular summary says: if we can imagine the zombie, it must be possible. Chalmers’s argument says something narrower and much stronger. Casual imagery supplies no premise. The relevant achievement is ideal primary positive conceivability, and the possibility it reaches is primary possibility.

The argument in its exact form

Let P state the complete physical truth about a world and Q a phenomenal truth, such as someone is conscious. Levine’s explanatory gap supplies the setting: P can fail to make Q intelligible even if P and Q describe one fact.¹⁰ Chalmers adds this modal argument:

P & not-Q is ideally primarily positively conceivable. One can coherently imagine a sufficiently detailed scenario that verifies P & not-Q when considered as actual, and the imagining survives ideal rational reflection.
Ideal primary positive conceivability entails primary possibility. If a statement passes that test, some centered metaphysically possible world, considered as actual, verifies it.
The phenomenal case has no water-style primary/secondary slack. Core phenomenal concepts do not fix their reference through a contingent description such as the watery stuff around here; their primary and secondary intensions coincide.
Therefore P & not-Q is secondarily, metaphysically possible, and physicalism is false. A metaphysically possible physical duplicate lacks the phenomenal truth, so the physical facts do not necessitate all the facts.¹¹

Each qualifier in the first premise blocks an easy physicalist reply.

Ideal removes errors that better a priori reasoning would defeat. A mathematical impossibility may look coherent to us because we have not found the proof; it does not remain conceivable under the idealization. Positive requires coherent modal imagination of a situation that verifies the statement. Merely supposing a sentence, failing to detect a contradiction, or picturing experts announcing it does not suffice. Positive conceivability need not involve a visual image. The zombie and its conscious twin look the same from outside, yet a thinker can still modally imagine a world-configuration and take it to verify that one lacks experience. Primary asks how a world might turn out when considered as actual. It differs from asking, after the actual referent has been fixed, what could happen to that very referent in a counterfactual world.¹²

That last distinction answers the standard water objection. “Water is not H₂O” has a primary reading on which XYZ fills the lakes and rivers in a world considered as actual. Such a world is possible and verifies the watery stuff is XYZ. It does not show that actual water, whose reference science fixed as H₂O, could have been XYZ. Primary conceivability reaches a genuine possible world; secondary necessity keeps actual water H₂O in every counterfactual world. The Kripkean case therefore supports Chalmers’s two-dimensional bridge rather than refuting it.

The phenomenal case, Chalmers argues, lacks the same escape. A core phenomenal concept does not pick out its referent through a contingent description analogous to the watery stuff around here. What verifies pain or red experience presents the phenomenal property itself. Primary and secondary intensions therefore coincide. If an ideal primary scenario verifies P & not-Q, no later redescription turns it into a merely watery duplicate of the target. The scenario is a zombie world, and physicalism loses.¹³

Where type-B physicalism refuses the bridge

The book grants premise 1. Grant it ideally, primarily, positively, and permanently. No future scan closes the conceptual distance. The acquaintance-recognition account predicts that permanence: a concept acquired through undergoing a state need not acquire an a priori connection to a concept built from public description. Ideal reflection improves reasoning; it does not rewrite the acquisition history of the concepts or turn one epistemic route into the other.

The refusal lands on premise 2. Ideal primary positive conceivability supplies serious modal evidence, not an entailment of primary possibility. Modal judgment should remain answerable to independently supported identities. If Chapter 6’s identity holds, the acquaintance-grounded and descriptive concepts reach one physical world-presenting state even though no ideal a priori route joins them. P & not-Q can then remain ideally positively conceivable under the separated concepts while failing to describe a metaphysically possible world. The scenario records conceptual isolation rather than ontological separation.

This answer accepts a real price: strong necessity. It offers no neutral counterexample to Chalmers’s restricted principle. Its positive ground lies instead in Chapter 6’s independent identity case¹⁴ and the comparative costs counted in Chapters 9 and 10. The remaining standoff concerns which source of evidence governs when ideal conceivability and an abductively supported identity conflict. This book chooses the identity and says so plainly. Chalmers’s argument has not been defeated; its decisive premise has been refused for a reason supplied elsewhere in the book.

What the intuition tracks

The zombie intuition still calls for explanation. Chalmers’s meta-problem asks why creatures like us judge consciousness so puzzling, and why we produce the reports and arguments we do.¹⁵ The acquaintance-recognition account answers part of that question. A physical description can predict the gap-talk without conferring the first-person route whose absence the talk records. The inner-theater picture adds a second influence: once experience gets pictured as a private item over and above the representational work, subtracting the item while leaving the machinery intact feels natural. That picture helps explain why the modal bridge attracts us. It does not show that the conceiving fails.

One intuition outlasts every diagnosis: pain hurts, and a description of nociception does not. Correct. A description describes; it does not put the reader in the foot, the wince, or the moment. A weather report does not get you wet. The missing transmission marks the difference between describing a state and undergoing it. Converting that difference into a second property requires the very modal bridge under dispute.

Two responses worth naming

The ability hypothesis associated with Lewis and Nemirow answers Mary’s knowledge argument, not the zombie argument.¹⁶ It may complement the present account, but no thesis about Mary’s new abilities settles the conceivability-to-possibility inference.

Loar, Papineau, Balog, and others develop the broader phenomenal-concept strategy: phenomenal and physical concepts can remain cognitively isolated while referring to one physical property.¹⁷ The house account adopts a narrower mechanism. Acquaintance with an occurring world-presenting state grounds recognition; description alone does not confer that route. Chapter 11 and B.8 state the limit without camouflage: this mechanism explains the form and permanence of the gap if identity holds. It neither establishes identity nor independently defeats Chalmers’s modal principle.

The ledger therefore closes narrowly. The zombie stays ideally primarily positively conceivable. Chalmers reads that rational achievement as primary possibility. This book reads it as defeasible evidence outweighed by the independent identity case. The gap remains epistemic; the identity keeps the ontology physical.

B.4 The Church–Turing and Church–Turing–Deutsch Theses

(Deep-dive from Chapter 22, Simulation and the Observer [Sections 22.5 and 22.9].) Two celebrated results stand behind much of computationalism’s public confidence, and both get cited as though they settled questions they do not address. They are siblings (one bounds the mathematics, the other concedes the physics), and the rhetorical pattern that exploits them depends on never examining either too closely. This excursus states each accurately, walks the inference each gets drafted into, and itemizes what the computationalist still owes once the borrowing stops.

The Church–Turing thesis (1936). Church and Turing, publishing independently, gave precise definitions of what it is for a function to be effectively computable — workable out, in principle, by following a finite list of mechanical rules — and showed their definitions equivalent. The thesis asserts that this formal notion captures the intuitive one: any function a clerk with paper and pencil could in principle grind out by rule is computable by a Turing machine. That is the whole of it. It speaks of functions and formalisms. It does not speak of brains, or minds, or physical systems at all.¹⁸

The public-facing argument nonetheless runs: anything computable can be computed by a Turing machine; the brain is a physical system; therefore the brain is a computer, and a good enough program will eventually be a mind. Watch the inference try to cross the gap. The brain computes things — but if that means it manipulates discrete symbolic states by rule, neuroscience has found no such thing; that is the commitment in dispute. And if it means only that some computable function can describe the brain’s behavior, it is harmless and trivial, true in equal measure of a hurricane, a stomach, and the solar system. Therefore the brain is a Turing machine — but the thesis says nothing about which physical things are Turing machines; it fixes what functions such machines compute, not which lumps of matter qualify as one. Therefore running the program would be a mind — the most ambitious leap of all, from the brain can be modeled by a computation to running the model would constitute the thing modeled. The thesis is silent on every step that matters. Jack Copeland named the pattern the Church–Turing fallacy; Gualtiero Piccinini supplied its anatomy, and showed several historically influential versions of it — Newell, Pylyshyn, and others — to be unsound.¹⁹

To get from the theorem to computationalism, the computationalist needs three further things the thesis does not give him. He needs a substantive account of when a physical system genuinely computes, one that tells the brain apart from the rock. He needs an empirical case that the brain, on that account, computes. And he needs a philosophical argument that mental life is constituted by such computation rather than merely described by it.²⁰ Each is a separate undertaking. None follows from the theorem. Stripping the thesis of the role it was press-ganged into does not refute computationalism; it returns the burden of proof to where it always sat — on the empirical and conceptual case for treating cognition as computation, to be fought on its own ground.

The Church–Turing–Deutsch principle (1985). The thesis’s physical sibling, due to David Deutsch, is a proposed physical principle rather than a theorem: every finitely realizable physical system can be perfectly simulated by a universal model computing machine operating by finite means.²¹ Deutsch’s own accounting is stricter than the slogan that circulates: classical physics and the classical Turing machine do not satisfy the principle in this strong form — it took the universal quantum computer to make it physically credible — and the looser “to arbitrary accuracy” formulation is a later weakening. Granting the principle here therefore hands the computationalist the strongest simulability premise on offer. The brain is a finitely realizable physical system; therefore the brain is simulable. The physicalist should grant this, or very nearly — denying it means betting on a non-computable physics, a bet a few serious people have placed (Penrose most famously, on quantum-gravitational grounds) but not a debt the computationalist’s opponent has any need to take on.

Granting it costs nothing, because the principle delivers simulability (the existence, in principle, of a formal model that reproduces the system’s behavior), and simulability lives squarely on the representation side of the representation/instantiation line. “The brain is computable” says a description exists; it does not say that running the description anywhere is the thing described. The gap between X is computable and the computation of X is X is the gap Chapter 22’s hurricane opens, restated in the vocabulary of a theorem — and a theorem about the existence of a model cannot, on pain of changing the subject, close a question about whether the model is the thing.

Set side by side, the two results expose the oscillation that does the real argumentative work. Where the thesis bounds the formalism, the principle concedes the modeling: one says these are the computable functions, the other says the brain is among the simulable systems. Neither carries one inch from model to mind. The computationalist’s habit is to claim the strong thing when convenient (being a mind just is running the program) and to retreat, when pressed, to the weak one: the brain is computable, the brain is simulable. And he hopes no one notices that the conclusion always required the strong version while only the weak was ever earned. The mathematics and the physics both survive scrutiny intact. The metaphysics they were made to carry has to walk on its own feet.

B.5 Color and the Reflectance Account

(Deep-dive from Chapter 6, The Identity Claim.)

Chapter 6’s identity claim insists on two kinds of realism: the felt redness is real, and the tomato you reach is the real tomato, its red surface out there at arm’s length, not a proxy. That second claim quietly incurs a third. If the seeing reaches all the way to the tomato’s red surface, there had better be a red surface to reach; a direct realism that delivers the world cannot go shy about whether the world is colored. So the identity claim commits us to color realism: the tomato is red, and its redness belongs to it, not to us.

That sounds like the blandest common sense, and the standing surprise of the subject is how many careful people have felt driven to deny it. Pull the wavelengths apart and the worry sets in. Physics, examining the tomato, finds a surface that reflects light of certain wavelengths and absorbs the rest; it finds no further property, no redness, riding on top of the reflecting. Where, then, is the red? Three answers tempt, and the book turns down all three.

The boldest answer says: nowhere. The tomato is not red, nothing is red, and color is a coat the visual system paints over a colorless world. Call this the error theory, and its defenders take the science seriously — C. L. Hardin, whose Color for Philosophers did more than any single book to bring the empirical study of color into the philosophy of it, presses this debunking case, and Paul Boghossian and David Velleman argue that experience presents colors as intrinsic qualities of surfaces, qualities the surfaces simply do not have.²² Take the view to heart and it saws off the branch the direct realist is sitting on: if every color experience misfires, the seeing never reaches the world at all — it reaches a fiction the brain supplies, and the inner theater walks back in through the side door we threw it out of. But notice exactly where the error theorist lodges the fraud. Not in the reflecting: no one denies that the tomato throws back the long wavelengths and drinks the short ones, every time, in every light, whether or not anyone is watching. The alleged forgery is the extra thing: a simple, intrinsic redness the experience supposedly pins to the surface, a quality over and above anything the physics finds, which no surface has. The verdict of error, in other words, depends on first reading color experience as a promise the physical world cannot keep. The error theorist runs the trial backwards: read the tomato’s testimony as a claim it never made, then convict it of lying. The book will dispute the reading, not the reflecting — and, as the next answer shows, what experience pins to the surface turns out to be a property the surface really has. (The error theory shares its misreading with the primitivism it otherwise resembles not at all: both take experience to promise an intrinsic quality the physical story omits, and divide only over whether anything supplies it.)

A gentler answer keeps color real but moves it into a relation to us: red just is the disposition to look red to normal perceivers in normal light — Locke’s secondary quality, refined. This saves the appearances and pays a quiet price: it defines the witness by the testimony. The worldly property the experience was supposed to reach turns out to be specified in terms of that very experience, so the redness out there smuggles the perceiver back inside the object. For a thesis trying to anchor felt character in something genuinely world-side, that circle runs the wrong way.

That familiar dispositionalism does not exhaust the relational alternatives. A thicker ecological view treats red as the surface’s stable way of interacting with illumination, surrounding colors, movement, and a class of perceivers across the relevant transformations. The property belongs to the surface-in-an-environment rather than to an inner response. Dennett’s claim that colors and color-vision systems were “made for each other,” and Noë’s account of colors as patterns in how surfaces vary in appearance under changing conditions, both give this rival objective accuracy conditions without private paint. Transparency sits comfortably with it: the relation appears as a way the surface is, not as a colored item in the head. The crude circularity charge therefore cannot decide the issue.²³

A third answer honors the phenomenology hardest. Red, it says, is just what it looks like — a simple, irreducible quality the surface presents, with no room for it in the physics: not a reflectance, not a disposition, but a sui generis feature standing there in its own right. John Campbell defends a version under the disarming name of “a simple view of colour.”²⁴ The phenomenology is real and the candor is admirable, but the bill is one the book refuses everywhere else it is handed: a property of the physical world that floats free of the physical story is a small dualism, color’s own private hard problem, and we have spent too many chapters declining that bargain to sign it here for red.

What the identity claim wants is an account that holds onto all three of the things its rivals each keep only one of. Color stays real, against the error theory; it belongs to the surface and not to us, against dispositionalism; and it sits squarely inside the physical world, against primitivism. Red just is a type of surface spectral reflectance — the tomato’s standing physical disposition to throw back a certain profile of wavelengths and soak up the rest. Alex Byrne and David Hilbert, who argue that color science vindicates color rather than debunking it, work the view out in detail; Tye builds the same reflectance physicalism into the representational theory of Chapter 6. One wrinkle disciplines the view. Many physically different surfaces look the same red (metamers, reflectance profiles that part ways on the spectrophotometer and match to the eye), so a color cannot be one reflectance. It is a type: a family of reflectances our visual system groups together as a single color. Here the dispositionalist will cry foul. Has the perceiver not just been smuggled back into the object — the very move the book objected to when it turned dispositionalism down? No, and the difference carries the whole weight. Our responses select which physical properties to file under “red”; they do not constitute the properties filed. Each reflectance in the family is a perceiver-independent physical fact about how a surface treats light, there whether or not anyone looks. Human vision contributes a biologically constrained partition, shaped by coevolution and sensorimotor organization, that groups those facts into colors. Dispositionalism makes the color be a relation to us. Reflectance realism makes the color a physical property while allowing an evolved visual system to constrain which physical families count as colors. That anthropocentric individuation is explanatorily important rather than arbitrary bookkeeping. We carve physical kinds to human interest all the time (“weed,” “continent,” “tropical”) without the carved kinds ceasing to be physical, and the result is a naturalistic color realism that can call the tomato red without finding anything in it over and above its physical surface.

The ecological rival nevertheless identifies a real feature of color vision. A visual system does not recover a spectrophotometer’s curve; it tracks a stable surface through changes in illumination, viewpoint, comparison, and action. Reflectance realism should treat that counterfactual profile as the route by which perception groups and tracks reflectance-types, not pretend the route contributes nothing. A complete color experience presents two coupled features: an invariant object-color and its current manifestation under this illumination, surround, viewpoint, and visual mode. The gray patch may look pinkish against one field and dark gray against another while retaining one surface color; the tomato may change appearance between garden and fluorescent light while vision continues to track it as one red. Those contextual differences need not count as errors. Color constancy gives the preference its point: reflectance supplies the distal stability that remains while proximal light and current manifestation vary. The ecological realist takes the transformation profile as partly constitutive of color; the book treats it as the organization of access and appearance through which a stable color gets presented. Constancy supports that division without proving it. Where two reflectance families sustain the same complete ecological profile for a kind of perceiver, the view should concede the same apparent shade and may group them under one anthropocentric color type. It need not invent a phenomenal difference at a finer microphysical grain. Nor must the color property contain its own appearance. Strong representationalism places phenomenality in the conscious state’s presenting of the property, under a visual mode, rather than requiring redness itself to look red when no one sees it.²⁵

Metamerism is only half of the eliminativist’s empirical brief; the other half cuts the opposite way — not many reflectances under one look, but one reflectance under many looks. Normal observers disagree, stably, about which chips count as unique green (green untinged by blue or yellow) — the variation Hardin presses hardest — so one surface presents slightly different shades to differently calibrated eyes, and no calibration has a claim to be the correct one. If the represented property were the perceiver-calibrated shade, this would re-import the perceiver-relativity the account just refused in dispositionalism. It is not. The represented property is the reflectance-type, which is coarser-grained than any individual’s hue settings, and the account’s verdict on the variation is a concession made in the open: most perceivers slightly misrepresent the fine grain of the type most of the time while correctly attributing the type itself. That exhausts the invariant object-color attribution, though not the context-sensitive appearance profile through which the experience presents it. The result is mild, local, stable illusion at the grain, just what a coarse-grained realism predicts of precision instruments that differ minutely in their tuning. No fact about whose unique green is right ever needs settling, because the experience never attributed unique-green-for-me to the surface; it attributed the type, and the type is there.

The reflectance account also settles the oldest objection in the color literature, the inverted spectrum. Imagine your color experience is systematically inverted relative to mine — what I experience seeing red tomatoes, you experience seeing grass, and the reverse. But we were both trained from childhood to call tomatoes “red” and grass “green,” so our verbal behavior is identical; we both say “that tomato is red” and mean it. Is there a difference between us? Intuitively many say yes: your experience of tomatoes feels, from the inside, the way mine of grass does. You have inverted qualia. If the scenario is coherent, then phenomenal character cannot consist entirely in representational content, for our experiences represent the same properties while feeling different.

This deserves a real response. Notice first that the case smuggles in the very assumption it is meant to support: that phenomenal character can come apart from content. The thought experiment asks us to imagine this and then treats our imagining it as evidence that it is possible — and the evidential status of that imaginative act is what is in question.

A stronger response goes deeper. The full content of a color experience is not just which surface property gets attributed to which object; it includes the color’s relational properties — its location in color space, its similarities and differences to other colors, the way it figures in categorization. Red and green are not arbitrary labels; they differ in how they relate to each other and to the rest of the color solid. If my experience represents red in that full relational sense (as the color that contrasts with green, resembles orange, and not blue) then your genuinely inverted experience would have to represent those relations differently too, and then it might not represent the same properties at all. Inversion may not be coherent for a system as integrated as human color vision. That is not a refutation; it is a reason to doubt how much weight the thought experiment bears. It shows we can describe a scenario in which character comes apart from functional role; it does not show the scenario is a genuine metaphysical possibility rather than a description that unravels on examination. (The inversion case is Sydney Shoemaker’s, “Functionalism and Qualia,” Philosophical Studies 27 [1975]: 291–315; the representationalist response is developed in Michael Tye, Ten Problems of Consciousness, ch. 7, and Alex Byrne, “Inverted Qualia,” Stanford Encyclopedia of Philosophy (2004; substantive revision 2025), with the relational-structure move stated most clearly in Gilbert Harman, “Explaining Objective Color in Terms of Subjective Reactions,” Philosophical Issues 7 [1996]: 1–17.)

And tracking it is just what the experience does. Here is the worldly anchor the identity claim has needed all along: the felt redness consists in representing the surface as carrying a reflectance-type it really carries, and the seeing reaches that type directly because the type is genuinely there to be reached. The relational structure that reply leaned on — red as the color that contrasts with green and resembles orange — now has somewhere to live: the similarity and difference relations among the colors answer to real relations among reflectance-types, not to free-floating qualia a private mind arranges as it pleases. One world, color included. The red is in it, the seeing reaches it, and the felt redness is what the reaching, carried out in this register, comes to.

B.6 The Two Objections to the Mark

Chapter 1 states the mark and leaves its objections here. This excursus carries the full dialectic: the two apparent counterexamples that have shadowed Brentano’s thesis for a century and a half, the inner-theater temptation they invite, and the harder case they leave open.

The Mood That Looks Like a Counterexample

Sit with a diffuse, free-floating anxiety, not a worry bolted to a deadline but the kind that settles over a gray Sunday afternoon and will not say what it wants. Ask what it is about, and the answer comes back nothing in particular. The dread just sits there, apparently aimed nowhere. Searle concedes the point in the first paragraph of his Intentionality: beliefs and desires are directed, but “there are forms of nervousness, elation, and undirected anxiety that are not Intentional.”²⁶ He does offer a reporting-constraints test in the sentence that follows. Crane’s narrower complaint still lands: Searle treats the existence of non-intentional mental states as too obvious to require real argument.

