The argument in eight lines

The phrase "hard problem" has a birthday. In April 1994, at the first Toward a Science of Consciousness conference in Tucson, Arizona, a young Australian philosopher put the question to his audience of what we need to explain when we explain consciousness.