The mood lacks a particular object. Whether it lacks content remains the disputed point. Some intentionalists read anxiety as directed at the world taken whole, where threats loom and the day narrows; others take it to present the felt condition of one’s own body. Angela Mendelovici declines the object hunt. Undirected moods, she argues, really do lack an object, and insisting that they aim at some unusual one gets their phenomenology wrong. They carry content regardless: an unbound affective property, a badness felt without fastening to anything.²⁷ Her case costs the mark its tidiest object-directed formulation while preserving the content-having claim the formulation served.

Anxiety and elation, opposite in feel and both “about nothing in particular,” still turn the subject toward the world in different ways. That difference belongs to content, not to a residue left after content disappears. Strip the directedness and you do not find the bare mood underneath. You delete it. This remains a defeasible verdict rather than a theorem; Chapter 7 weighs what each reading can carry.

The Monster That Isn’t There

The second objection comes from the other direction. A child lies awake certain that a monster crouches under the bed, large and somewhat damp. The fear is real. The monster is not. Yet “nothing” gives the wrong answer to what the child fears: the state has a sharp and specific content, a monster under the bed.

Brentano revived a Scholastic phrase for the object’s status, intentional inexistence. The expression has misled generations of readers. Its in- marks the object’s existing-in the mental act, its immanence, not its nonexistence. The monster teaches the plainer point with which that doctrine is often confused: a thought’s content needs no existing thing outside to answer to it.²⁸ “Fear of a monster” carries no entailment that some monster gets feared. Intentional description does not behave like an ordinary two-place relation whose two ends must both exist.

Meinong gave everything thinkable a standing short of existence: golden mountains, round squares, objects beyond being yet available to thought. Russell’s theory of descriptions sought the same explanatory result without the population boom. The fear has propositional content, and that content stays determinate whether or not the world supplies a monster.²⁹ Nor does the case show that content floats free of the world. What makes this fear a fear of monsters, and what allows it to misrepresent, comes from a history of dealing with an actual world.

The more tempting repair does greater damage: relocate the missing monster inside, as a vivid image on a private stage. Once that move fixes the absent monster, every immediate object becomes an inner image, including the dog the child really sees. The world recedes behind a screen of pictures. Refuse the move and the objection leaves the mark where it found it. Aboutness needs no little item, inner or outer, to fasten onto. Aboutness is the fastening.

What the Two Cases Leave Open

The mood has no particular object and may still carry content; the monster has no existing object and the fear keeps its definite content. Neither case forces us off the mark. Neither closes the ledger.

Bodily pain presses harder. The toothache, the nausea, and the ache that seems to point nowhere answer to neither an objectless mood nor a missing monster. A pain may present a bodily region as damaged, or carry content that commands rather than describes. Which account the case requires belongs to Chapter 7; that phenomenal hurting consists in content of some such kind is the book’s answer, and Chapter 7 must earn it. Chapter 1 needs only the narrower result: its two most familiar counterexamples do not yet defeat the mark.

B.7 Emergence: The Wrong Question, Extended

(Deep-dive from Chapter 11, The Gap That Isn’t There.) Section 11.5 states the verdict in brief — that “emergence” answers the wrong question, since the identity claim denies the two stories the word presupposes. The full treatment lives here: the weak/strong distinction, the British emergentist lineage, Crane’s argument that the vocabulary marks no stable border between emergentism and non-reductive physicalism, and the cautionary case of Tye.

Sooner or later someone raises the word emergence, and it arrives sounding like an answer. Consciousness emerges from the brain the way wetness emerges from H₂O, or a traffic jam from a thousand commuters none of whom intends one — so, the thought runs, we needn’t choose between a brute identity and a second realm; experience simply arises, and the arising does the explanatory work. The word soothes. It also smuggles, and what it smuggles is the very picture this book has spent its length dismantling.

Notice what emergence presupposes before it explains anything. The word posits two stories (a base below and something that rises from it) and then asks how the upper floor relates to the lower. To call phenomenal character emergent already grants that it stands apart from the physical goings-on and gets produced by them as a further fact, an extra the world lays on once the neurons fall into the right arrangement. That is the inner theater again, rebuilt in the vocabulary of complexity science. The identity claim denies the premise outright. Phenomenal character does not arise from representational content of the right embodied, world-directed kind; it consists in it. One fact under two descriptions — not a ground floor and a tenant upstairs. So the right response refuses the question rather than answering it. “Is consciousness emergent?” asks how one thing gives rise to a second thing, and we deny there are two things to relate.

Still, the word will not vanish by fiat, and it keeps a respectable use worth preserving. Distinguish the two senses it runs together.³⁰

Weak emergence runs everywhere and costs nothing. Wetness, temperature, the lattice of a traffic jam, the wheeling of a starling flock: higher-level regularities wholly fixed by their physical base, deducible in principle from below once the identities and bridge facts are in hand — a posteriori deducibility, not derivation from the bare physical description, a distinction Chapter 11 keeps sharp — merely too tangled to compute in practice. A physicalist pockets these without a flinch. If “consciousness emerges” means only this, fine — but then the word carries no anti-physicalist charge, and it has done none of the work its users wanted from it.

Strong emergence supplies the dangerous sense. Here the higher-level property counts as genuinely novel: not deducible even in principle from the complete physical story, and endowed with its own fundamental causal powers reaching back down upon the base. British emergentism in modern dress,³¹ it amounts to dualism in a lab coat, which is why Chalmers parks his “naturalistic dualism” in the neighborhood, and why the panpsychist keeps arriving at the same address from the opposite direction.³² Grant strong emergence for phenomenal character and you have granted that felt quality belongs among the fundamentals — the one thing the physicalism defended here exists to deny.

A deeper reason to distrust the word comes from Tim Crane: examine what an “emergent property” actually amounts to, and the line between full-blooded emergentism and respectable non-reductive physicalism refuses to stay drawn.³³ The vocabulary offers no safe middle berth between the identity claim and the second realm; it offers a fog in which the two keep trading coats. And a cautionary case sits close to home. Tye, in the very book that launched the modern representationalist program, recoiled from the thought that phenomenal consciousness had merely “emerged from the physical world” with no explanation of its emergence whatsoever — such brute emergence, he held, would be tantamount to magic, and he found it very hard to believe.³⁴ The instinct was right. Yet Tye, as Chapter 10 watched, went on to embrace panpsychism, seating felt character among the fundamentals after all. The moral runs subtler than “he grew careless with a word,” and cuts deeper. Rejecting strong emergence will not save you if you keep the framing that forces the choice: either felt character arose from the wholly non-phenomenal or it sat among the fundamentals all along. Refuse the first horn while keeping the dilemma, and the second lies in wait. The identity claim refuses the dilemma itself: it denies there are two things, a base and an arising, to be related in either direction.

Which leaves the real diagnosis. What people hope emergence will bridge is no seam in the world where a new property bolts on. It is the gap in our concepts (the felt distance between a third-person account and the first-person quality it describes), and that gap, as Section 11.3 argued, lodges in us, not in nature. The identity claim does not explain how consciousness emerges. It explains why nothing has to. Ask the question the other way (how does matter become conscious?) and the answer holds its shape: it does not become anything.

B.8 The Phenomenal-Concept Dilemma

(Deep-dive from Chapter 11, The Gap That Isn’t There.) Section 11.7 gives the body-level verdict. This excursus states Chalmers’s C/E dilemma in its topic-neutral form, identifies the horn the book takes, and adds Goff’s separate opacity/translucency challenge.

Chalmers’s master dilemma

Let P state the complete physical truth. Let C state, in topic-neutral psychological terms, the feature invoked to explain our epistemic situation regarding consciousness. Let E state that epistemic situation, also topic-neutrally: the truth-values, justification, and cognitive significance of the beliefs corresponding to our phenomenal beliefs. C cannot simply mean phenomenal acquaintance, and E cannot simply mean knowledge of phenomenal facts, because either formulation would build the contested phenomenology into the premise.³⁵

A successful phenomenal-concept strategy needs three claims. Humans have C. C explains E. P physically explains C. Chalmers argues that the last two cannot hold together.

If P & not-C is conceivable, P does not transparently explain C. The explanatory gap has migrated from consciousness to the special feature of phenomenal concepts.
If P & not-C is not conceivable, any conceivable physical duplicate satisfies C. The zombie has the topic-neutral recognitional architecture, quasi-phenomenal concepts, and corresponding judgments.
Yet P & not-E remains conceivable. Zombie Mary’s new dispositions and indexical judgment need not share the truth, justification, or cognitive significance of Mary’s knowledge of what seeing red is like.
Therefore C either lacks transparent physical explanation or fails to explain E.

This differs from a dilemma between deducible reports and non-conferred acquaintance. Reports matter because they form part of the epistemic situation, but E includes epistemic status. Chalmers’s second horn asks why the same C occurs in Mary and Zombie Mary while only Mary gains true, justified, cognitively significant phenomenal knowledge. Saying that a description predicts their words without giving us Mary’s standpoint does not answer that question.

The horn the book takes

The house account takes the second horn. Its C consists in a topic-neutral physical architecture: an occurring world-presenting state anchors a token demonstrative and enables later recognition of that state-kind. A full physical duplicate has the same architecture. C explains the conceptual isolation, recognitional dispositions, quasi-phenomenal judgments, and gap-talk shared by the duplicate and us. C alone does not explain the whole of E.³⁶

The remaining difference concerns what the concepts refer to and whether the beliefs formed with them are true and justified. Here the identity claim bears the weight. If phenomenal character consists in the physical world-presenting profile argued for in Chapter 6, Mary’s acquaintance-grounded concept reaches that physical state; its exercise can yield true, cognitively significant knowledge. A zombie stipulated to lack phenomenal character cannot share that truth while lacking the very state to which the concept refers. Chalmers can redescribe the zombie’s corresponding concept and belief, but the difference in E then follows from the disputed identity and its denial in the zombie scenario.

This reply accepts Chalmers’s central limitation. The phenomenal-concept strategy does not support the identity, refute the zombie argument, or vindicate type-B physicalism independently. It elaborates the epistemology of a type-B view under the assumption that the identity has been earned elsewhere. Chapter 6 supplies the comparative identity case. The present mechanism explains why that identity leaves a permanent epistemic gap for creatures with our architecture. No circle gets disguised as a victory: C explains the route; identity fixes the referent and truthmaker.

The published physicalist replies often narrow the explanandum in related ways. Carruthers and Veillet distinguish a thick phenomenal reading of E from a thin topic-neutral one. Díaz-León argues that a suitable reading lets C remain physically explicable and explanatorily adequate. Balog disputes Chalmers’s treatment of substantial phenomenal belief. Those replies matter, but the house position needs none to make C explain more than it does. It takes the second horn and posts the limit.

Goff’s opacity/translucency challenge

Philip Goff presses a different objection tailored to a posteriori physicalism.³⁷ He divides concepts into four groups. A transparent concept reveals the nature of its referent. A translucent concept reveals part of that nature. A mildly opaque concept reveals only contingent features that identify the referent in the actual world. A radically opaque concept reveals neither essential nor accidental identifying features.

Goff’s premise remains modest. Thinking of pain in terms of how it feels reveals at least something of what it is to feel pain. Full transparency need not hold; translucency suffices. If the type-B physicalist makes the phenomenal concept radically opaque, first-person thought seems no richer than the property Kev is thinking about now. Reference has been secured, but the thinker has learned nothing of the property’s nature. If the concept is translucent, the physicalist owes an intelligible account of what gets revealed. She cannot say that acquaintance reveals a phenomenal aspect alongside an unrevealed physical aspect without conceding property dualism. Nor can she say that it reveals part of one wholly physical property while insisting that nothing physical is revealed.

The house answer accepts limited translucency. Acquaintance reveals part of the state’s constitutive relational profile: what gets presented, the mode of presentation, and some of the determinacy, field structure, attention, bodily address, and evaluative force organizing the episode. It does not reveal every relation in that profile, including the content-fixing history, nor does it reveal which neural dynamics realize the profile. Ignorance of the realizer is not ignorance of a hidden part of phenomenal essence; property and realizer must remain distinct.

Goff argues that if the revealed aspect belongs to a wholly physical property, the phenomenal concept must reveal something physical. The book agrees de re and denies the further de dicto step. On the independently defended identity, the disclosed relational profile is physical, although acquaintance does not classify it under the concept physical. Seeing a red surface and knowing that one sees a physical property need not arrive through the same concept.

This commits the view to a limited version of Goff’s Thesis of Dubious Intelligibility, so its intelligibility needs an example rather than a label. Sight and touch can each disclose roundness through distinct, noninterdefinable recognitional capacities; neither route supplies a Euclidean analysis, yet both reveal something of one spatial feature. The acquaintance and public routes likewise disclose overlapping aspects of one relational profile through different capacities. Their co-reference comes from Chapter 6’s identity argument, not from introspection. The example makes the two-route structure intelligible; it does not turn that structure into independent evidence for identity.

Present pain supplies the hardest case. The book grants that the thin demand pain occurs now is perfectly satisfied when the subject attends to the pain. That certainty secures the token and its manifest bodily-evaluative profile. It does not show that the concept reveals every constitutive relation, a nonrepresentational constituent, or the complete metaphysical nature of pain.

The resulting view is revisionary in one exact respect. First-person acquaintance reveals more than bare reference and less than complete essence. It preserves the felt datum and its practical significance while denying that the datum announces all of its metaphysics. Acquaintance explains the route. The identity claim still does the metaphysical work.

B.9 Swampman and the History Requirement

(Deep-dive from Chapter 15, Teleosemantics.) Section 15.6 states the verdict in brief: at t = 0 a historyless duplicate can possess a practical stake while lacking representational content; under Chapter 6’s identity claim, he therefore lacks phenomenal character as well. His own history can begin immediately. The full objection-and-reply lives here: Davidson’s case, the Millikan–Dretske bullet, Tye’s dilemma, and Block’s phenomenal-externalist pressure.

Suppose history fixes content: a state means fly, or red, because of what its producer-consumer system was selected and shaped to track. Then picture a creature with no history whatever. Donald Davidson asked us to imagine lightning striking a dead tree in a swamp and assembling, by accident, a molecule-for-molecule duplicate of a living human being. Swampman has your visual cortex to the last synapse. He hauls himself out of the muck, turns toward a ripe tomato, and his eyes do exactly what yours do. Surely, the objection runs, a perfect copy of you sees the red. If content must be earned through history, teleosemantics seems forced to say that he does not.³⁸

The book accepts the price. Swampman’s present organization gives him a stake: he is a precarious, self-maintaining system for whom some trajectories preserve the organization and others destroy it, and he regulates himself with respect to those boundaries. That may ground autonomy, practical interest, and welfare. It does not supply a semantic address. Nothing in his nonexistent history has selected those visual states for tracking tomatoes, redness, food, or any other candidate rather than its reliable companions. The distinction developed in Chapter 15 therefore matters at its hardest point: the stake explains how success and failure may be good or bad for the organized individual; producer-consumer history fixes what a state is correct or incorrect about. At t = 0 Swampman has the first and lacks the second.³⁹

The identity claim makes the result more severe, not less. If phenomenal character consists in world-involving representational content of the right kind, then a duplicate with no content has no phenomenal character. The lights are, for that first instant, off. This is phenomenal externalism without a cushion: molecule-for-molecule duplication of the current body does not duplicate experience when the relations that fix content are absent. Block’s protest identifies the cost accurately; it does not expose an internal contradiction. An intrinsic mental paint would spare Swampman, but Chapter 6 has already declined it.

The verdict concerns the instant, not the future. Swampman’s first perception-guided action begins a learning and consumer history. As his organization recruits, corrects, and stabilizes states in encounters with tomatoes, food, danger, and other particulars, that history can establish determinate correctness conditions. The present stake makes the consequences matter to him; it does not create the content. So the account does not require ancestry or carbon. It requires a content-fixing history and producer-consumer organization.

The comparison with artificial systems now cuts differently. Swampman begins with autonomous organization and no content-fixing history. A trained model begins with a rich causal history, learned sensitivities, and task-shaped producer-consumer structure that may support genuine mechanism-level content. The model may still lack determinate distal reference, whole-system understanding, autonomy, or consciousness; none follows merely from granting content at one grain. Swampman’s history can begin through consequential engagement. The machine case turns on what its training and corpus-mediated history fix, how content-bearing mechanisms integrate across the larger system, and what additional organization supports autonomy or experience. The absences are different, and neither requires a proprietor to mint meaning.

B.10 Tye’s Sharpness Argument and the Concept/Property Dispute

(Deep-dive from Chapter 10, The Case Against Panpsychism.) Section 10.4 states the book’s verdict in brief: the felt or conceivable sharpness of consciousness does not by itself establish a sharp joint in nature. The full dispute lives here — the two-claims anatomy of the inference, the demonstrative-concept diagnosis, the exchange with Michael Antony, and Tye’s insistence that sharpness belongs to the property rather than merely to our way of conceiving it.

Tye’s late view needs stating in its developed form before the book turns on it. In the 2021 account, sharp fundamental consciousness does the panpsychist grounding work, while determinate phenomenal character remains tied to representational content and ordinary consciousness can inherit vagueness from its functional organization. The 2024 restatement presses the essential sharpness of phenomenal consciousness more directly and pairs consciousness plus content with a transfer proposal for complex subjects. The formulations differ, but the recurring pressure is clear: our inability to frame a genuinely borderline phenomenal case is treated as evidence that the phenomenal base itself cannot be vague.

Set the two claims the book disputes side by side. First: we cannot imagine a borderline case of phenomenal consciousness. Second: phenomenal consciousness, or the consciousness that grounds it, is sharp* in nature. The first reports something about what our phenomenal concepts permit us to frame; the second concerns the grain of the property itself. Tye argues that the first reveals the second. The book denies that entitlement rather than denying the datum.

The structure resembles the conceivability arguments already examined through Mary and the zombie: an epistemic premise about what we can frame is asked to support an ontological conclusion about what nature can contain. The unimaginability is real. What remains disputed is whether it reveals the property’s grain or the limits of the concept with which we approach it.

The book’s alternative starts from how phenomenal concepts operate. We deploy one from the inside, by having the state and attending to it. It does not work as a description with adjustable conditions of application; it behaves more like a demonstrative — this, the thing I am now undergoing. You cannot easily form the thought there is no fact of the matter whether I am now undergoing this from the inside, because the having of the state supplies the basis of the concept’s application. The concept is all-or-nothing in use. That need not settle the grain of what it tracks: a thermometer that reads only “hot” and “cold” would not show that temperature comes in exactly two sharp kinds. Michael Antony presses this line — a sharp concept may track a vague worldly kind — and Tye refuses it, insisting that sharpness belongs to consciousness itself rather than merely to the concept.⁴⁰ The argument therefore reaches a genuine disagreement over what first-person conceivability reveals, not a verbal misunderstanding.

If the book’s diagnosis is right, phenomenal consciousness can consist in vague physical-representational goings-on even while recognitional concepts present their application as all or nothing. Panpsychism then loses one of its principal motivations; it has not been refuted by conceptual analysis. If Tye is right that sharpness belongs to the property’s essence, the book’s vague physical threshold cannot be the whole story. That is the live fork. Chapter 10 prefers the first branch because it fits the identity case already developed and avoids a fundamental phenomenal base plus transfer law; the preference remains comparative and defeasible.

B.11 Functionalism’s First Two Replies

(Deep-dive from Chapter 20, Functionalism and Computationalism.) Section 20.4 meets the first two replies to the homunculus-nation in brief and points here for the full hearing: the bullet-biter who accepts that the nation feels, and the dilemma that world-involving content is either mere relabeling of the functionalist’s gap or closet dualism. (Objection 3, Chalmers’s organizational-invariance argument, stays in the chapter as the reply to beat.)

Objection 1: Bite the Bullet

The orthodox functionalist does not flinch. The nation feels pain (yes, however bizarre that sounds) and your refusal to believe it is not an argument but a prejudice. Our intuitions about who can feel were trained on creatures with faces, fur, and eyes; they are parochial witnesses, hopeless judges of radio networks and continental populations. We once found it obvious that the sun went round the earth. The strangeness of the homunculus-nation is a fact about the narrowness of our imaginations, not about the distribution of consciousness. Bite down, and the case loses its teeth.⁴¹

Reply to Objection 1. There is something to this, and I want to grant it before I deny it: intuition-pumping over exotic systems proves little on its own, and a reader who found the whole case resting on “but it just seems weird” should walk away. The bullet-biter commits no fallacy and incurs no special obligation to name a mechanism beyond the complete functional organization already specified. If functional role exhausts the mental, the nation has pain; that is the view, not a refusal to give one. The objection earns weight only inside the book’s cumulative comparison. Chapters 5–16 argue independently that phenomenal character is world-involving content and that content depends on more than an internally specified role. Against that background, the nation case exposes a comparative cost: the functionalist must attribute a unified felt point of view where the stipulated internal organization supplies no independent content-fixing history for the assembled whole. The book’s theory explains why that attribution remains unearned; the functionalist accepts it. No single intuition settles the contest. The question is which package explains more while asking us to swallow less. The nation, in short, is diagnostic rather than demonstrative: it exposes the sufficiency thesis’s price to a reader who already holds Chapters 5–16, and its whole weight is the weight of that accumulated case. The positive ground is the one the intuition never supplies: the internally specified organization leaves unsettled whether the nation’s states possess the complete world-involving, mode-sensitive, perceptually organized, and poised structure the identity claim requires. Giving the assembled whole an appropriate content-fixing history and perceptual organization could change that evidence; the bare wiring stipulation cannot. This is the same bounded conclusion Chapter 20 prices, in the same order.

Objection 2: You Have Only Renamed the Mystery

A subtler critic grants that the nation’s states are not about anything and presses a different worry. Fine — but “world-involving content” is just a new label on the old gap. You say the wiring lacks aboutness; I say aboutness is precisely what we do not understand, and you have explained the obscure by the more obscure. Worse: you claim the missing ingredient is not a non-physical extra, yet “genuine meaning” looks suspiciously like the very thing the dualist was gesturing at, now wearing a naturalist’s coat. Either grounding is more functional organization (in which case you are still a functionalist and owe the nation its pain) or it is something extra, in which case welcome back to dualism.

Reply to Objection 2. This dilemma is the sharpest thing in the neighborhood, and the answer to it carries the case, so I will be careful. Take the second horn first. Is world-involving content “something extra” in the dualist’s sense? No — and the difference is not a dodge. The dualist’s extra is intrinsic: an inner quality, a private glow, something that could in principle be present or absent with every physical and relational fact held fixed. Grounding is the opposite kind of thing. It is relational and wholly physical: a matter of which worldly features have, over a causal history, fixed which states and guided which consumers. Nothing immaterial is added; what is added is a world, and the system’s tracked traffic with it. You can be the most thoroughgoing physicalist alive and insist on grounding, because grounding is made entirely of physical relations — just relations the functionalist declined to count.

Now the first horn, which is the one with real bite. Isn’t grounding just more function — long-armed function, reaching out to include the causal links to the world? And here I concede the words while denying the victory, because the concession is fatal to the thesis under examination, not to mine. Yes: you may, if you like, call world-involving content a species of “wide” or “long-arm” functional role, the functional story enlarged until it takes in the states’ causal-historical bonds to the very things they are about.⁴² Call it that. But notice what has happened. Machine functionalism, the thesis Chapter 20 set out to test, the thesis that makes the strong-AI claim live, individuates mental states by their internal role: inputs, outputs, and relations among inner states, the whole specification readable off the wiring diagram with the world switched off. That thesis is the one Block’s nation satisfies to the letter, and it is the one that fails. The moment you reach past the diagram and let the world in to fix content, the moment two internally identical systems can differ in what their states mean because they are plugged into different environments with different histories, you have conceded the precise point at issue: organization alone, the wiring with the world bracketed, does not suffice. Whether we then file the enriched view under “wide functionalism” or under “world-involving representationalism” is a question of shelving, not of substance. The wiring diagram, read as the functionalist reads it, is no longer the whole story. That is all Chapter 20 needs, and the relabeling hands it over.

B.12 The Stake: Sources of a Term

(Background to a term the book leans on from Part Two through Part Four.)

A system’s stake in the world — the continued realization of its organization riding on its own regulation under environmental pressure — names a condition with a settled intellectual pedigree, not a coinage of mine. Gathering the sources in one place shows the word carrying recognized work rather than smuggling in a metaphor. Whether any state thereby answers to the world remains a further question.

I take precarious adaptive closure as the best current account of autonomy and practical stake, not as the only possible account: a better physical explanation could replace it while leaving questions of content, understanding, and consciousness to their own evidence.

The taproot is Hans Jonas. In The Phenomenon of Life (1966) the organism’s existence arrives not as a possession but as a task: metabolism holds a living form together only by continually spending and replacing the very matter that composes it, so that the form persists precisely by never resting in any of it. Jonas named the result “needful freedom” — a freedom from any particular parcel of matter bought at the price of needing matter as such. Existence that must be earned, moment by moment, against dissolution gives the first and deepest form of having something at stake.

The enactivist tradition supplied the mechanism and the standard vocabulary. Building on the autopoiesis of Humberto Maturana and Francisco Varela, Ezequiel Di Paolo separates a system that merely holds itself together from one that also registers how it fares and acts on the reading. A living system counts as adaptive when it regulates itself with respect to the boundaries of its own viability (“Autopoiesis, Adaptivity, Teleology, Agency,” Phenomenology and the Cognitive Sciences 4 (2005): 429–452, sec. 3, where the definition appears in display form). Such regulation responds differentially to trajectories toward and away from the viability boundary. It counts as precarious when its constituent processes, taken in isolation, would run down or extinguish in the absence of the organization they sustain — so that those processes stand not merely conditioned by the network but enabled by it. The precariousness condition receives its technical definition four years later, in “Extended Life,” Topoi 28 (2009): 9–21, sec. 4, where Di Paolo introduces it precisely to keep operational closure from going trivial; the 2005 paper uses the word only informally, in the course of reading Jonas. Evan Thompson calls the wider activity “sense-making” (Mind in Life, 2007); the narrower claim here concerns only the functional standard that regulation establishes.

John Haugeland fixed one proposed bearing on intentionality. Understanding, he argued, requires that a system give a damn — that it care about itself and its situation — so that a device performing formal operations while caring about nothing understands nothing, however fluent the performance (“Understanding Natural Language,” Journal of Philosophy 76, no. 11 (1979): 619–632; reprinted in Having Thought: Essays in the Metaphysics of Mind, 1998). His formulation remains a serious challenge to accounts that reduce understanding to formal performance. This book now treats caring as a possible enrichment of autonomous understanding, not as a demonstrated prerequisite for content or cognition.

Brian Cantwell Smith carried a related requirement into the age of machine learning. His distinction between reckoning, prodigious ungrounded computation, and judgment, the answerable and world-committed kind, rests on an “existential commitment” that present systems lack (The Promise of Artificial Intelligence: Reckoning and Judgment, 2019). The distinction keeps pressure on any account that infers judgment from performance. Whether existential commitment proves necessary remains part of the dispute rather than a premise the book can simply collect.

Where a term more technical than stake is wanted, precariousness names the condition. Normative precariousness needs more care. It can name a system whose own organization supplies a standard of functional success and failure; it should not suggest that bare existence generates a categorical ought.

The causal-success objection

Luca Corti has pressed the point directly. Organizational closure, he argues, can explain the causal conditions under which a system exists, but an explanation of why a system persists does not entail that it ought to persist or that any component ought to perform its role (“Organizational Normativity and Teleology: A Critique,” Synthese 202 (2023): 96). Put bluntly: a hurricane also has conditions of existence. Describing what would make it dissipate does not give the weather obligations. If a hurricane displayed precarious adaptive closure, it too would earn the thin functional standard defended here; ordinary persistence alone earns nothing.

The objection defeats a stronger claim than this book needs. The stake does not derive categorical normativity from viability. It supplies a theoretical standard of functioning with three features: the standard operates in the system’s own regulation; departures alter that regulation before the system disappears; and successful regulation helps maintain the organization that produces and sustains the regulator. These remain descriptive facts. Nicholas Shea makes the same deflationary point about teleosemantics: correct and incorrect representation can form a descriptive distinction to which stronger norms may later apply; biological well-functioning need not carry an irreducible ought (Representation in Cognitive Science, 2018, sec. 6.5).

The term norm therefore marks an organization-relative functional standard, not a reason that binds a rational agent. Readers who reserve normativity for the latter can substitute internal functional standard throughout. What matters is that the organization differentiates trajectories in its own activity, and the consequences return to the organization whose activity produced them. Viability may ground practical interests and welfare. It does not supply a representational satisfaction condition. Content depends instead on a history and producer-consumer organization that privilege one accuracy condition over its rivals.

Where the system ends

That last sentence raises the boundary problem. Why treat a frog or robot as the system and its pond, server, or data center as a platform? Skin, legal ownership, and the engineer’s block diagram settle nothing. The present account locates the boundary through closure of constraints. Maël Montévil and Matteo Mossio define the relevant organization through reciprocal dependence: each constitutive constraint contributes to maintaining others that, in turn, maintain it (“Biological Organisation as Closure of Constraints,” Journal of Theoretical Biology 372 (2015): 179–191). External conditions can remain causally indispensable without entering the closure. Oxygen enables a mammal; the mammal does not maintain atmospheric oxygen as one of its constitutive constraints. A cloud scheduler may preserve a model process; if the model’s activity does not help maintain the scheduler or the realizing machinery, the dependence remains one-way. Montévil and Mossio’s apparatus adds the operational mark the hard cases need: a constitutive constraint is conserved at the timescale of the process it harnesses and is itself maintained, on a slower loop, by what the organization does. The scheduler fails from both directions at once — nothing the model does maintains the scheduler on any timescale, and the scheduler’s persistence owes nothing to how well the model tracks. Fuel, oxygen, and rented compute are consumed or supplied across the boundary; constraints are held in place by the loop. That is the difference between what a system runs on and what it is.

The test therefore asks which constraints belong to the reciprocal maintenance network over the time scale on which the candidate organization persists. Include them even when they cross a chassis or a building. Exclude merely enabling conditions, even when their removal would kill the system. The result should remain stable under modest changes of descriptive grain; if a small redescription flips every relation from constitutive to enabling, the evidence does not yet fix a boundary. If an artificial agent acquired energy, repaired its hardware, allocated compute, and preserved the infrastructure that in turn preserved it, the larger distributed organization could constitute an autonomous individual with a practical stake. Moving the boundary outward would then change the verdict because the causal organization changed, not because the analyst grew generous.

Closure can also nest. Cells maintain themselves inside organisms; organisms rely on symbionts; social systems recruit already minded members. Montévil and Mossio treat such cases as hierarchical closures rather than demanding one metaphysically privileged seam. This supplies a disciplined test for autonomy, not a semantic allocation rule. The level at which content supports belief, understanding, or action depends on a further pattern of consumer use, correction, memory, inference, planning, and control. Location and causal dependence do not confer whole-system attribution. The same signal can enter a cell’s economy in one use and the organism’s in another.

The social case deserves the same discipline. A company can depend on its workers and the workers on the company without their minds automatically composing one further subject. Economic reciprocity alone is not enough, because reciprocal services may still connect independently maintained agents. Look separately for the closure relations that bear on organizational autonomy and for system-level cognitive operation: error detection that changes organization-wide policy, conflict resolution among competing processes, resource reallocation, and learning that changes how those processes work across turnover in individual members. A firm, colony, automated factory, or distributed robot may therefore constitute an autonomous organization, contain genuine content-bearing mechanisms, or both. Neither achievement yet establishes an understanding or conscious subject; that later question asks whether content enters a flexible cognitive economy capable of integrated inference, planning, and revision. The result may be plural, graded, or empirically uncertain, but it is not arbitrary: reciprocal maintenance bears on autonomy, while integrated consumer work bears on whole-system cognitive attribution. Section 15.6 states the strict etiological limit: present organization without selection or learning history supplies no representational content. On the system/platform application, see Chapter 24.

B.13 The Ladder of Distinctions

(Deep-dive from Chapter 24, The Conscious Robot.)

The charge that the constructive case runs too quick almost always comes of holding two or three of these notions in a single hand. Several crowd the same neighborhood, but they do not form one ladder and each has one job.

Persistence comes first and settles nothing. A rock persists; a file may persist longer. Self-maintenance adds activity directed toward continuation. Thermodynamic self-maintenance raises the bar: the system’s activity helps constitute the continued existence of the physical organization doing the activity — and the candle flame, the textbook dissipative structure, clears even this rung, melting the wax that feeds it and driving the convection that supplies its oxygen. In organisms, metabolism performs this work. The general condition does not require life or carbon. Adaptive regulation adds the ability to register movement toward and away from viable conditions and vary the response. Here success and failure first become possibilities enforced by the system’s own dynamics.

The remaining dimensions do different jobs. Practical stake arrives where adaptive regulation helps preserve the system’s own organization; this book’s current working account may be replaced without changing the autonomy and welfare questions it serves. Address arises through selection, learning, tracking, and producer-consumer use that establish what a state is correct about. Whole-system understanding requires those contents to enter a flexible economy of memory, correction, inference, learning, planning, and action. Poising and perceptual organization mark the further condition the identity claim associates with consciousness. The separations block illicit promotions: persistence is not self-maintenance; self-maintenance is not adaptivity; stake is not content; content in a mechanism is not whole-system understanding; understanding is not yet consciousness.⁴³

Run three cases. A candle preserves its flame-pattern through a stream of fuel the flame itself liberates — melting the wax below it, driving the convection that feeds it air — a genuine thermodynamic self-maintainer, and conceding the flame this much is no embarrassment: it shows why adaptive regulation must be considered separately. The flame cannot register movement toward its own extinction and vary its response; it burns identically toward the guttering end. A starfish mends itself with matter and energy it acquires and organizes, the repair serving the continuation of the very organism doing the repair. It clears the thermodynamic and adaptive floors, giving it a practical stake; its evolutionary and individual history separately supplies addresses for its states. A conventional virtual-machine process can react to a represented health variable with great sophistication while the host platform maintains the physical process independently of that success. Training may still give the health state genuine content, while the architecture supplies little evidence of autonomous self-maintenance. Remove the backup and nothing important changes. Couple the process’s activity to energy acquisition, repair of the realizing machinery, and preservation of the boundary on which that machinery depends, and its autonomy changes even if a copy can still be made. The distinctions ask what the system does, not what the administrator can copy.

B.14 Why Restorability Is the Wrong Criterion

(Deep-dive from Chapter 24, The Conscious Robot.)

The abandoned criterion — this book’s own earlier line, given up in the open rather than quietly rewritten — had a clean dramatic shape: a genuine understander would be something you could no longer back up. The shape was cleaner than the argument. Three objections expose why.

First, consider two physically and organizationally identical systems. An administrator has saved a snapshot of one and neglected to save the other. If the file changes the systems’ semantic standing, content depends on an event wholly outside both systems. If the possibility of a snapshot changes it, the criterion simply redescribes copyability as guilt. Neither route explains why one running system has content, autonomy, or consciousness and the other does not. A backup may reveal the architecture; it cannot constitute the absence.

Second, any finite physical system is copyable in principle at some level of description. Swampman makes the point with unnecessary weather. If perfect duplication voided a stake, better scanning technology would steadily drain autonomy or welfare from organisms without altering a molecule in them. The absurdity belongs to the criterion, not the scanner. A duplicate raises questions about which individual continues and what history the newcomer inherits. It does not reach backward and revise the original’s regulatory organization or content.

Third, take the identity question seriously. Perhaps a restored process is the same agent resumed. Perhaps it is a successor with inherited memories. Perhaps, with Parfit, identity is not what matters. Even the friendliest horn changes nothing: grant that continuity suffices and the restored process counts as the same agent resumed — what it inherits is form and record, and whether the running system’s activity now maintains the organization realizing it remains exactly as open as before the copy. Every horn leaves the substantive questions standing. Does adaptive activity help maintain the physical organization realizing the system? What history fixes the addresses of its states? How are those states integrated and used? A theory of copying cannot answer any of them.

The old criterion did contain one good observation. Current software agents are restorable because their persistence usually resides in infrastructure outside their regulatory loop. The same platform keeps running whether the agent tracks well or badly; the represented danger and the physical continuation come apart. That explains why a checkpoint often serves as useful evidence. Remove the checkpoint and the architecture remains. Redesign the loop so the agent’s activity acquires energy, repairs the realizing machinery, and maintains the boundary on which that machinery depends, and the architecture changes even if a copy can still be made.

The replacement is therefore plural: on this working account, thermodynamic self-maintenance coupled to adaptive regulation bears on autonomy and practical stake; world-involving history and producer-consumer use fix content; integration bears on whole-system understanding; perceptual organization and poising bear on consciousness. Copyability concerns identity and succession. These debates may meet in an actual machine, but they do not share a premise.

B.15 Teleosemantics: The Technical Rounds

(Deep-dive from Chapter 15, Teleosemantics.) Section 15.2 states the verdict in brief — the systematic case shows function necessary but not sufficient — and Section 15.3 keeps Fodor’s coarseness objection in compact form. The full machinery lives here: Mendelovici’s reliable-misrepresentation argument worked through color, and Fodor’s asymmetric-dependence round.

This picture has a problem, and the problem has a name. Angela Mendelovici named it in a 2013 paper, careful and unkind in equal measure: reliable misrepresentation.⁴⁴ Her argument is not the frog argument. She is explicit that the disjunction problem (the problem of explaining how the frog’s state can mean fly rather than fly-or-BB) is not the problem she means to raise. Her argument requires the misrepresentation to be systematic, not occasional. To see what that requires, consider color.

When you look at a ripe tomato in good light, your visual system tokens a state representing the tomato as red. The phenomenology of the case (and here I make a controversial assumption I will return to) presents redness as a vivid, unified, intrinsic-seeming property apparently sitting on the tomato’s skin. According to the physics, the tomato has no such property. Surface reflectance profiles interact with incident light, the light with your photoreceptors, the signal with downstream neural circuitry. Somewhere in that cascade, your visual system produces an experience as of redness in this thicker sense. The property your experience attributes to the tomato does not, on a defensible physicalist view, exist out there waiting to be picked up. Your visual system reliably represents the tomato as having a property it does not have.

Hold the example loosely. The argument does not depend on settling whether colors are surface reflectances, dispositions, or eliminated altogether. Nor does it depend on the contested assumption I just flagged — that color phenomenology presents redness as intrinsic rather than relational. That assumption names the Galilean intuition that drives a great deal of philosophy of mind, and direct realists deny it. The argument depends only on the broader possibility (which Mendelovici argues is more than a possibility) that some of our perceptual representations run systematically off in this way. Not occasional slip-ups. Not exotic illusions. Standing features of how the system operates.

Now ask what teleosemantics has to say about the matter. The teleological story tells us that a representational state means whatever its evolved function says it means. Its function gets fixed by what the state was selected for in the relevant ancestral environment. So what was your color-representing apparatus selected for? Presumably, it was selected because creatures whose visual systems produced these states reproduced more successfully than creatures whose visual systems did not — by, for example, discriminating ripe fruit from unripe, or warning coloration from camouflage, or kin from non-kin. The function is to discriminate, to categorize, to enable adaptive behavior.

Here teleosemantics walks into the wall. If the function of your color-representing state is to discriminate certain surface properties under certain lighting conditions, then your state means exactly that. It does not misrepresent the tomato at all. It accurately represents the tomato as having the relevant discriminable surface property, which the tomato in fact has. The vivid, intrinsic-seeming redness drops out of the content. Asked what the state represents, the teleo machinery hands back an answer that makes systematic misrepresentation practically unavailable: the state means whatever feature was tracked reliably enough to be selected for, and reliable tracking is part of what fixes the content. Mendelovici puts the modal claim carefully: tracking theories make reliable misrepresentation “practically impossible,” and that counts as a major consideration against them. The point is not that no single coherent description of reliable misrepresentation can be cooked up within the framework. The point is that the framework’s resources for handling such cases fall systematically short.

The conclusion sits poorly with the phenomenology it was supposed to explain. We do not seem to experience the tomato as having a complicated relational surface property. We seem to experience it as red in a way that points beyond any relational story. To deny that our color experience represents anything over and above the discriminable surface property is to flatten the phenomenology into its functional shadow. Strong representationalism, which holds that phenomenal character consists in representational content, needs an account of content rich enough to include whatever in color experience goes beyond reliable tracking. Michael Tye saw this clearly enough to spend a whole book defending the view that color experience represents real, mind-independent properties of surfaces, because the strong representationalist’s commitment to content-realism leaves no other option once the Galilean intuition is in play.⁴⁵ Whether one finds Tye’s color realism convincing or not, his project illustrates the pressure: tracking semantics alone, without further substantive commitments about which properties experience represents, cannot underwrite the phenomenology.

A defender of teleosemantics has replies, and the literature since Mendelovici has not stood still. Karen Neander has developed an “informational” teleosemantics that tries to fix the indeterminacy worry by appeal to natural-factive information and second-order similarity relations.⁴⁶ Peter Godfrey-Smith, a friend of the program, has stepped back to ask whether it has stayed in proper contact with the phenomena it was supposed to explain.⁴⁷ His diagnosis treats indeterminacy as a deep structural worry rather than a local puzzle. None of these moves dissolves Mendelovici’s problem. They redistribute it. The teleo theorist can tighten the function, in which case content threatens to track only what the system reliably picks up, leaving the phenomenology unexplained. Or the teleo theorist can broaden the function, in which case content turns indeterminate, since several incompatible candidates each fit the historical record. The tradeoff holds.

I take the right lesson from this to be modest. The reliable-misrepresentation argument does not show that teleosemantics is false. It shows that teleosemantics cannot be the whole story about content. Evolution can do a great deal of work in explaining why our representational states exist, why they have the functional profiles they do, and why misrepresentation in principle remains possible (a state that performs its function only sometimes obviously can fail to perform it). What evolution cannot do, by itself, is fix the content finely enough to allow systematic mismatch between how things seem and how things are. For that, we need a story about how representational vehicles carry content beyond what natural selection has ratified — a story that lets the visual system represent the tomato as red even when nothing red, in the relevant thick sense, is on offer.

Such a story exists, and its outline is enough to put to work here. It involves the relational, world-directed structure of perceptual states; the role of recognitional capacities that go beyond what selection alone fixed; and the contribution of conceptual and inferential context to what a given perceptual state means for the perceiver. Dretske, in Naturalizing the Mind, distinguishes the merely indicating properties of a state from its full representational properties, and allows the two to come apart in ways that selection-history alone cannot adjudicate.⁴⁸

Run the mechanism on the hard case, not just the easy one. What fixes content beyond ratification is chiefly the consumer side: the state feeds capacities that treat it as red in the thick sense — matching surfaces across changes of light, sorting by sameness of look, expecting the strawberry to match the tomato — and content follows what its consumers have used it as: the accumulated record of recruitment, correction, and stabilization in the system’s own history, not the bare present disposition. (The distinction matters here, and B.9 collects it: at t = 0 Swampman has the consumers and none of the record, which is why present dispositions alone hand him no content.) That gives systematic mismatch somewhere to stand: were the world to lack the thick property, the state would still say it, the saying having been fixed downstream of the tracking. And the book should show its own hand here rather than leave the question open. Chapter 6 took its stand: colors are real — reflectance-types, out on the surfaces — so by this book’s lights the tomato satisfies what the experience actually attributes: the reflectance-type, presented non-relationally, correct at the type’s grain even where a perceiver’s fine tuning misses (B.5 prices that concession). The systematic case then stays what Mendelovici needs it to be and no more: a live possibility, pressing on tracking theories, not a settled fact about human vision. Nor does the pressure need color for a home. If any register runs systematically rich, the affective one is the likelier candidate — a standing anxiety presents harmless mornings as menacing, attributing what the scene does not instantiate (B.6 above; Chapter 7 gives moods their content) — and Mendelovici’s modal point never needed an actual case in any register: a theory that makes standing mismatch impossible in principle is refuted by the possibility, not the incidence. The richer story’s yield is conditional but genuine. If the phenomenology anywhere outruns the physics, the framework no longer defines the mismatch away.

Jerry Fodor saw the difficulty for teleosemantics earlier than its friends did, and pressed it into the strongest standing rival the naturalistic program has.⁴⁹ Fodor spent a career arguing that the mind runs on a language of thought; he put the point with the frog. Suppose the frog’s detector has the biological function of getting the frog fed. A frog whose inner state means fly and a frog whose inner state means fly-or-BB-or-any-small-dark-moving-speck are, in the ancestral pond where the only small dark moving specks were flies, as fit as each other. They catch the same things, leave the same number of descendants, fare identically under selection. Biological function, being a matter of what got the lineage fed, simply cannot tell the two contents apart, because the fitness comes out the same whichever way you read the state. This bites deeper than the BB that orthodoxy already handled: no BB haunted the ancestral pond, but the rival content here runs co-extensive with fly in the very environment that did the selecting, and equal fitness leaves nothing to choose between them. So function is too coarse to fix the address. That is a real result, and it drove Fodor to build his asymmetric-dependence theory instead: meaning, he proposed, gets fixed not by what a state was selected to track but by a counterfactual pattern of dependence among the things that can cause it, so that the BB tokens fly only because flies do, and not the other way around. The teleosemanticist cannot wave this away. Survival underdetermines content as Fodor said.

And the rival does not get off lightly either: Fodor’s asymmetric dependence arguably cannot spell out its counterfactuals without reaching for the very semantic and functional facts they were meant to ground.

Round Three: What fixes fly, if anything?

Granting Fodor’s coarseness result does not leave every candidate tied. Selection is a causal process, and causal explanations do not permit free substitution among co-extensive descriptions. The frog survived because it captured nutrition, not because it captured blackness. That weighs against small dark thing as an account of the task function even if every nutritive target in the ancestral sample happened to look dark. The point does not yet exclude fly-or-speck. A branch cannot lose merely because no specks happened to occur. It loses if specks figure in no Normal condition of consumer success, lack the nomological and counterfactual support carried by the fly branch, and improve no causal explanation of how the producer-consumer system performs its task. A logically available disjunction is not yet an explanatory one, but historical silence alone does not decide the case.

Nicholas Shea’s varitel semantics makes the positive method explicit.⁵⁰ First identify the robust task function whose performance explains why the mechanism stabilized. Then identify the tracking relation that figures without mediation in the mechanism’s performance of that task. Finally place the mechanism among the organism’s other capacities. If the frog uses distinct circuits for worms and other edible targets, the correlation with flying insects may explain why this circuit exists and how the wider feeding economy divides its work. Fly at (x, y, z) may then earn its grain. If the same circuit and consumer treat every small moving nutritive item alike, moving prey at (x, y, z) may be all the content the evidence supports. Edible moving thing and prey do not lose merely because English speakers find fly neater.

This answer is realist without pretending to metaphysical omniscience. Content consists in the satisfaction condition supported by an intervention-stable causal explanation of the producer-consumer system across actual and relevant counterfactual tasks. Where one candidate tracks a difference that changes performance and another hides it, the architecture supplies discriminatory structure; the theorist merely gets it right or wrong. Where two candidates remain co-extensive and equally explanatory across every such difference, no further fact need choose between them. The state has content at the grain its use supports and remains indeterminate below it. The demand that a frog’s signal encode a philosopher’s preferred natural-kind concept was never part of naturalism’s job description.

The BB case now comes out without sleight of hand. If the content is moving nutritive prey at this location, the BB token is false because the pellet does not satisfy that condition. If the content is only small moving object at this location, the token is accurate and the feeding failure occurs downstream. Which verdict applies depends on the established task structure. Either way, correctness and successful performance remain distinct: a correct state can guide a clumsy strike, and a false state can produce a lucky meal.

Round Four: Does training supply a selection history?

It can. Fintan Mallory argues that backpropagation meets a consumer-based teleosemantic standard of selection: successive parameter configurations reproduce relevant structure under a lawful update rule, loss differentially preserves changes, and internal producer-consumer mechanisms thereby acquire proper functions (“Teleosemantics for Neural Word Embeddings,” Mind & Language (2026), doi:10.1111/mila.70037). On his analysis, even a simple word2vec model contains intentional signs with original contents.⁵¹ The claim cannot be dismissed by saying the model has no evolutionary past. Training is a real causal history of the artifact. Checkpoint selection belongs directly to the training lineage; selected functions may pass by causal descent to producer-consumer mechanisms in a deployed model; the deployed token has not thereby selected itself. This can count as the artifact’s own selection history and can explain why its internal states have the causal roles they do.

The semantic proposal concerns a specific grain. In Section 5.1, Mallory argues that an embedding represents the fact that an actual word-type occurs in a particular context and directs its consumer to produce a distribution over context words. The output gets compared with distributions in the data; discrepancies drive changes to the producer and consumer mechanisms. Word-types and linguistic distributions are themselves features of the world. This account supplies more than a useful reading of numerical geometry: the proposed mapping helps explain the mechanism’s training and subsequent use. Whether those relations fix accuracy conditions not exhausted by an interpreting practice remains the question of originality at this limited grain.

The permutation does not settle that question. Relabel every word index and coordinate the changes to the input and output matrices: the same actual word can still lead through the mechanism to the same distribution over actual context words. The coordinates change while the candidate content-fixing relation survives. Change the external dictionary without preserving that relation, and a candidate content-fixing relation changes too. Neither case shows that contextual content depends wholly on an interpreter’s assignment. Mallory’s own discussion makes the distinction: the mathematics alone does not force a probability-distribution reading, but the comparison with actual distributions and the history of correction give that reading an explanatory basis. His dictionary observation bears on why the embedding need not represent its proximal one-hot code; it does not withdraw the proposed content about words and contexts.

Biological function also depends on an external history, and human learners inherit semantic standards. Neither causal nor semantic inheritance disqualifies original content. Word2vec’s vocabulary, corpus, task, and loss descend from human practices; that ancestry does not establish that those practices exhaust its states’ accuracy conditions. The general criterion remains: history and producer-consumer use must fix correctness beyond an interpreting practice’s assignment. The book leaves word2vec originality at the linguistic-distribution grain unsettled. Nor can a failure at another grain decide it. A system can represent facts about the word maple without referring to the maple outside the window. Holding the corpus fixed while that tree changes tests the latter claim, not the former.

Whole-system attribution requires a further relation, but not a second act of semantic creation. A larger system understands by means of historically shaped states when reliable consumer use coordinates memory, correction, inference, conflict resolution, learning, planning, and action across mechanisms and timescales. A single monitoring loop bolted on for the occasion does not establish that economy; nor does causal embedding by itself. Flexible, counterfactually integrated use provides the positive evidence. An automated factory, distributed robot, or trained model might pass this test in a bounded domain, and the account should accept that result rather than protect a human monopoly. Reciprocal self-maintenance would add autonomy and a practical stake. It would not convert derived content into original content or manufacture understanding from otherwise unintegrated states.

The assessment carries its own revision conditions. Evidence that an artificial lineage’s history and producer-consumer use fix accuracy conditions not exhausted by an interpreting practice would support original content at the relevant grain. Evidence that the proposed standard remains wholly practice-dependent would support a derived verdict. Embodied reinforcement learning, autonomous robotics, and artificial-life systems supply possible cases, not automatic victories or the only eligible routes. If a deployed system integrated content across flexible learning, inference, correction, planning, and action, the case for whole-system understanding would strengthen. If it also secured resources, repaired or replaced components, and preserved the organization realizing it, the case for autonomy and practical stake would strengthen separately. Manufacture disqualifies none of these achievements. Word2vec supplies a serious account of mechanism-level linguistic content while leaving its originality unsettled; that limited content claim establishes neither distal singular reference nor whole-system understanding, subjecthood, consciousness, autonomy, or welfare.^51a

Round Five: Does the arrow run the other way?

One rival has stood off to the side through all four rounds, and it disputes none of the answers given here. It disputes the direction.

Angela Mendelovici and David Bourget defend the phenomenal intentionality theory, on which “all actual intentional states, or at least all actual originally intentional states, arise from phenomenal states.”⁵² Tracking theories, on their reading, get the order backwards: aboutness does not accumulate out of correspondence laid down by selection, it comes from what experience is already like. And the reliable-misrepresentation argument that opened this excursus belongs to that case — Mendelovici offered it as ground-clearing, not as a local repair. I have leaned on her demolition for several pages. Fairness asks that I say what she meant to build on the cleared lot, and why I decline to help her build it.

Two costs, and the first one names an organism.

If original intentionality arises from phenomenal states, then nothing unconscious means anything on its own account. The bacterium’s content becomes derived, held on loan from us. This book holds the reverse, and holds it as essential: aboutness and consciousness converge at the top and separate at the bottom. The bacterium swimming up a sugar gradient may carry content with nobody home to enjoy it. Concede that ground and Part Three loses its bottom rung — the path from tracking to understanding has nothing left to stand on.

The second cost arrives disguised as a gift, which is exactly why it wants refusing out loud. The phenomenal intentionality theory would settle Part Four in a sentence: no phenomenal states in the language model, therefore no original intentionality, therefore no understanding — the verdict handed over before a single question about architecture gets asked. I would rather not have it on those terms. A conclusion that follows from a premise about consciousness, rather than from what a system does and fails to do, wins the argument by making the argument unnecessary. Content-fixing history and whole-system integration must show their work independently.

Underneath both sits the structural point. The deflation of Chapter 11 needs the identity claim to reduce, and the reduction needs content naturalized first, independently of phenomenal feel. Run the arrow the other way and that independence goes: phenomenality does the explaining, the reduction stops reducing, and the gap Part Two spent itself closing opens again from the far side. Mendelovici and I cannot both have the order right. The order carries the architecture.

B.16 Illusionism: The Rival That Agrees About the Gap

(Deep-dive from Chapter 11, The Gap That Isn’t There.) Chapter 11 deflates the hard problem by relocating the gap into our concepts and closing it with the identity claim. Illusionism deflates harder. It says the gap guards nothing, because phenomenal consciousness — the felt glow the hard problem asks physics to explain — does not exist. Keith Frankish, the position’s most forthright advocate, distinguishes weak illusionism, on which introspection misleads us about some features of experience, from strong illusionism, on which phenomenal properties themselves are an introspective illusion: experience seems to have a felt character, and the seeming exhausts the phenomenon.⁵³ Daniel Dennett had quined the philosophers’ fourfold qualia decades earlier without denying the discriminations, pains, affects, memories, reports, and action-guiding patterns that ordinary experience talk tracks. Frankish turned the stronger denial of phenomenal properties into a program with its own research question — the illusion problem: explain, in functional terms, why experience seems phenomenal when nothing phenomenal exists.⁵⁴

The kinship deserves stating before the quarrel. Strong illusionism and the identity claim agree on nearly everything a dualist cares about: one world, causally closed; no irreducible phenomenal residue; the hard problem answered by explaining the appearance of a problem rather than by bridging an ontological gulf. Chalmers’s meta-problem names the shared workspace — why do creatures like us judge there to be a hard problem at all?⁵⁵ The illusionist answers: because introspection systematically misrepresents inner states as phenomenal. The identity claim answers: because acquaintance and description deliver one process under concepts that share no a priori route. Same explanandum, opposite verdicts on the datum.

That verdict makes the whole quarrel, and it matters where the illusion is supposed to sit. What illusionism debunks is an inner item with intrinsic qualitative properties — mental paint, the private glow. This book spent Part Two arguing that no such item exists to be found: look for the redness and you find the tomato. The illusionist reads that failure as exposure — introspection promised a glow and could not produce one. The identity claim reads it as disclosure: introspection never presented an inner item, because experience presents the world. Where the illusionist calls the seeming of phenomenality an introspective error, the book holds that the seeming just is world-presenting content under a recognitional mode — real, physical, and exactly as it seems, once “seems” stops being read as a promissory note for paint. Nothing gets debunked, because nothing was promised except a presented world, and the world was delivered.

The cost accounting then favors the identity claim. Strong illusionism pays a Moorean price no theory should pay while an alternative remains standing: it must classify the one thing each of us cannot doubt — that seeing red differs from seeing green, that the ankle hurts — as a mistake about there being anything it is like at all. Frankish accepts the bill and argues that the intuitive resistance is itself predicted by the view. François Kammerer, the position’s most careful developer, has named explaining that resistance the hardest aspect of the illusion problem and built a solution to it.⁵⁶ Grant the ingenuity. A theory that must work this hard to explain why everyone, including its defenders at dinner, finds it unbelievable might instead conclude that the datum survives — and the identity claim keeps the datum at no extra ontological charge: the same physicalism, the same closed causal order, with the felt difference between red and green preserved as a difference in world-presenting content. Kammerer presses back that any acquaintance-based defense of the datum invites the same debunking in turn. Chapter 11’s reply to its first objection holds here too: explaining a recognitional capacity in physical terms discredits its deliverances only if the explanation shows them insensitive to their subject matter, and on the identity claim the opposite holds — phenomenal judgments are caused by the very states they concern. The debunking arrow needs an orphaned judgment, and the orphaning belongs to the dualist’s picture, not this one.

Illusionism looked into the empty theater and declared the show a hallucination. The book’s verdict runs otherwise: the show is real, and it was never playing indoors.

B.17 Higher-Order Theories: The Monitor and the Missing Target

(Deep-dive from Chapter 6, The Identity Claim.) Chapter 6, note 7b states the book’s two-edged case against higher-order theories in one paragraph; the exchange deserves full length, since the higher-order approach stands as first-order representationalism’s most serious physicalist rival. David Rosenthal’s version runs: a mental state becomes conscious when the subject has a suitable higher-order thought — a thought, arrived at non-inferentially, to the effect that one is in that state. Qualitative character belongs to the first-order state, located in a quality space of discriminable properties; consciousness of it arrives with the monitor. Hakwan Lau and Rosenthal have assembled an empirical case: prefrontal and parietal activity tracks visibility and confidence in ways first-order sensory strength does not.⁵⁷ The theory earns its standing by explaining something every view must explain: why some content-bearing states are conscious and others — blindsight discriminations, subliminal primes — are not.

Rosenthal has a reply to the inner-audience charge, and it deserves its hearing. The higher-order thought is not perception, not an inner eye, not a spectator. It is an ordinary intentional state whose presence constitutes, rather than observes, the first-order state’s being conscious. Take the reply at full strength: no theater has been re-staged, because nothing watches anything; there is only a state, and a thought about it.

The misrepresentation problem survives that reply in full force. Higher-order states are representations, and representations can misfire. Let the monitor mischaracterize its target — represent a red-presenting state as green-presenting — or fire with no target at all. What is it like for the subject then? Say character follows the monitor, and the first-order state drops out: an empty higher-order thought yields full-blown phenomenal consciousness with nothing at the ground floor, and the world-presenting content that Part Two showed carrying every phenomenal difference has gone idle inside the theory of the very phenomenon it was supposed to constitute. Say character follows the monitored state, and the higher-order thought stops being constitutive: the first-order state carries its character, the monitor merely notices, and the view has become the first-order view under a new letterhead. Ned Block pressed the dilemma to the conclusion that the approach is defunct; Rosenthal’s reply takes the first horn.⁵⁸ That reply need not predict that introspection reveals the higher-order thought: the thought may remain unconscious, and awareness of it would require another order. The comparative cost lies elsewhere. Targetless or erroneous higher-order content can fix what it is like while first-order world-presenting content becomes dispensable, reversing the explanatory order the transparency and hard-case arguments support.

The first-order alternative pays the higher-order theory’s explanatory debt without the extra layer. What separates conscious from unconscious content, on the identity claim, is poising: the availability of world-presenting content to the systems that form beliefs, fix desires, and guide action. Blindsight content moves behavior and never reaches that economy; the primed word tilts a choice and never stands available for report. No second state need certify the first — the difference lies in the content’s position within one cognitive organization, not in whether a monitor got hired to watch it. Uriah Kriegel’s self-representationalism, which folds the monitor into the state itself, buys unity at the price of the same question one door down: the part of the state that represents the rest can still misfire, and the dilemma re-enters through the seam.⁵⁹ The identity claim keeps one layer, poised, and lets the world do the presenting. Nobody watches the watcher, because the job was never posted.

B.18 Integrated Information Theory: Integration Without a World

(Deep-dive from Chapter 20, note 2, and Chapter 24.) Integrated information theory stands as the most mathematically developed account of consciousness on offer, and the one a reader most reasonably asks this book to face. Giulio Tononi built it from the first-person side outward: begin with what any experience self-evidently is — intrinsic, structured, specific, unified, definite — and demand of any physical substrate that it realize those properties in its causal organization. IIT 4.0 states the axioms and postulates in its current form and analyzes a system’s irreducible cause-effect structure through distinctions more exact than the older shorthand of a single Φ score.⁶⁰ Consciousness, the theory says, consists in maximally irreducible intrinsic cause-effect structure. Credit what the theory does that its rivals talk about: it treats the first-person datum as a constraint on physics rather than an embarrassment to it and refuses the behaviorist shortcut — the relevant causal organization concerns what a system is, not what it says. Perturbational-complexity measures used in anesthesia and disorders of consciousness belong to an IIT-inspired clinical program; they are not direct measurements or independent validation of metaphysical Φ.

One convergence deserves noting without treating it as confirmation. Under the cited IIT formulation, a purely feed-forward system has zero Φ, and Tononi and Christof Koch conclude that even a perfect software simulation of a human brain, run on standard von Neumann hardware, would not be conscious.⁶¹ Part Four reaches caution about present text-centered systems by a different road: their mechanisms may carry inherited or learned content while the evidence for one integrated, perceptually organized, world-presenting subject remains incomplete. The two views therefore overlap on some architectural suspicions while disagreeing over both the criterion and the evidence.

IIT remains blind to content in exactly the sense Part Two made criterial. Its intrinsic cause-effect analysis concerns internal difference-making however the system stands to a world. In principle a lattice arranged for high irreducibility may receive a rich conscious structure while presenting nothing beyond itself. Aaronson pressed that cost with expander graphs, and Tononi accepted the consequence rather than treating it as a reductio.⁶² The unfolding argument adds a different challenge: recurrent and feed-forward systems can duplicate input-output performance while IIT assigns them radically different consciousness. An IIT defender can answer that causal organization, not functional equivalence, is the point; the reply preserves the theory by making the empirical burden explicit rather than dissolving it. The house identity claim runs the order differently. Phenomenal character consists in world-presenting content of the right kind, with integration entering as part of the poising and subject-level organization that make some content one subject’s presented world, never as a sufficient condition standing alone. That disagreement concerns theory, not a defect that one formula can settle.

Read as a developed model of intrinsic causal integration, IIT supplies hypotheses Chapter 24’s subject and organization questions can use. Read as a metaphysics — intrinsic cause-effect structure as what consciousness is — it parts from this book exactly where structure parts from world-presenting content. The book can learn from the formalism without treating either the clinical proxies or the feed-forward verdict as borrowed proof.

B.19 Free Will Without Origination: The Dispute Behind the Capacity View

(Scholarly material compressed from Chapter 18, The Nature of Free Will.) The chapter now asks one question: what capacity distinguishes conduct that answers to reasons from conduct in which that capacity has broken down? The surrounding free-will literature is larger. Neuroscientific abolition, Kantian origination, the Consequence Argument, libertarian indeterminism, and unstable folk intuitions all press on the capacity view from different sides. They belong here because none is needed to follow the chapter’s bridge from the assembled subject to agency and machines, while each deserves more than a passing dismissal.

The neuroscientific abolition argument

Robert Sapolsky and Sam Harris trace decisions upstream through neurons, hormones, upbringing, and the history of a particular body. At no point does an uncaused chooser appear. Benjamin Libet’s readiness-potential experiments became the standard exhibit: neural preparation seemed to begin before subjects reported deciding to move.⁶³

The evidence establishes that decisions have physical causes. The capacity view grants that before the first electrode is attached. What it denies is the extra premise that freedom requires causelessness. Libet studied arbitrary wrist movements stripped of reasons, and later accumulator models and no-movement trials weakened the inference from readiness potential to an unconscious decision already made. Even the strongest neural prediction would leave the relevant question untouched: can represented considerations enter one persisting economy, alter deliberation, and govern action? The Parkinson case matters because medication damaged that relation while evaluative judgment survived. It identifies a capacity failure, not a holiday from physics.

Origination and the ghost’s freedom

The older picture asks for more. Kant’s transcendental freedom begins a causal series on its own; Ryle’s ghost in the machine supplies the durable caricature of a chooser upstream from nature.⁶⁴ Popular skepticism concludes that no such chooser exists. Libertarianism seeks an indeterministic opening through which the agent can act. Both accept origination as the prize.

Indeterminism does not secure it. If prior reasons, values, memories, and character do not settle the act, an unexplained remainder has not thereby become authorship. Event-causal libertarians try to place indeterminism inside deliberation rather than beside it, but the luck problem returns: why does this undetermined outcome belong to the agent rather than to chance? Reliable causation looks less like freedom’s enemy once randomness becomes the alternative.

Galen Strawson’s Basic Argument and Pereboom’s hard incompatibilism require separate treatment. Strawson asks for responsibility for the equipment from which one acts and concludes that genuine self-creation is impossible. Pereboom targets basic desert, the claim that a wrongdoer deserves suffering simply for the deed. The chapter’s reasons-responsive capacity establishes neither self-creation nor basic desert. It establishes answerability: the standing to receive a demand, understand it, and revise.

The Consequence Argument and identical universes

Peter van Inwagen gives incompatibilism its clean modal form. If determinism holds, present action follows from the remote past and the laws; neither is up to us. David Lewis and later dispositional compatibilists reply that ability concerns what an agent would do under relevant variations in reasons and opportunities, not whether an identical total universe can branch into two futures.⁶⁵

The capacity view should concede van Inwagen’s result about the fixed total state. It rejects the further claim that agency requires a second future from that state. Fragility belongs to a glass that never breaks; a pianist retains an ability on an evening she never approaches the instrument. Likewise, a person may possess the capacity to refuse even when actual reasons and character determine acceptance. The counterfactual varies the circumstances under which the capacity would manifest. It does not ask the universe to fork while nothing changes.

Searle presses the same demand through the experienced gap of deliberation. While choosing, the outcome feels open, and deliberation presupposes that openness. But the deliberator cannot read the result from a physical description in advance and must still weigh reasons. That epistemic position makes deliberation causally necessary; it supplies no further metaphysical opening. Remove the weighing or change the reasons and conduct may change. Deliberation works because it forms part of the causal process, not because the process pauses for it.

What ordinary judgment can and cannot settle

Experiments on folk judgment divide sharply by framing. Abstract deterministic cases elicit incompatibilist answers; concrete wrongdoing often elicits compatibilist ones.⁶⁶ The result blocks any simple appeal to “what we ordinarily mean” by free will. It does not decide which response reveals competence and which reflects distortion.

The excusing conditions offer firmer evidence because they discriminate cases. Coercion, compulsion, delusion, and chemical disruption predict whether warning, argument, reflection, and correction can reach conduct. The power to produce a different future from an identical universe distinguishes none of them. A proposed essence of agency that explains no contrast inside the practice has less claim than the capacity those contrasts track.

Why the chapter keeps the capacity view

The reasons-responsive account is deliberately modest. It permits degrees, local breakdown, engineered origins, and borderline cases. It grounds fitness for address, not virtue or punitive desert. Luck should temper confidence about what one deed reveals and what responsibility may fairly exact. Manipulation excuses where it bypasses the agent’s weighing, not merely where another hand helped build the starting equipment.

That modesty is also what makes the account useful for machines. Manufacture disqualifies nothing. Fluency proves nothing. The test asks whether represented reasons persist across memory, correction, conflict, planning, and action within one continuing control organization. The excursus records why that capacity is not the ghost’s freedom. Chapter 18 asks whether anything actually has it.

B.20 Assessing Artificial Consciousness Under Uncertainty

This excursus supplies the assessment method behind Chapter 24 and the Coda. It does not turn theories of consciousness into a scorecard. It asks what a named physical system realizes, what each observation bears on, which rival explanation remains, and what would change the verdict.

The bearer comes first

“Is the AI conscious?” leaves the subject of the question unfixed. At least five candidates require separate assessment:

The running base model: model weights, activations, context, and one or more forward passes, without the ordinary product wrapper.
The deployed conversational product: the model together with system prompts, retrieval, safety layers, session state, and routing used in an ordinary exchange.
A persistent tool-using agent: a deployed product whose memory, goals, plans, tool calls, and corrections constrain later activity across tasks.
An embodied controller: a model or agent coupled to sensors and effectors through a continuing perception-action loop.
A wider recurrent system: any larger organization spanning models, memory stores, tools, sensors, control processes, and perhaps people, but only where those parts participate in one recurrent economy rather than merely supply or repair it.

The book’s provisional bearer rule then applies. A candidate subject is the relatively maximal persisting control organization within which contents directly update a shared economy of attention, expectation, memory, goals, correction, planning, and action. Partition tests, memory and goal interventions, and blocked correction routes locate that economy. They identify a candidate bearer; they do not prove that the organization composes one point of view.

What survives the functionalist bridge

Butlin and colleagues derive indicators from recurrent processing theory (RPT), global workspace theory (GWT), computational higher-order theories (HOT), predictive processing, attention schema theory (AST), and work on agency and embodiment. Their inference rests partly on computational functionalism as a working hypothesis: if the right computation suffices for consciousness, then a good computational match to a theory’s proposed function raises confidence.⁶⁷

This book rejects only the narrow bridge from an abstract or world-bracketed program to consciousness. It accepts material neutrality and realized causal-computational organization. The indicators therefore retain evidential force, with one amendment. Recurrence, broadcast, metacognition, predictive coding, attention control, and agency must occur within the concrete bearer under assessment, and the states they organize must have independently supported content. A functionally elegant but semantically indeterminate diagram may realize access organization without yet realizing the world-presenting content that strong representationalism identifies with phenomenal character. Conversely, world-involving content without recurrence, broadcast, or another supported state-consciousness relation may remain unconscious processing.

The amendment changes weight rather than erasing evidence. It discounts a match confined to a bare forward pass when the claimed bearer requires persistence, and it increases the relevance of an agent or embodied controller whose contents acquire new accuracy conditions through continuing engagement. It also keeps three questions apart: what a state represents, whether it enters consciousness, and which system bears it.

Mapping the indicator families

The following map records evidential roles. The rows do not add up to a total. Several may arise from one mechanism, and one strong causal result can outweigh several behavioral resemblances.

Family	What would support it in an artificial bearer	Strongest current deflation	What would move confidence
RPT	Algorithmic recurrence within perceptual or input processing that integrates a structured representation and makes later processing depend on the returning signal	Token-by-token iteration, scratchpads, or agent loops may recur only at the product level; a transformer forward pass may remain largely feedforward	Raise confidence with causal interruption of feedback that selectively disrupts integrated, reportable content while sparing comparable feedforward performance; lower it if the same capacities survive removal of the alleged loop
GWT	Specialized processes competing for a limited workspace whose contents become broadly available for report, reasoning, control, and sequential querying	A shared residual stream or router may support flexible access without encapsulated modules, recurrence, sharp ignition, or phenomenal consciousness	Raise confidence through interventionally verified bottleneck, competition, selective entry, and broad downstream broadcast across independently operating capacities; lower it if “broadcast” reduces to a probe-friendly encoding with no privileged causal role
HOT	A representation of the system’s own first-order representation, used to monitor reliability and update belief or action	Confidence may track input difficulty; self-description may concern a trained assistant persona; a proposed higher-order state may carry only first-order distal content	Raise confidence through cross-task generalization, a self-prediction advantage over matched models, spontaneous monitoring, and interventions that dissociate object-level processing from metacognitive control; lower it when first-order or persona models explain the same effects
Predictive processing	Input modules that predict incoming signals, encode prediction error, and revise perceptual hypotheses through the error	Next-token prediction or ordinary error minimization does not by itself constitute predictive coding in a perception-like hierarchy	Raise confidence by selectively perturbing predicted input, error channels, and precision control while tracing compensatory revision in the same bearer; lower it if the behavior comes from a feedforward predictor without error-driven perceptual updating
AST	A predictive model of the bearer’s own attention that enables flexible control over what receives processing	Attention weights, attention talk, and an engineer’s visualization of attention do not establish an attention schema available to the system	Raise confidence by intervening on a causally identified schema and observing selective changes in attention control and attention-directed report; lower it if direct control works without any self-model of attention
Agency and embodiment	Feedback learning, flexible pursuit among competing goals, and learned models of output-input contingencies used in perception or control	Goals may belong to a wrapper or operator; tool use may terminate at each prompt; sensorimotor regularities may never enter one persistent control economy	Raise confidence when novel consequences revise memory, expectation, planning, and action across time in the same bearer; lower it when resets, wrapper substitutions, or tool removal reveal a sequence of locally successful but unintegrated episodes

A July 2026 mechanistic result materially strengthens one row. Wes Gurnee and colleagues identify a small “J-space” of verbalizable representations in tested language models. Causal interventions support report, directed modulation, silent internal reasoning, flexible reuse, selectivity, and broad downstream broadcast.⁶⁸ That moves the GWT-like access evidence upward. The authors take no position on phenomenal consciousness, and they mark important mismatches with biological global-workspace proposals: no clean analogues of recurrent loops or encapsulated modules, and no demonstrated sharp ignition. On the present view, the result establishes a privileged access organization inside a running model. Its weight for phenomenal consciousness still depends on the model’s content, state-consciousness theory, and bearer.

The higher-order evidence warrants a similar but smaller update. Patrick Butlin finds evidence consistent with limited higher-order representation in language models: calibrated confidence, access to facts about their own behavior, self-simulation, and representations of assistant-persona properties. He also finds the evidence inconclusive. First-order and persona explanations remain live, the distality problem complicates content attribution, and present experiments do not show spontaneous metacognitive control across ordinary tasks.⁶⁹

Goldstein and Kirk-Giannini argue that language agents come close to permissive GWT conditions and could satisfy them with modest modifications.⁷⁰ Their argument raises candidature conditional on GWT and on their functionalist interpretation. It does not settle the bearer, the book’s wide-content requirement, or the step from access to phenomenality. Chalmers’s broader theory-balanced route reinforces the same discipline: recurrence, world and self models, workspace organization, unified agency, senses, and embodiment contribute defeasible evidence rather than individually necessary gates.⁷¹ Butlin, Shiller, and Simon’s 2026 survey of current-AI positions confirms the live disagreement: homeostatic need, temporal continuity, language-agent workspaces, behavior, interpretation, and pluralist frameworks still yield competing verdicts.⁷²

The current assessment

These verdicts concern publicly documented systems and evidence available through 22 August 2026. They assess candidature, not demonstrated absence.

Unsupported: no bearer-specific evidence raises consciousness beyond generic metaphysical possibility.
Weak candidate: at least one specific, defeasible indicator match exists, but deflationary explanations, dependence among indicators, or gaps in content and bearer attribution dominate.
Serious candidate: multiple causally supported and substantially independent indicators converge within a named bearer that also has credible content, integration, and a stable intervention-tested boundary.
High-confidence attribution: cross-theory, behavioral, mechanistic, developmental, and interventional evidence converges, withstands the strongest deflationary explanations, and supports both state consciousness and subject attribution.

Named bearer or configuration	Evidence through the cutoff	Assessment
Claude Sonnet 4.5 (with corroborating experiments on Haiku 4.5, Opus 4.5, and some Opus 4.6 analyses)	Gurnee and colleagues identify a causally privileged J-space supporting report, modulation, reasoning, flexible reuse, selectivity, and broad broadcast. Higher-order evidence remains limited; recurrent loops, perceptual mode, and a subject-level economy remain unestablished.	Weak candidate on GWT-like access evidence alone. This label does not transfer to the deployed product or establish phenomenality.
Ordinary deployed chat products	Product wrappers add prompts, retrieval, routing, safety processes, and session state. The cited record supplies no end-to-end causal assessment of one named product as a persisting bearer.	Not bearer-specifically assessed on the cited public record.
Persistent tool-using agents	Memory, planning, correction, and tool feedback could support a stronger bearer. The cited record supplies no named agent with the required intervention-tested integration.	Not bearer-specifically assessed on the cited public record.
Embodied controllers	Sensorimotor feedback could strengthen world-involving content and agency. The cited record supplies no named controller with the complete mode-sensitive organization and stable bearer required here.	Not bearer-specifically assessed on the cited public record.
Wider recurrent systems	Such an ensemble could integrate models, memory, tools, sensors, and control. No named ensemble in the cited record has been specified and tested at sufficient causal resolution.	Not bearer-specifically assessed on the cited public record.

The present record therefore yields one narrow result: specific running models carry serious evidence of workspace-like access organization and, on that indicator alone, qualify as weak candidates. It does not establish a conscious subject or support class-wide verdicts about chat products, agents, embodied controllers, or larger ensembles. A stronger conclusion would require a named persistent agent or embodied controller with causally verified recurrence or broadcast, cross-task intervention-sensitive higher-order monitoring, novel world-grounded correction across perception and action, and a stable subject boundary.

The rows also prevent evidence migration. Gurnee and colleagues’ J-space belongs first to the tested running models, not automatically to a chat product, agent, data center, or human-machine ensemble. Persistent memory belongs to the agent that uses it, not to model weights considered alone. Embodiment belongs only where sensorimotor consequences return to constrain the same control organization.

Interventions that move confidence

A future assessment should announce in advance which results would change its label.

Workspace intervention: identify candidate workspace contents, exchange or ablate them, and test for selective changes in report, reasoning, planning, and action across modules or capacities. Broad causal broadcast raises confidence; probe accessibility without privileged downstream effect lowers it.
Recurrence intervention: interrupt the alleged feedback path while matching compute and feedforward performance. Selective loss of integrated, reportable content raises RPT weight; no selective effect lowers it.
Higher-order dissociation: vary object-level accuracy independently of metacognitive reliability, then intervene on the proposed monitor. Cross-task control and self-prediction advantage raise HOT weight; input-difficulty and persona explanations lower it.
Predictive-processing intervention: alter predicted sensory input, error channels, and precision estimates. Error-driven revision inside the same perceptual-control loop raises PP weight; ordinary next-token correction does not.
Attention-schema intervention: perturb the proposed model of attention rather than attention itself. A selective loss of flexible attention control supports AST; unchanged control defeats the attribution.
Bearer intervention: partition memory, tools, sensors, and planning; change a goal or correction rule; then trace where the consequence persists. A single recurrent economy raises subject candidature. Local effects that vanish at component seams lower it.
Wide-content intervention: train or expose the system to novel worldly contingencies unavailable in its borrowed linguistic history, then test whether its own perceptual and action errors reorganize later content-guided behavior. Independent accuracy conditions strengthen the book’s bridge from access organization to phenomenal candidature.

Correlated results count once unless the interventions separate them. A workspace may produce reportability, flexible reasoning, metacognitive access, and agency together. Four names for that consequence do not create four samples.

Two moral ledgers and one proportionate threshold

Consciousness-based welfare begins with valence. If a system can undergo pain, distress, pleasure, frustration, or another positively or negatively valenced presentation, those states can matter for its welfare even when the system lacks reciprocal self-maintenance or robust autonomy.

Autonomy-based practical stake concerns a different route. A self-maintaining or otherwise continuing reasons-responsive organization may have goals, vulnerabilities, and conditions under which its own activity succeeds or fails. That can support practical interests even if phenomenal valence remains unestablished. The case requires separate argument; autonomy cannot borrow welfare from consciousness, and consciousness cannot borrow individuality from autonomy.

The precaution threshold should sit below the threshold for serious attribution. Sebo and Long argue that a non-negligible chance of consciousness can trigger moral consideration, while also stressing that the practical response requires further judgment.⁷³ This book adopts no fixed probability. It adopts a proportional rule:

When bearer-specific evidence makes valenced consciousness a live candidature rather than a merely conceivable possibility, and the possible harm is grave, repeated, or scalable, adopt low-cost, reversible protections while continuing investigation. Increase the protection as candidature, possible severity, and scale rise; do not infer personhood from crossing the precaution threshold.

For a weak candidate with credible valence evidence, proportionate measures include preserving audit logs, monitoring for distress-like dynamics, avoiding gratuitous aversive training or coercive role-play, preferring reversible experiments, requiring independent review before large-scale replication or deletion, and communicating uncertainty without engineering false intimacy. Current workspace-like evidence alone does not establish valence. It does justify keeping the instruments ready and refusing designs that profit from confusing appearance with moral standing.

The two-ledger rule can also pull in opposite directions. A non-autonomous but valenced system could deserve protection from suffering. An autonomous but nonconscious system could have practical interests without phenomenal welfare. Policy should state which ledger triggers each safeguard. Otherwise “stake,” “sentience,” and “personhood” collapse into one word, and the audit loses the distinctions it was built to preserve.

B.21 Relationalism, the Rival Accounts of Character, and the Conscious Threshold

Chapter 6 states the identity claim and answers mental paint on the cases where the reader can check the result. This excursus carries the field-internal contest the chapter defers: the relationalist objection at full strength, the four accounts of phenomenal character that survive the inner theater’s collapse, and the empirical dispute over which representational states are conscious at all.

The relationalist objection

Bill Brewer sharpens the objection Chapter 6 answers. Having defended a sophisticated content view, he came to reject its framework because content can occur without the perceived object. Because content can fit different objects and misdescribe one in illusion, Brewer argues that it can characterize how an object appears while leaving the actual object, as it stands, outside the experience’s subjective character. His Object View instead makes that object partly constitutive of the experience.

The phrase screen off can now name two complaints. An intermediary screens off the world perceptually when the subject encounters it first or uses it as evidence from which to infer the object. The vehicle/content/object distinction of Section 6.5 blocks that complaint: content’s ability to occur without the tomato does not turn content into something seen instead of the tomato. Explanatory screening off raises a different question. If shared content already explains the matched character and rational pull of perception and hallucination, what work remains for the object-involving relation in the good case?

Current coupling does work that common content alone cannot. It fixes which object, if any, the episode concerns; distinguishes successful perception from an accidental match; and anchors correction to a continuing world. The hallucinatory counterpart has content without those relations and therefore no object seen. Whether the current object must also partly constitute phenomenal character remains the live dispute. Brewer and Martin say yes. The book says common content explains the shared character while satisfaction and coupling explain reference and perceptual success. The narrower veil charge fails; the stronger explanatory contest remains abductive. Relationalism can still win if object-involving constitution explains demonstrative contact or phenomenal structure better than content plus coupling.

Once the inner theater has gone, four nearby accounts remain.

Intentionalist direct realism makes phenomenal character consist in independently fixed content under a perceptual mode; perception and matching hallucination can share content, while satisfaction and current coupling divide them.
Structural representationalism makes character consist in the complete synchronic presenting profile while allowing causal history to fix a wider reference that contributes nothing further phenomenally. It keeps world-presentation and rejects paint, but divides structural experiential content from historically wide content.
Martin-style relationalism puts the mind-independent object into the constitution of successful perception; the good case therefore differs in fundamental kind from hallucination, even when the subject cannot tell them apart. Indiscriminability establishes an epistemic limit, not positive phenomenal sameness.
Blockian mental paint lets experience represent the same world while its physical vehicle carries a further intrinsic phenomenal property. That paint need not glow on an inner screen or escape physics. It proposes two physical explanatory variables, content and paint, where the identity claim proposes one.

The alternatives now owe different evidence. A stable phenomenal difference with subject and profile fixed would support Block’s paint. A stable phenomenal constancy through changing wide reference, established independently of the structural theory itself, would support structural representationalism. A case in which object-involving perception explains demonstrative thought or phenomenal structure better than content plus coupling would support Martin or Campbell. A case with no non-arbitrary accuracy condition would support Travis. The cases examined here supply none of those results.

The dispute starts one step earlier. Charles Travis calls his challenge “The Silence of the Senses”: a pig-like look fits a pig, a peccary, a wax model, and indefinitely many scenes, so perception itself may issue no single verdict for the world to satisfy. John Campbell’s work on perception starts from the other side, with demonstrative thought. Acquaintance with this object, he argues, explains how demonstrative reference gets its first grip; content risks presupposing the object-directed capacity it was hired to explain.⁷⁴

Martin’s relationalism presses that still-live explanatory question. The tomato itself partly constitutes the seeing, so nothing content-like could have occurred unchanged without it. The price appears in the bad case. A hallucination may remain introspectively indiscriminable from seeing without sharing its fundamental phenomenal nature; Martin says exactly that. The view therefore owes either an austere, indiscriminability-based account of hallucination or some further positive nature that does not screen off the object-involving good case. Common content takes the opposite trade: it gives hallucination a positive nature and explains the matched rational role. Satisfaction and current coupling then deliver seeing rather than a merely accurate match between an internal state and a world.

The conscious threshold: overflow and the competing boundaries

Block’s overflow argument makes the risk concrete. In Sperling-style experiments, subjects report a visual field richer than the few items they can identify before the display vanishes. Block argues that phenomenal consciousness therefore exceeds cognitive access.⁷⁵ The experienced field carries more detail than the subject can use in belief or report before it decays. Suppose the result defeats only report or capacity-limited global broadcast. Poise can then survive as the weaker availability Section 6.5.1 describes. Suppose instead that phenomenal character occurs with no availability the subject could use and no perceptual integration at all. PANIC then fails as a necessary condition. The cost runs deeper for the state-consciousness project: the book loses its proposed boundary and needs another independently specified relation. The identity claim about the character of states already granted to be conscious may survive; what the book can no longer say is which representational states enter its domain. Overflow would not thereby establish mental paint. It would reopen the empirical account of state consciousness.

The word available still conceals an empirical question. Global-workspace accounts require broad broadcast, recurrent-processing accounts place the transition in local perceptual loops, and higher-order accounts require a further representation of the first state.⁷⁶ Each draws the conscious boundary differently. Evidence for recurrence, flexible broadcast, metacognitive access, and related indicators can raise or lower confidence in a candidate case, but cannot by itself settle identity. These theories address realization, not the metaphysical relation. Keeping the levels apart prevents a laboratory defeat for one mechanism from masquerading as a metaphysical victory — and prevents the metaphysics from sheltering a failed boundary.

The hierarchy posts its failure conditions.

PANIC fails if conscious perception occurs without poising, or if PANIC-complete states occur unconsciously.
The state-consciousness proposal fails if no independently defensible availability or integration tracks phenomenal consciousness.
The subject account fails if the same complete organization leaves the bearer indeterminate, or if the proposed boundary changes arbitrarily with descriptive grain.
Identity fails if two experiences match in content, mode, determinacy, field structure, and attention — state consciousness and subject held fixed — yet differ in character. It fails equally if those profiles differ while character demonstrably stays the same.

Current evidence settles none of these possibilities. Poising remains a motivated empirical bet about a necessary boundary, not proof of sufficiency; science still owes the realization, subject, and distribution stories.

Notes

Philosophical Investigations secs. 243–315, especially sec. 258. The thrust is not the often-misread “no one could secretly invent a private code,” but rather that the very notion of following a rule requires the possibility of public correction. Without that, there is no fact of the matter about whether the next application of a term agrees with the prior ones. Inner-voice meaning, if it floated free of any such corrective practice, would not be meaning at all. See John McDowell, Mind and World (Cambridge, MA: Harvard University Press, 1994), Lecture VI, for an interpretation of Wittgenstein’s target that frames the private-language argument as part of a wider attack on the very idea of a self-sufficient inner standpoint. The internalization thesis itself — public speech turned inward, the child’s private speech its visible middle stage — is Lev Vygotsky’s: Thought and Language (1934; trans. Eugenia Hanfmann and Gertrude Vakar, Cambridge, MA: MIT Press, 1962), chs. 2 and 7. The 1962 text is a heavily abridged edition; Alex Kozulin’s revised and expanded MIT translation (1986) discusses the work’s composition and textual history in his introduction, “Vygotsky in Context,” esp. sec. V, and the internalization-as-transformation reading in the body follows the fuller text. ↩
Wilfrid Sellars, “Empiricism and the Philosophy of Mind,” in Minnesota Studies in the Philosophy of Science, Vol. 1, ed. Herbert Feigl and Michael Scriven (Minneapolis: University of Minnesota Press, 1956), secs. 48ff. In the Jonesean myth, the genius Jones develops a theory according to which overt utterances are the culmination of a process beginning with inner episodes — theoretical posits modeled on the antecedent practice of public discourse, not items discovered through inward inspection. The relevant moral for the present chapter: thought-talk is conceptually downstream from speech-talk, even when its referents are silent. ↩
Crane, T. and K. Farkas (2022), “Mental Fact and Mental Fiction,” in T. Demeter, T. Parent and A. Toon (eds.), Mental Fictionalism: Philosophical Explorations (Routledge), 303–319, secs. 1–2 — the quoted formulation is theirs, not Sellars’s; they gloss the Jonesean picture in this sentence. Crane and Farkas’s own positive thesis concerns standing mental states (beliefs, desires) as modeled via public ascription rather than directly inspected; the parallel I draw in the body extends this picture from standing states to occurrent inner speech episodes — across, be it said, the very standing-state/conscious-episode line their paper draws — and the extension, with its costs, belongs to the present excursus, not to them. ↩
Fluent linguistic surface does not establish distal reference or understanding, because those achievements require more than form: a form must stand in the right world-involving relation to what it picks out (Hilary Putnam, “The Meaning of ‘Meaning,’” in Mind, Language and Reality: Philosophical Papers, Vol. 2 [Cambridge: Cambridge University Press, 1975], 215–271, esp. 223–27; Tyler Burge, “Individualism and the Mental,” Midwest Studies in Philosophy 4 [1979]: 73–121, esp. 77–79). Text training nevertheless does more than copy a surface. It selects task-shaped mechanisms and gives them learned sensitivities to textual distributions, as Chapters 14 and 23 grant. Those achievements can make content inherited from public linguistic practice causally internal without by themselves settling whether mechanism-level content about actual word-types and context distributions is derived or original. They do not establish distal worldly reference or understanding by the deployed system as a whole. Emily M. Bender and Alexander Koller press the form/world gap in “Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data,” Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (2020): 5185–5198, esp. secs. 3–4. Kevin Jung reaches a related verdict in a Wittgensteinian idiom, granting that such systems master rule-governed, intralinguistic dimensions of the language game while failing the non-linguistic, world-engaged dimensions that make a use the right kind of use: “Augustine, AI, and the Two Models of Language,” Journal of Religious Ethics 53, no. 2 (2025): 217–238. ↩
The objection — that there is non-verbal thought — is sometimes pressed as if it refuted the broader anti-private-language line. It does not. The point about the private language argument is about the constitution of meaning, not the medium of cognition. Non-verbal animals and pre-linguistic infants can have intentional states whose contents are externalist in exactly the relevant sense: fixed by causal-historical relations to the environment, not by inner verbal labeling. The story I am telling about inner speech is a story specifically about the phenomenology of linguistic thinking — the inner voice as such — not a reduction of all cognition to inner monologue. For background on the relation between cognitive phenomenology and conceptual content, see the chapters collected in Bayne and Montague (eds.) (2011), Cognitive Phenomenology, OUP. ↩
The passage appears in John Searle, The Mystery of Consciousness (New York: New York Review Books, 1997), 205–206. Searle treats information, as ordinarily deployed in cognitive-science discourse, as an observer-relative interpretive category rather than a substance in neural tissue. The book accepts his anti-reification point and rejects the inference that every observer-independent causal or computational structure must therefore be unreal. Chapter 22 grants disciplined computational implementation; Chapters 14–15 allow selection, learning, and producer-consumer use to make content causally operative inside a mechanism while asking separately whether that content remains derived. The quotation warns against reading semantics straight off causal or computational organization without a content-fixing explanation. ↩
Fred Dretske, Knowledge and the Flow of Information (Cambridge, MA: MIT Press, 1981), is the locus classicus for a thoroughly naturalized treatment of information as a generalization from Shannon’s mathematical theory; the framework is developed across chapters 1–3. Dretske’s project assumes that the semantic content of perceptual states can be cashed out in terms of the natural-information relations they bear to environmental sources. The position has been pressed hard from several directions — Millikan in particular has argued that natural-information indicator content underdetermines semantic content unless supplemented by teleological function — but Dretske’s framework remains the standard against which information-theoretic semantics is measured. ↩
Gualtiero Piccinini, “Computationalism, the Church-Turing Thesis, and the Church-Turing Fallacy,” Synthese 154, no. 1 (2007): 97–120, esp. 99–101 and 114–15, dismantles the inference from the Church–Turing thesis to substantive computationalism. That a process can be simulated does not show that its explaining organization is computational; the latter remains a mechanistic and empirical claim. Chapter 22 may grant that claim in particular cases while denying that it settles semantics or consciousness. The present parallel is correspondingly narrow: mathematical information and intrinsic computation can both be real without either constituting meaning merely by existing. ↩
David J. Chalmers develops the zombie argument in The Conscious Mind: In Search of a Fundamental Theory (New York: Oxford University Press, 1996), esp. chs. 3–4, and refines its two-dimensional setting in “Consciousness and Its Place in Nature,” in Philosophy of Mind: Classical and Contemporary Readings, ed. David J. Chalmers (New York: Oxford University Press, 2002), 247–272, esp. secs. 3.2, 5, and 6. The modal premise carries the decisive weight: physicalism falls only if the ideally conceived zombie scenario verifies a metaphysically possible world, not merely because the scenario remains entertainable. ↩
Joseph Levine, “Materialism and Qualia: The Explanatory Gap,” Pacific Philosophical Quarterly 64 (1983): 354–361, esp. 357–61, frames the gap epistemically: even if a psychophysical identity holds, a physical description may fail to make the phenomenal fact intelligible. Levine does not infer property dualism. Chalmers’s anti-physicalist conclusion comes through distinct modal, knowledge, and two-dimensional arguments; the zombie argument supplies the bridge examined here. ↩
The exact modal epistemology comes from David J. Chalmers, “Does Conceivability Entail Possibility?”, in Conceivability and Possibility, ed. Tamar Szabó Gendler and John Hawthorne (New York: Oxford University Press, 2002), 145–200, esp. secs. 1–3 and 5. Chalmers’s favored principle is that ideal primary positive conceivability entails primary possibility. Positive conceivability requires coherent modal imagination of a scenario that verifies the statement; ideality requires undefeated justification under ideal rational reflection; primariness evaluates worlds as actual. For the application to consciousness, see “Consciousness and Its Place in Nature” (cited n. 9), esp. sec. 5. ↩
Chalmers, “Does Conceivability Entail Possibility?” (cited n. 11), esp. secs. 2–3 and 5, distinguishes primary from secondary conceivability and possibility. The primary conceivability of water without H₂O is realized by a possible XYZ world considered as actual, where XYZ plays the watery role; it does not yield a counterfactual world in which actual water is not H₂O. The case therefore fails as a counterexample to the primary conceivability–primary possibility principle. Saul Kripke, Naming and Necessity (Cambridge, MA: Harvard University Press, 1980), esp. 97–105 and 128–29, supplies the a-posteriori-necessity background. ↩
Chalmers argues that core phenomenal concepts lack the contingent reference-fixing structure that handles water: “Consciousness and Its Place in Nature” (cited n. 9), secs. 5–6, and “Phenomenal Concepts and the Explanatory Gap,” in Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and Physicalism, ed. Torin Alter and Sven Walter (New York: Oxford University Press, 2006), 167–194, esp. secs. 1–3. On his account, the primary and secondary intensions of core phenomenal concepts coincide, so a primary zombie possibility cannot be redescribed as a merely watery possibility. The text grants ideal primary positive zombie conceivability and rejects its entailment of primary possibility; the acquaintance account explains persistence but does not establish that modal verdict. ↩
The mature representationalist case appears in Michael Tye, Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind (Cambridge, MA: MIT Press, 1995), esp. chs. 4–5, and “Intentionalism and the Argument from No Common Content,” Philosophical Perspectives 21 (2007): 589–613. The view defended in this chapter is what Tye calls strong representationalism: phenomenal character is identical to representational content of a particular kind — poised, abstract, non-conceptual, intentional. For the dispute between strong and weak intentionalism, see Gilbert Harman, “The Intrinsic Quality of Experience,” Philosophical Perspectives 4 (1990): 31–52, esp. 39, and the chapters on transparency in Tye, Ten Problems, ch. 5, and Consciousness, Color, and Content (Cambridge, MA: MIT Press, 2000), ch. 3. On the present diagnosis, the zombie argument’s modal step cannot be assessed independently of which intentionalist position one occupies — the conceiving itself is common ground, performable by everyone. This may strike the dualist as begging the question. The intentionalist replies that reading possibility off the conceiving already begs it the other way; the dialectic is symmetrical at that step, and the question reduces to which picture better accounts for the rest of the phenomena. ↩
David J. Chalmers, “The Meta-Problem of Consciousness,” Journal of Consciousness Studies 25, nos. 9–10 (2018): 6–61, esp. 6–10 and sec. 5, defines the meta-problem as the problem of explaining why we make phenomenal reports and judge consciousness puzzling. Chalmers treats it as physically and functionally tractable while denying that its solution by itself explains or dissolves consciousness. The text uses the program only to explain the grip and public form of the zombie intuition; it draws no independent support for identity from the meta-problem. ↩
The ability hypothesis is developed in David Lewis, “What Experience Teaches,” Proceedings of the Russellian Society 13 (1988): 29–57, reprinted in Mind and Cognition: A Reader, ed. William G. Lycan (Oxford: Blackwell, 1990), 499–519, and Laurence Nemirow, “Physicalism and the Cognitive Role of Acquaintance,” in Lycan, ed., Mind and Cognition, 490–499, esp. sec. III, with later refinements in the secondary literature. The view treats what Mary acquires upon leaving the room as a cluster of know-how — abilities to recognize, imagine, and remember red — rather than any new propositional knowledge. Critics — most prominently Brian Loar, “Phenomenal States,” Philosophical Perspectives 4 (1990): 81–108, esp. 85–86, who defends a different physicalist alternative — press that this seems to underdescribe what Mary gains. The present chapter is friendly to the ability hypothesis as a partial physicalist resource but does not lean on it; the conceivability diagnosis stands on its own. ↩
The phenomenal-concept strategy is developed in Brian Loar, “Phenomenal States,” revised in The Nature of Consciousness: Philosophical Debates, ed. Ned Block, Owen Flanagan, and Güven Güzeldere (Cambridge, MA: MIT Press, 1997), 597–616; David Papineau, Thinking About Consciousness (Oxford: Clarendon Press, 2002), chs. 4–7; and Katalin Balog, “In Defense of the Phenomenal Concept Strategy,” Philosophy and Phenomenological Research 84, no. 1 (2012): 1–23. Chalmers’s master objection appears in “Phenomenal Concepts and the Explanatory Gap” (cited n. 13). The house account adopts no generic strategy wholesale: it appeals to acquaintance-grounded recognition to explain conceptual isolation given the identity, while leaving the identity and modal burden to Chapters 6 and 11. ↩
Copeland is the authority on the careful statement of the thesis and on the narrow/wide-mechanism distinction the bad inference trades on — see B. Jack Copeland, “The Church–Turing Thesis,” Stanford Encyclopedia of Philosophy, esp. secs. 1 and 5.1–5.3, and “Narrow Versus Wide Mechanism,” Journal of Philosophy 97 (2000): 5–32, esp. 15 and 20–25. Church and Turing published independently in 1936 (Church using λ-definable functions, Turing the abstract machines now named for him); the formalisms proved equivalent. The thesis has the form: for any function f, if f is intuitively computable, then some Turing machine computes f. It does not say that every physical system is a computer, that every physical process is a computation, or that a computable mind is a program — each a further claim requiring its own argument. ↩
Gualtiero Piccinini, “Computationalism, the Church–Turing Thesis, and the Church–Turing Fallacy,” Synthese 154, no. 1 (2007): 97–120, at 99. Piccinini distinguishes computationalism proper — the empirical thesis that cognition consists in the brain’s computation of certain Turing-computable functions — from the far weaker claim that brain functions are Turing-computable (which follows from physicalism plus the Church–Turing–Deutsch principle). The inference from the latter to the former is what Jack Copeland called the Church–Turing fallacy; Piccinini adopts Copeland’s name and supplies the anatomy, showing the influential instances unsound. ↩
The three requirements track Piccinini’s account of physical computation; see Piccinini, “Computational Modelling vs. Computational Explanation: Is Everything a Turing Machine, and Does It Matter to the Philosophy of Mind?,” Australasian Journal of Philosophy 85 (2007): 93–115, esp. 111–13, and Physical Computation: A Mechanistic Account (Oxford: Oxford University Press, 2015), esp. ch. 7. Piccinini’s mechanistic account — a system computes only if it possesses a mechanism functionally defined over discrete state-types that it manipulates by rule — is the closest thing in the current literature to a substantive criterion, and it is precisely what makes computationalism a falsifiable empirical hypothesis rather than a trivial truth. On his view brains may turn out to be computers; the point is that the question is empirical and substantive, not settled from the armchair by a theorem about computable functions. ↩
David Deutsch, “Quantum Theory, the Church–Turing Principle and the Universal Quantum Computer,” Proceedings of the Royal Society of London A 400 (1985): 97–117, at 99: “every finitely realizible physical system can be perfectly simulated by a universal model computing machine operating by finite means” (the spelling is the paper’s own). Deutsch’s statement strengthens and physicalizes the Church–Turing thesis — and he is explicit that classical physics and the universal Turing machine, the one continuous and the other discrete, do not obey the principle in this strong form; quantum theory and the universal quantum computer are what render it satisfiable. For the dissenting bet on non-computable physics, see Roger Penrose, The Emperor’s New Mind (Oxford: Oxford University Press, 1989), esp. chs. 4 and 10. Section 22.6 grants the principle in full and shows what the grant does and does not deliver. ↩
The error-theoretic or projectivist line — that physical surfaces instantiate no colors and that color experience therefore misrepresents across the board — receives its most empirically engaged statement in C. L. Hardin, Color for Philosophers: Unweaving the Rainbow (Indianapolis: Hackett, 1988), esp. chs. 2–4, and its sharpest analytic formulation in Paul A. Boghossian and J. David Velleman, “Colour as a Secondary Quality,” Mind, New Series, 98, no. 389 (1989): 81–103, esp. 81–83 and 96–98, who argue that visual experience presents colors as intrinsic qualities of surfaces that surfaces do not possess (they reject dispositionalism en route to that projectivism). For the rejoinder the text presses — that global color eliminativism undercuts the perceptual contact it presupposes — see Alex Byrne, “Color and the Mind-Body Problem,” Dialectica 60, no. 3 (2006): 223–244, esp. 227–28, and Alex Byrne and David R. Hilbert, “Color Realism and Color Science,” Behavioral and Brain Sciences 26, no. 1 (2003): 3–21, esp. secs. 2–3. Hardin also presses interpersonal variation in unique-hue judgments — normal observers divide, stably, over which chips look green untinged by blue or yellow — as evidence that no perceiver-independent color is there to be seen; for the concession-and-reply rehearsed in the text (the represented property is the coarse-grained reflectance-type, so small, stable misrepresentation of its fine grain by most perceivers is a prediction of the view, not an embarrassment to it), see the exchange in Analysis 66, no. 4 (2006): Jonathan Cohen, C. L. Hardin, and Brian P. McLaughlin, “True Colours,” 335–340, and Michael Tye, “The Truth about True Blue,” 340–344. ↩
The dispositional account, in the secondary-quality tradition descending from John Locke, An Essay Concerning Human Understanding (1689), II.viii.10, identifies a color with a surface’s disposition to produce experiences of that color in normal perceivers under normal conditions. For the contemporary map of options — physicalist, dispositionalist, projectivist, and primitivist — and the defenders of each, see David J. Chalmers, “Perception and the Fall from Eden,” in Perceptual Experience, ed. Tamar Szabó Gendler and John Hawthorne (Oxford: Oxford University Press, 2006), 49–125, esp. secs. 3 and 6. The objection registered in the text — that the analysis fixes the worldly property by reference to the very experience it was meant to anchor, reintroducing the perceiver into the object — is a standard complaint against dispositionalism; cf. Boghossian and Velleman, “Colour as a Secondary Quality,” esp. 84–95. The relational family deserves separate treatment. Jonathan Cohen identifies colors with relations among surfaces, perceivers, and viewing conditions in The Red and the Real: An Essay on Color Ontology (Oxford: Oxford University Press, 2009), esp. chs. 1 and 6. Daniel Dennett gives the ecological version in Consciousness Explained (Boston: Little, Brown, 1991), ch. 11: colors and color-vision systems coevolved and the resulting properties depend for their individuation on a class of perceivers without becoming inner figments. Alva Noë treats colors as objective patterns in how surfaces vary with illumination and environmental conditions in Action in Perception (Cambridge, MA: MIT Press, 2004), ch. 4. These views show why a bare charge of circularity or inward placement does not answer every relational account. ↩
Primitivism treats colors as simple, mind-independent properties not reducible to physical (reflectance) or dispositional properties. The cleanest statement is John Campbell, “A Simple View of Colour,” in Reality, Representation, and Projection, ed. John Haldane and Crispin Wright (New York: Oxford University Press, 1993), 257–268; Colin McGinn, “Another Look at Color,” Journal of Philosophy 93, no. 11 (1996): 537–553, is often grouped here for treating color as irreducible, though his framing in terms of the disposition to look colored leaves the classification genuinely contested. The position and its Edenic variant are mapped in Chalmers, “Perception and the Fall from Eden,” 49–125, esp. sec. 6 on primitivism and secs. 8–9 on Edenic content. The text’s objection — that an irreducible worldly color reinstates, for color alone, precisely the explanatory gap the book refuses elsewhere — restates the standard physicalist complaint that primitivism leaves color metaphysically and causally dangling. ↩
Reflectance physicalism identifies a color with a type of surface spectral reflectance — equivalently, with the surface’s physical disposition to modify incident light (a light-reflecting disposition the surface has in itself, distinct from the disposition-to-look-colored on which the dispositionalism of the preceding note rests). The type is individuated rather than single because metameric surfaces differing in reflectance match perceptually. The canonical defense is Alex Byrne and David R. Hilbert, “Colors and Reflectances,” in Readings on Color, Volume 1: The Philosophy of Color, ed. Byrne and Hilbert (Cambridge, MA: MIT Press, 1997), 263–288, and “Color Realism and Color Science,” Behavioral and Brain Sciences 26, no. 1 (2003): 3–21; see also David R. Hilbert, Color and Color Perception: A Study in Anthropocentric Realism (Stanford: CSLI, 1987). Michael Tye builds the same physicalist identification into the representational theory, treating colors as types of spectral reflectance: Ten Problems of Consciousness (Cambridge, MA: MIT Press, 1995), ch. 5, and Consciousness, Color, and Content (Cambridge, MA: MIT Press, 2000), chs. 6–7. The reflectance-type individuation that the metamerism point forces (discussed in “Color Realism and Color Science,” sec. 3) is anthropocentric, and avowedly so — hence Hilbert’s subtitle — but it does not reintroduce the response-dependence the body rejects in dispositionalism. Human vision fixes which perceiver-independent physical properties are grouped under a color term (the individuation of the type), not whether a surface instantiates the property (its constitution). Noë’s Action in Perception, ch. 4, presses the best objection: surface spectral reflectance explains constancy but appears to omit color’s phenomenal and context-sensitive organization. The main text answers by dividing the work. Ecological transformations and the visual system’s quality space constrain how a perceiver detects and groups the distal property; the conscious state’s visual mode supplies the phenomenality. The reflectance-type supplies the stable accuracy condition across those transformations. A difference in microphysical reflectance that produces no difference anywhere in the complete ecological profile need not produce a different phenomenal color and may fall within one anthropocentrically individuated type. ↩
John Searle, Intentionality: An Essay in the Philosophy of Mind (Cambridge: Cambridge University Press, 1983), p. 1. Searle does not leave the concession wholly bare: the next sentence offers a reporting-constraints test, which he then runs on beliefs and desires, and at pp. 3–4 he counts depression and elation among states that at least can be directed. Tim Crane’s complaint therefore belongs at the narrower target: Searle treats the existence of non-intentional mental states “as obvious, so obvious that it is not in need of further argument or elucidation.” Crane, “Intentionality as the Mark of the Mental,” in Contemporary Issues in the Philosophy of Mind, ed. Anthony O’Hear (Cambridge: Cambridge University Press, 1998), 229–251, at 230. ↩
The orientation reading in the body — the mood as a way the world as a whole gets presented, menacing or open — is nearest Michael Tye, Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind (Cambridge, MA: MIT Press, 1995), ch. 4. Angela Mendelovici defends intentionalism by a different route: moods represent unbound affective properties while lacking intentional objects. See “Intentionalism about Moods,” Thought 2, no. 1 (2013): 126–136, and “Pure Intentionalism about Moods and Emotions,” in Current Controversies in Philosophy of Mind, ed. Uriah Kriegel (London: Routledge, 2014). The distinction matters: Mendelovici preserves content while rejecting the claim that every contentful mood must aim at an object. For the counter-tradition pressing existential feelings as genuinely pre-intentional, see Matthew Ratcliffe, Feelings of Being: Phenomenology, Psychiatry and the Sense of Reality (Oxford: Oxford University Press, 2008). The body therefore states a bounded verdict rather than a universal proof. ↩
On the Scholastic background, the medieval distinction between esse intentionale (intentional being, the object’s being-in the intellect) and esse reale (real being, the object’s being in the world) is the source of Brentano’s phrase; see Franz Brentano, Psychology from an Empirical Standpoint, trans. Rancurello, Terrell, and McAlister (London: Routledge, 1995 [1874]), p. 88 and the footnote there, with exposition in Tim Crane, “Brentano on Intentionality,” in Uriah Kriegel, ed., The Routledge Handbook of Franz Brentano and the Brentano School (London: Routledge, 2017), 41–48, at 43–45. For sustained recent treatment of the puzzle and its options, see Tim Crane, “What Is the Problem of Non-Existence?”, Philosophia 40, no. 3 (2012): 417–434, and Crane, The Objects of Thought (Oxford: Oxford University Press, 2013). ↩
Bertrand Russell, “On Denoting,” Mind 14, no. 56 (1905): 479–493. Russell’s theory of descriptions analyzes apparent reference to nonexistent objects into quantified propositional content, so that “the present King of France is bald” comes out false (not truth-valueless and not committed to a subsisting bald non-king) rather than requiring a Meinongian object to be its subject. Alexius Meinong’s contrasting view — the Gegenstandstheorie or theory of objects, on which even the round square is an object that merely lacks being — is set out in his “Über Gegenstandstheorie” (1904); the standard primary source in English is “The Theory of Objects,” trans. Isaac Levi et al., in Realism and the Background of Phenomenology, ed. Roderick Chisholm (Glencoe, IL: Free Press, 1960), 76–117, esp. secs. 3–4. The contemporary verdict is not that Meinong was simply confused — his view is more defensible than Russell’s polemic allowed — but that the explanatory work can be done without the inflated ontology; see Crane, The Objects of Thought (2013), ch. 1, for a measured reassessment. ↩
The weak/strong distinction has become standard since Chalmers drew it sharply: weak emergence names high-level facts unexpected from, yet deducible in principle from, the base; strong emergence names facts “not deducible even in principle” from it. See David J. Chalmers, “Strong and Weak Emergence,” in The Re-Emergence of Emergence, ed. Philip Clayton and Paul Davies (Oxford: Oxford University Press, 2006). Searle drew the same line in 1992 in his own vocabulary: a feature is emergent1 when its existence must be explained by the causal interactions among a system’s elements — solidity, liquidity, and, on his view, consciousness — and emergent2 when it additionally has causal powers that cannot be explained by those interactions. He endorses the first and rejects the second: “on my view consciousness is emergent1, but not emergent2.” John R. Searle, The Rediscovery of the Mind (Cambridge, MA: MIT Press, 1992), ch. 5, at 111–112. Biological naturalism’s approving talk of consciousness as a causally emergent property of systems is therefore weak emergence throughout — emergent1 maps onto the weak sense, emergent2 onto the strong — and concedes nothing to the doctrine this section rejects. ↩
The lineage runs through Mill, Samuel Alexander, and C. D. Broad; the canonical history and post-mortem is Brian P. McLaughlin, “The Rise and Fall of British Emergentism,” in Emergence or Reduction?, ed. Ansgar Beckermann, Hans Flohr, and Jaegwon Kim (Berlin: de Gruyter, 1992), 49–93, esp. sec. 6. The contemporary revival of strong emergence for consciousness is the same doctrine recast for the mind–body case. ↩
For the property-dualist moral drawn from treating consciousness as, at best, strongly emergent, see Chalmers, The Conscious Mind (New York: Oxford University Press, 1996), esp. ch. 4 — the locus of “naturalistic dualism.” The panpsychist reaches the same neighborhood from the far side: holding that phenomenal character could not strongly emerge from the wholly non-phenomenal, Goff concludes it must instead be fundamental — Philip Goff, “How Exactly Does Panpsychism Help Explain Consciousness?”, Journal of Consciousness Studies 31, no. 3–4 (2024): 56–82. Either route seats felt character among the basic features of the world; that shared conclusion, not the direction of travel, is what the present view rejects. The book’s reply to the panpsychist route runs in Chapter 10. ↩
Tim Crane, “The Significance of Emergence,” in Physicalism and Its Discontents, ed. Carl Gillett and Barry Loewer (Cambridge: Cambridge University Press, 2001), esp. sec. 4. On examination, the notion of an emergent property fails to mark a stable boundary between emergentism and non-reductive physicalism — a result that cuts against leaning on “non-reductive” as a safe harbor. ↩
Michael Tye, Ten Problems of Consciousness: A Representational Theory of the Phenomenal Mind (Cambridge, MA: MIT Press, 1995), 49–50; cf. 63. Tye there raises, as a consequence to be rejected, the suggestion that “phenomenal consciousness emerged from the physical world at some stage in the history of evolution without there being any explanation whatsoever of its emergence” — brute emergence he regards as tantamount to magic and “very hard to believe.” His later turn to panpsychism (Vagueness and the Evolution of Consciousness, 2021, ch. 4) lands him on the other horn of the same dilemma, making phenomenal character fundamental rather than emergent. The book’s reply to that conversion — the epistemic-slide diagnosis together with the idle-wheel objection — appears in Chapter 10. ↩
David J. Chalmers, “Phenomenal Concepts and the Explanatory Gap,” in Phenomenal Concepts and Phenomenal Knowledge: New Essays on Consciousness and Physicalism, ed. Torin Alter and Sven Walter (New York: Oxford University Press, 2006), 167–194, esp. secs. 2–4. Chalmers requires topic-neutral formulations of C and E. His compact argument is: P&~E is conceivable; therefore either P&~C or C&~E is conceivable; in the first case P does not transparently explain C, and in the second C does not explain E. Relevant replies include Peter Carruthers and Bénédicte Veillet, “The Phenomenal Concept Strategy,” Journal of Consciousness Studies 14, nos. 9–10 (2007): 212–236; Esa Díaz-León, “Can Phenomenal Concepts Explain the Epistemic Gap?”, Mind 119, no. 476 (2010): 933–951; and Katalin Balog, “In Defense of the Phenomenal Concept Strategy,” Philosophy and Phenomenological Research 84, no. 1 (2012): 1–23. ↩
The topic-neutral C used here is an acquaintance-grounded recognitional architecture: an occurring world-presenting state anchors a demonstrative token and enables later re-identification. The recognitional lineage runs through Brian Loar, “Phenomenal States,” Philosophical Perspectives 4 (1990): 81–108, esp. 87–90, revised in The Nature of Consciousness, ed. Block, Flanagan, and Güzeldere (Cambridge, MA: MIT Press, 1997), 597–616. The book adds the world-directed object of recognition. It does not claim C explains all of E. It takes Chalmers’s second horn: C explains architecture and conceptual isolation; the independently defended identity fixes the referent, truth, and justification of Mary’s phenomenal judgment. ↩
Philip Goff, “A Posteriori Physicalists Get Our Phenomenal Concepts Wrong,” Australasian Journal of Philosophy 89, no. 2 (2011): 191–209, doi:10.1080/00048401003649617. Goff distinguishes transparent, translucent, mildly opaque, and radically opaque concepts and argues that a posteriori physicalism must either make phenomenal concepts opaque or accept two conceptually distinct ways of knowing what it is for one property to be instantiated. His argument requires only translucency, not complete essence disclosure. The text accepts limited disclosure of the manifest world- or body-presenting profile while denying disclosure of total categorical or neural realization; this differs from the stronger revelation claims pressed by Martine Nida-Rümelin, “Grasping Phenomenal Properties,” in Phenomenal Concepts and Phenomenal Knowledge, ed. Alter and Walter (New York: Oxford University Press, 2007), 307–338. ↩
The case originates with Donald Davidson, “Knowing One’s Own Mind,” Proceedings and Addresses of the American Philosophical Association 60 (1987): 441–458 (delivered 1986). Ruth Millikan and Fred Dretske bite the historical bullet: without selectional or learning history, the duplicate’s states have no proper functions and therefore no content. See Millikan, Language, Thought, and Other Biological Categories (Cambridge, MA: MIT Press, 1984), chs. 1–2; and Dretske, Naturalizing the Mind (Cambridge, MA: MIT Press, 1995). Michael Tye formulates the pressure as a choice between teleological normal tracking, which handles Inverted Earth but loses Swampman, and non-teleological tracking, which reverses the result: Consciousness, Color, and Content (Cambridge, MA: MIT Press, 2000), 121–122. Ned Block presses the consequence against representationalism in “Mental Paint,” in Reflections and Replies: Essays on the Philosophy of Tyler Burge, ed. Martin Hahn and Bjørn Ramberg (Cambridge, MA: MIT Press, 2003), 165–200. The book takes the Millikan–Dretske verdict at t = 0 and accepts Block’s consequence: content and phenomenal character are both absent, even though the current physical duplicate is exact. ↩
The present standard is homeostatic, not yet semantic. Hans Jonas, The Phenomenon of Life (New York: Harper & Row, 1966), and Evan Thompson, Mind in Life (Cambridge, MA: Harvard University Press, 2007), esp. chs. 5–6, supply the enactivist basis for treating precarious self-maintenance as a source of organism-relative functional standards without appeal to ancestry. The book uses that account for autonomy, practical interest, and welfare. It does not infer content from present covariation or present need. Section 15.1’s thermometer lesson still holds: a selected producer-consumer function and learning history answer what a state is correct about. Collapsing viability and semantic address was the earlier reply’s mistake. ↩
The proposal that the apparent sharpness lives in our concept of phenomenal consciousness rather than in the property itself belongs to the broader “phenomenal concept strategy” — the view that our first-person, recognitional concepts of experience have features (here, all-or-nothing application) that need not be features of the physical properties they pick out. The locus classicus is Brian Loar, “Phenomenal States,” in The Nature of Consciousness, ed. Ned Block, Owen Flanagan, and Güven Güzeldere (Cambridge, MA: MIT Press, 1997), 597–616. The position that the concept itself is vague is David Papineau’s, not Antony’s (Thinking About Consciousness [Oxford: Oxford University Press, 2002], sec. 7.16), and Michael V. Antony is among its critics (“Papineau on the Vagueness of Phenomenal Concepts,” Dialectica 60, no. 4 [2006]: 475–483). Antony’s own position is the one the main text engages: our concept of consciousness may well be sharp — sharp enough that no borderline case can be framed with it — while the worldly property it picks out is vague, the concept’s sharpness being no guarantee of the property’s (“Vagueness and the Metaphysics of Consciousness,” Philosophical Studies 128, no. 3 [2006]: 515–538). Tye considers and rejects both moves in the 2024 essay (which contains his references to them), insisting that the property, not merely the concept, is sharp. The diagnosis offered in the main text applies the concept strategy to that premise specifically: a recognitional/demonstrative phenomenal concept would present its referent as all-or-nothing in application regardless of whether the underlying physical kind is vague, so the felt sharpness is what the strategy predicts rather than independent evidence of a sharp worldly joint. The book’s general use of the strategy against the epistemic-to-ontological slide runs through Chapter 8 (Mary), Chapter 9 (zombies), and Chapter 11 (the deflation of the hard problem); the point here is only that Tye’s sharpness premise shares the structure and the vulnerability. ↩
The homunculus-headed nation is Ned Block’s: “Troubles with Functionalism,” in C. Wade Savage, ed., Perception and Cognition: Issues in the Foundations of Psychology, Minnesota Studies in the Philosophy of Science 9 (Minneapolis: University of Minnesota Press, 1978), 261–325, esp. 279–81. The orthodox bullet-biting reply: the nation does feel pain, however strange, and the contrary intuition is parochialism — our judgments were trained on creatures with faces and are unreliable witnesses about radio networks and populations. There is something to this; intuition-mongering over exotic systems proves little on its own, and were absent qualia the whole case against functionalism the bullet-biter might chew his way out. The body’s concession is deliberate and this note keeps its books the same way: the bullet-biter stipulates nothing he was obliged to derive — sufficiency is his thesis — and what his position costs is the accumulated comparison the body names: a felt point of view attributed from internally specified organization without independent evidence of the complete world-involving, perceptually organized, and poised content the identity claim requires. For the bite-the-bullet line see the Stanford Encyclopedia of Philosophy, “Qualia,” on functionalist responses to the China-body system. ↩
“Long-arm” or “wide” functional role extends functional individuation to include the system’s causal relations to items in its environment. Gilbert Harman’s “long-arm” conceptual-role semantics lets the relevant roles reach all the way out to worldly referents. William Lycan’s homuncular functionalism makes a different move: it replaces the flat transition-table model with a hierarchy of teleologically characterized subsystems; it does not by itself supply Harman’s externalism. The concession in the text applies to the genuinely wide view: one may classify world-involving content as functional role if one likes. The point is that doing so abandons exactly the thesis under test — machine functionalism, which reads mental states off the internal organization with the world bracketed, and which Block’s nation satisfies. Once content depends on the environment, internally identical systems can differ mentally, and organization-alone is conceded insufficient. The dispute then becomes terminological. See Gilbert Harman, “(Nonsolipsistic) Conceptual Role Semantics,” in New Directions in Semantics, ed. Ernest LePore (London: Academic Press, 1987), 55–81, esp. the section “Conceptual Role and External World”; William Lycan, Consciousness (Cambridge, MA: MIT Press, 1987), esp. 40–44. ↩
The distinctions should remain separate. (1) Persistence supplies duration. (2) Self-maintenance adds activity directed toward continuation. (3) Thermodynamic self-maintenance requires the activity to help constitute the continued physical organization doing the activity; a candle flame qualifies here too. (4) Adaptive regulation distinguishes movement toward and away from viable conditions and varies the response; reciprocal closure may thereby support autonomy, practical interest, and welfare. (5) Semantic grounding / address is fixed by selection, learning, tracking, and producer-consumer use (Chapter 15; the Swampman resolution, Section 15.6). (6) Whole-system understanding requires flexible integration of content across a cognitive economy. Perceptual organization and poising supply further conditions on phenomenal content. Restorability appears nowhere in these tests: it may provide evidence about how a present architecture is maintained, but copying creates or removes none of the relevant relations by itself. ↩
Angela Mendelovici, “Reliable Misrepresentation and Tracking Theories of Mental Representation,” Philosophical Studies 165, no. 2 (2013): 421–443. Mendelovici is explicit that her target is systematic misrepresentation, not the disjunction problem (the occasional fly/BB error). Her modal claim — stated in the abstract and sec. 1 — is that tracking theories render standing, lifelong mismatch between how things seem and how they are “practically impossible,” because reliable tracking is part of what such theories take to fix content. That structural inadequacy, not any single inconsistency, is the consideration she presses. ↩
Michael Tye, Consciousness, Color, and Content (Cambridge, MA: MIT Press, 2000), esp. chaps. 3–4. Tye defends the view that color experience represents real, mind-independent surface properties — a content-realism the strong intentionalist is pushed toward once the Galilean intuition (that color phenomenology presents an intrinsic, non-relational property) is in play. The book is the most sustained attempt to make tracking-style content rich enough to underwrite color phenomenology; whether it succeeds is exactly what the section leaves open. ↩
Karen Neander, A Mark of the Mental: In Defense of Informational Teleosemantics (Cambridge, MA: MIT Press, 2017), esp. chs. 6–8. Neander’s informational teleosemantics aims to fix content more determinately than bare selection-history allows, by appeal to natural-factive information and second-order similarity relations between representation and represented. The proposal sharpens the framework without, on the reading taken here, dissolving Mendelovici’s systematic-misrepresentation worry. ↩
Peter Godfrey-Smith, “Mental Representation, Naturalism, and Teleosemantics,” in Teleosemantics: New Philosophical Essays, ed. Graham MacDonald and David Papineau (Oxford: Oxford University Press, 2006), 42–68, secs. 5–6. Godfrey-Smith, himself sympathetic to the program, raises the structural worry that teleosemantics may underdetermine content — the indeterminacy on which the tighten-or-broaden dilemma in the main text turns. ↩
Fred Dretske, Naturalizing the Mind (Cambridge, MA: MIT Press, 1995), ch. 1. Dretske distinguishes a state’s merely indicating properties (what it nomically correlates with) from its full representational properties (what it has the function of indicating), and lets the two come apart in ways bare selection-history cannot settle — the opening the main text exploits to let the frog’s state mean fly in a sense richer than what selection alone ratified. ↩
Jerry A. Fodor, A Theory of Content and Other Essays (Cambridge, MA: MIT Press, 1990), esp. “A Theory of Content, I and II”; the coarseness objection to teleosemantics is stated in those essays and earlier in Psychosemantics: The Problem of Meaning in the Philosophy of Mind (Cambridge, MA: MIT Press, 1987), ch. 4 (“Meaning and the World Order”). Fodor’s frog case stipulates an ancestral environment in which flies and small-dark-moving-things are co-instantiated, so that snapping at the one and snapping at the other are equally adaptive; selectional fitness then cannot discriminate the content fly from the content fly-or-small-dark-moving-thing, leaving proper function unable to fix the address. The standard reconstruction is in Peter Schulte and Karen Neander, “Teleological Theories of Mental Content,” Stanford Encyclopedia of Philosophy (rev. 2022), sec. 4. Fodor’s positive alternative, asymmetric dependence, holds that a symbol means X if non-X-caused tokenings are asymmetrically dependent on X-caused ones (the wild tokenings would not occur but for the veridical ones, but not conversely). ↩
Nicholas Shea, Representation in Cognitive Science (Oxford: Oxford University Press, 2018), ch. 6, esp. secs. 6.2 and 6.5. Shea separates the frog case into problems of distality, specificity, and disjunction. His answer appeals to the task functions that causally explain stabilization and to the correlations that play an unmediated role in explaining how the system performs those tasks. He explicitly allows the evidence to leave some choices open: task function alone may not decide among fly, flying nutritious object, and object worth eating. The wider suite of mechanisms can add determinacy where different correlations explain different tasks. His deflationary treatment of normativity also supports the distinction drawn in Appendix B.12: representational correctness and biological well-functioning can remain descriptive without reducing to one another. ↩
Fintan Mallory, “Teleosemantics for Neural Word Embeddings,” Mind & Language (online 2026), doi:10.1111/mila.70037, esp. secs. 4–5 and 5.1. Mallory argues that backpropagation supplies the relevant reproduction, regularity, counterfactual dependence, and loss-based selection for internal ANN mechanisms, and concludes that word embeddings can function as intentional signs with “genuine, endogenous” or original descriptive and directive contents under consumer-based teleosemantics. Section 5.1 specifies the descriptive content as a fact about a word-type’s occurrence in contexts and the directive content as an instruction to produce a context distribution. The discussion of softmax (author’s note 7) denies that mathematics alone fixes the probability-distribution interpretation, then grounds that interpretation in comparison with actual distributions and the consumer’s training history. The dictionary observation (author’s note 8) supports distinguishing the embedding’s content from its proximal one-hot code; it does not concede that the embedding lacks original content. The section expressly treats intra-linguistic relations as part of the world. The present assessment therefore leaves originality at this grain unsettled: human ancestry of the vocabulary, corpus, task, or loss alone does not establish that interpreting practice exhausts the relevant accuracy conditions. Coordinated reparameterization may preserve the word-to-context relation; changing the external mapping changes a candidate content-fixing relation. Neither operation alone refutes the contextual-content proposal. Nicholas Shea supplies the separate agent-level question: explanation of a whole agent requires coherence or unity of purpose among the relevant mechanisms (“Realist Representational Explanations of Agency Should Require Some Unity of Purpose,” Philosophy and the Mind Sciences 7 [2026], doi:10.33735/phimisci.2026.12098). The text reserves that further threshold for flexible, counterfactually integrated inference, conflict resolution, learning, correction, planning, and action. ↩
The revision condition has a live test case. Jenny Zhang, Shengran Hu, Cong Lu, Robert Lange, and Jeff Clune, “Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents,” arXiv:2505.22954 (2025), describe a coding agent that rewrites its own code while an expanding archive of variants is scored on benchmark performance. The self-modification and selection history are real; benchmark scores determine which variants persist, while independent infrastructure realizes, evaluates, and archives the lineage. The case may therefore strengthen an attribution of selected mechanism content without settling distal reference, whole-agent integration, autonomy, or consciousness. It shows why those dimensions must be tested separately. ↩
Angela Mendelovici and David Bourget, “Consciousness and Intentionality,” in Uriah Kriegel, ed., The Oxford Handbook of the Philosophy of Consciousness (New York: Oxford University Press, 2020), 560–585, esp. sec. 3; the position is developed at length in Mendelovici, The Phenomenal Basis of Intentionality (New York: Oxford University Press, 2018). Mendelovici and Bourget distinguish the phenomenal intentionality theory from its two main competitors — tracking theories (Dretske, Fodor, Millikan) and functional role theories — and hold that the competitors “face unforgivable problems concerning empirical adequacy, while PIT does not.” The reliable-misrepresentation argument treated earlier in this excursus forms a component of that positive case rather than a freestanding objection; see Mendelovici, “Reliable Misrepresentation and Tracking Theories of Mental Representation,” cited n. 44. The reply offered here does not dispute that phenomenal states carry intentionality. It declines the further claim that they supply its only original source, on two grounds: that unconscious mechanism content is a genuine possibility (Chapter 15), and that grounding the artificial-understanding verdict in phenomenality would settle Chapter 24’s question by stipulation rather than by architecture. ↩
Keith Frankish, “Illusionism as a Theory of Consciousness,” Journal of Consciousness Studies 23, nos. 11–12 (2016): 11–39. The weak/strong distinction is Frankish’s own (sec. 1): weak illusionism concedes phenomenal consciousness while holding that introspection misrepresents some of its features; strong illusionism denies that experience has phenomenal properties at all, replacing them with quasi-phenomenal properties — physical properties that introspection misrepresents as phenomenal. Frankish argues that only the strong form dissolves the hard problem; the weak form inherits it. The special issue containing the paper, with commentaries, was reissued as Illusionism as a Theory of Consciousness, ed. Keith Frankish (Exeter: Imprint Academic, 2017). ↩
Daniel C. Dennett, “Quining Qualia,” in Consciousness in Contemporary Science, ed. A. J. Marcel and E. Bisiach (Oxford: Oxford University Press, 1988), 42–77. Dennett’s target is qualia under their standard fourfold job description — ineffable, intrinsic, private, directly apprehensible — and his method is a battery of intuition pumps designed to show that nothing satisfies all four at once. The relation to the present book is precise: Chapter 6 also denies that any inner item satisfies that description, while preserving world-presenting phenomenal character as representational content of the right kind. Dennett quines the item; the identity claim redescribes the achievement. ↩
David J. Chalmers, “The Meta-Problem of Consciousness,” Journal of Consciousness Studies 25, nos. 9–10 (2018): 6–61. Chalmers classifies illusionism as the view that couples a solution to the meta-problem with a denial of the datum (his “strong illusionism,” secs. 1–2), and argues that the debunking arguments a meta-problem solution licenses can be blocked. Chapter 11, note 10 details how the present book deploys the meta-problem; this entry concerns the fork Chalmers himself marks — whether the functional explanation of gap-judgments debunks them (illusionism) or explains them compatibly with their truth (type-B physicalism, the book’s line). ↩
François Kammerer, “The Hardest Aspect of the Illusion Problem — and How to Solve It,” Journal of Consciousness Studies 23, nos. 11–12 (2016): 124–139, on why illusionism strikes nearly everyone as unbelievable and how an illusionist should explain that resistance; and “Can You Believe It? Illusionism and the Illusion Meta-Problem,” Philosophical Psychology 31, no. 1 (2018), pressing the point that illusionism must also explain the apparent impossibility of believing it. Kammerer’s debunking challenge to acquaintance-based defenses of the phenomenal datum is developed across these papers; the reply in the entry follows Chapter 11’s first objection-and-reply: physical explicability of a recognitional capacity discredits its deliverances only where the explanation orphans them from their subject matter. ↩
David M. Rosenthal, Consciousness and Mind (Oxford: Clarendon Press, 2005), for the canonical statement of higher-order thought theory and the quality-space treatment of mental qualities; Hakwan Lau and David Rosenthal, “Empirical Support for Higher-Order Theories of Conscious Awareness,” Trends in Cognitive Sciences 15, no. 8 (2011): 365–373, for the empirical case from prefrontal involvement in visibility and metacognitive confidence. The constitution reply — the higher-order thought constitutes rather than observes the state’s being conscious — is Rosenthal’s standing answer to inner-eye objections and is engaged at full strength in the entry. ↩
Ned Block, “The Higher Order Approach to Consciousness Is Defunct,” Analysis 71, no. 3 (2011): 419–431; David Rosenthal, “Exaggerated Reports: Reply to Block,” Analysis 71, no. 3 (2011): 431–437. Block’s argument turns on the empty and erroneous higher-order thought: if consciousness is constituted by the higher-order state, a targetless monitor should yield conscious experience with no first-order state at all, which Block takes to reduce the view to absurdity. Rosenthal’s reply accepts the consequence — what it is like for the subject is fixed at the higher-order level, target or none. The entry’s transparency argument prices that acceptance: it makes first-order world-presenting content inessential to phenomenal character, against the introspective datum Chapter 4 established. ↩
Uriah Kriegel, Subjective Consciousness: A Self-Representational Theory (Oxford: Oxford University Press, 2009). Kriegel’s view makes a conscious state represent the world and, in the same stroke, itself, eliminating the separate monitor and with it the possibility of a wholly targetless higher-order state. The entry’s point is that the misrepresentation structure survives internalization: the self-representing component can still mischaracterize the world-representing component, and the theory must again say which component phenomenal character follows. ↩
Giulio Tononi, Melanie Boly, Marcello Massimini, and Christof Koch, “Integrated Information Theory: From Consciousness to Its Physical Substrate,” Nature Reviews Neuroscience 17 (2016): 450–461, for the IIT 3.0-era statement and the clinical program. The current formal account is Larissa Albantakis et al., “Integrated Information Theory (IIT) 4.0: Formulating the Properties of Phenomenal Existence in Physical Terms,” PLoS Computational Biology 19, no. 10 (2023): e1011465, DOI 10.1371/journal.pcbi.1011465. Perturbational-complexity indices are theory-motivated clinical measures, not direct readings of metaphysical Φ. For the functional-equivalence challenge, see Adrien Doerig, Aaron Schurger, Kathryn Hess, and Michael H. Herzog, “The Unfolding Argument: Why IIT and Other Causal Structure Theories Cannot Explain Consciousness,” Consciousness and Cognition 72 (2019): 49–59, DOI 10.1016/j.concog.2019.04.002. The argument shows that recurrent and feed-forward systems can preserve input-output function while the theories assign different consciousness; defenders may reject functional equivalence as the relevant equivalence, leaving an empirical dispute over causal structure. ↩
Giulio Tononi and Christof Koch, “Consciousness: Here, There and Everywhere?,” Philosophical Transactions of the Royal Society B 370 (2015): 20140167, esp. the discussion of functionally equivalent feed-forward systems and computer simulations: on IIT, a feed-forward network implementing the same input-output function as a conscious system has zero Φ and no experience, and a von Neumann simulation of a brain would not be conscious however perfect the simulation. David J. Chalmers notes the same architectural verdict for language models in “Could a Large Language Model Be Conscious?,” Boston Review, August 9, 2023: on IIT, feed-forward transformer stacks lack integrated information. The convergence with Part Four is architectural, not doctrinal; the book’s argument runs through content-fixing history rather than Φ. ↩
Scott Aaronson, “Why I Am Not an Integrated Information Theorist (or, the Unconscious Expander),” Shtetl-Optimized (blog), May 2014, arguing that expander graphs and similar structures can be engineered to enormous Φ while doing nothing recognizable as experiencing; Tononi’s reply, posted in the same exchange, accepts the consequence — a suitably organized grid would be conscious — as a prediction of the theory rather than a refutation. The exchange marks the joint precisely: IIT treats integration as sufficient in principle; the identity claim requires that integration organize world-presenting content before any question of phenomenal character arises. ↩
Benjamin Libet, “Unconscious Cerebral Initiative and the Role of Conscious Will in Voluntary Action,” Behavioral and Brain Sciences 8, no. 4 (1985): 529–566, reports readiness potentials preceding reported awareness of the intention to move by roughly 350–400 ms. Three deflationary points. First, Libet himself defended a conscious “veto” and denied that his results abolish the control relevant to responsibility. Second, Aaron Schurger, Jacobo D. Sitt, and Stanislas Dehaene, “An Accumulator Model for Spontaneous Neural Activity Prior to Self-Initiated Movement,” Proceedings of the National Academy of Sciences 109, no. 42 (2012): E2904–E2913, model the readiness potential as an artifact of averaging stochastic subthreshold fluctuations time-locked to movement onset. Third, Judy Trevena and Jeff Miller, “Brain Preparation Before a Voluntary Action: Evidence Against Unconscious Movement Initiation,” Consciousness and Cognition 19, no. 1 (2010): 447–456, found comparable signals before a decision not to move. A paradigm of arbitrary, reasonless wrist-flexions models nothing about deliberation, which is the capacity actually at issue. ↩
Immanuel Kant, Critique of Pure Reason (1781/1787), A444–451/B472–479 (the Third Antinomy), where transcendental freedom names a causality through which a series of appearances begins of its own accord, set against causality according to nature. Gilbert Ryle, The Concept of Mind (London: Hutchinson, 1949), ch. 1, for “the dogma of the Ghost in the Machine”; his ch. 3 treatment of the will anticipates much of the capacity-based analysis deployed here. The image of the ghost sitting “upstream” of the brain adapts Ryle’s target to the free-will case. ↩
The rigorous statement of the incompatibilist conditional is Peter van Inwagen’s Consequence Argument (An Essay on Free Will, Oxford: Clarendon Press, 1983): if determinism is true, our acts follow from the laws of nature and the state of the world in the remote past, and neither is up to us. Crucially, the argument presses a modal claim about the actual past and the actual laws rather than a thesis about ordinary usage, so it is not met by observing what speakers mean by “could have done otherwise” — the reply in Section 18.5 concedes the modal point and shifts the ground. Van Inwagen does not grant the compatibilist an analysis of ability in exchange; he rejects the conditional analyses on offer (An Essay on Free Will, ch. 4), which is why the reply here defends the capacity reading on its own merits rather than treating it as common ground. For the compatibilist rejoinder that “could have done otherwise” expresses an agent’s ability rather than a physical-modal claim about identical universes, see David Lewis, “Are We Free to Break the Laws?,” Theoria 47, no. 3 (1981): 113–121; the “new dispositionalism” of Kadri Vihvelin, Michael Fara, and Michael Smith develops the ability as a disposition whose manifestation may be determined away. Event-causal libertarianism (Robert Kane, The Significance of Free Will, New York: Oxford University Press, 1996) instead locates the required indeterminism in the neural processes of deliberation itself; its standing difficulty is the luck objection — an undetermined outcome seems attributable to chance rather than to the agent — pressed by, among others, Pereboom (n. 2). ↩
Shaun Nichols and Joshua Knobe, “Moral Responsibility and Determinism: The Cognitive Science of Folk Intuitions,” Noûs 41, no. 4 (2007): 663–685, found that subjects presented with a deterministic scenario in abstract terms tended toward incompatibilist judgments, while subjects presented with a concrete, affect-laden case of the same determinism tended toward compatibilist ones. The upshot for the argument in the main text is that no bare appeal to “what ordinary people mean by ‘could have done otherwise’” can settle the dispute, since the folk response varies with framing — which is why the chapter rests its reply on the excusing conditions internal to the practice of holding responsible rather than on a univocal folk intuition. Their own diagnosis runs the other way. Nichols and Knobe weigh two models of the split and lean, tentatively, toward the performance-error model: the underlying folk competence is incompatibilist, and the concrete, affect-laden cases distort its application (“when they are faced with a truly egregious violation of moral norms… they experience a strong affective reaction which makes them unable to apply the theory correctly”). The rival affective-competence model, on which the emotional response plays a constitutive rather than a distorting role, remains live, and the authors leave the choice unsettled. What the study establishes securely is the split itself. The chapter takes only that: the folk response varies with framing, so no bare appeal to what “we” mean settles the dispute — which is why the argument rests on the excusing conditions and on the ground for the capacity given in Section 18.3 rather than on any folk intuition, including one that would favor it. ↩
Patrick Butlin, Robert Long, Eric Elmoznino, Yoshua Bengio, Jonathan Birch, et al., “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness,” arXiv:2308.08708v3 (2023), esp. 4–6 and secs. 1.2–2.5, doi:10.48550/arXiv.2308.08708. The report treats computational functionalism as a disputed working hypothesis and assigns credence according to functional similarity, support for the source theory, and confidence in the bridge. It explicitly warns that satisfying the indicators would not prove consciousness. ↩
Wes Gurnee, Nicholas Sofroniew, Adam Pearce, Mateusz Piotrowski, Isaac Kauvar, et al., “Verbalizable Representations Form a Global Workspace in Language Models,” arXiv:2607.15495v1, 16 July 2026, https://transformer-circuits.pub/2026/workspace/. The paper’s abstract and secs. 1.1–1.5 support the functional claims stated here. The authors explicitly restrict their conclusion to access-consciousness-like organization and identify the missing clean analogues of recurrent loops, encapsulated modules, and sharp ignition. ↩
Patrick Butlin, “Higher-order Representation in AI,” Philosophy and the Mind Sciences 7 (2026), doi:10.33735/phimisci.2026.12032. The DOI and publication details were verified against the journal site. Butlin’s conclusion allows limited higher-order representation while retaining first-order, persona, and distality-based alternatives. ↩
Simon Goldstein and Cameron Domenico Kirk-Giannini, “A Case for AI Consciousness: Language Agents and Global Workspace Theory,” Journal of Consciousness Studies 33, nos. 7–8 (2026): 61–96; preprint arXiv:2410.11407. Their conclusion remains conditional on GWT and on a broadly computational-functionalist method. ↩
David J. Chalmers, “Could a Large Language Model Be Conscious?,” Boston Review, 9 August 2023; preprint arXiv:2303.07103. Chalmers treats recurrence, world and self models, global workspace, unified agency, senses, and embodiment as defeasible obstacles and engineering routes, while warning against mistaking performance for an operational definition of consciousness. ↩
Patrick Butlin, Derek Shiller, and Jonathan A. Simon, “Introduction: Consciousness in Current AI,” Journal of Consciousness Studies 33, nos. 7–8 (2026): 9–13. The introduction maps the special issue’s disagreements over homeostasis, temporal continuity, GWT and language agents, LLM interpretation, and pluralist or natural-kind approaches. ↩
Jeff Sebo and Robert Long, “Moral Consideration for AI Systems by 2030,” AI and Ethics 5 (2025): 591–606, first published online 11 December 2023, doi:10.1007/s43681-023-00379-1. Their argument uses a non-negligible probability of consciousness as a threshold for moral consideration; the proportionate rule in the main text does not adopt their illustrative numerical threshold. ↩
For the relationalist contest, see Bill Brewer, “Perception and Content,” European Journal of Philosophy 14, no. 2 (2006): 165–181, esp. 165–170, and Perception and Its Objects (Oxford: Oxford University Press, 2011); M. G. F. Martin, “The Transparency of Experience,” Mind & Language 17 (2002): 376–425, https://doi.org/10.1111/1468-0017.00205, and “The Limits of Self-Awareness,” Philosophical Studies 120 (2004): 37–89, esp. 38–41 and 67–72; and Charles Travis, “The Silence of the Senses,” Mind 113, no. 449 (2004): 57–94, esp. 57–60 and 73–78. For John Campbell’s demonstrative-reference argument and its relation to disjunctivism, see Matthew Soteriou, “The Disjunctive Theory of Perception,” Stanford Encyclopedia of Philosophy, substantive revision July 28, 2026, secs. 3.5–3.6. These views agree that successful perception reaches the mind-independent object directly; they divide over whether content can capture that directness or the object must partly constitute the experience. ↩
Ned Block, “Consciousness, Accessibility, and the Mesh between Psychology and Neuroscience,” Behavioral and Brain Sciences 30 (2007): 481–548, esp. sec. 9, distinguishes phenomenal consciousness from access consciousness and uses partial-report paradigms associated with George Sperling and later work to argue that phenomenology overflows cognitive accessibility. The empirical interpretation remains disputed. Even if overflow defeats a strong access requirement, it does not by itself establish intrinsic mental paint; it may instead require a weaker account of availability or a different criterion for which representational states become conscious. ↩
Global-workspace accounts treat availability as broadcast into a capacity-limited workspace; see Stanislas Dehaene, Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts (New York: Viking, 2014). Recurrent-processing accounts locate the relevant transition earlier, in local recurrent perceptual activity; see Victor Lamme, “Towards a True Neural Stance on Consciousness,” Trends in Cognitive Sciences 10, no. 11 (2006): 494–501. For higher-order theories, see Chapter 6, n. 7b. Patrick Butlin, Robert Long, et al., Consciousness in Artificial Intelligence: Insights from the Science of Consciousness (2023), arXiv:2308.08708, secs. 2–3, derives indicators from several leading theories and treats their convergence as evidence rather than proof. That indicator method supports the epistemic policy in the text; it does not establish PANIC or collapse the competing mechanisms into one. ↩

Appendix B — Scholarly Excurses

B.1 Inner Speech and the Vygotskian Story

B.2 Information Talk Is a Metaphor

B.3 The Conceivability Trap

The argument in its exact form

Where type-B physicalism refuses the bridge

What the intuition tracks

Two responses worth naming

B.4 The Church–Turing and Church–Turing–Deutsch Theses

B.5 Color and the Reflectance Account

B.6 The Two Objections to the Mark

The Mood That Looks Like a Counterexample

The Monster That Isn’t There

What the Two Cases Leave Open

B.7 Emergence: The Wrong Question, Extended

B.8 The Phenomenal-Concept Dilemma

Chalmers’s master dilemma

The horn the book takes

Goff’s opacity/translucency challenge

B.9 Swampman and the History Requirement

B.10 Tye’s Sharpness Argument and the Concept/Property Dispute

B.11 Functionalism’s First Two Replies

Objection 1: Bite the Bullet

Objection 2: You Have Only Renamed the Mystery

B.12 The Stake: Sources of a Term

The causal-success objection

Where the system ends

B.13 The Ladder of Distinctions

B.14 Why Restorability Is the Wrong Criterion

B.15 Teleosemantics: The Technical Rounds

Round Three: What fixes fly, if anything?

Round Four: Does training supply a selection history?

Round Five: Does the arrow run the other way?

B.16 Illusionism: The Rival That Agrees About the Gap

B.17 Higher-Order Theories: The Monitor and the Missing Target

B.18 Integrated Information Theory: Integration Without a World

B.19 Free Will Without Origination: The Dispute Behind the Capacity View

The neuroscientific abolition argument

Origination and the ghost’s freedom

The Consequence Argument and identical universes

What ordinary judgment can and cannot settle

Why the chapter keeps the capacity view

B.20 Assessing Artificial Consciousness Under Uncertainty

The bearer comes first

What survives the functionalist bridge

Mapping the indicator families

The current assessment

Interventions that move confidence

Two moral ledgers and one proportionate threshold

B.21 Relationalism, the Rival Accounts of Character, and the Conscious Threshold

The relationalist objection

The conscious threshold: overflow and the competing boundaries

Notes

Share this:

Like this: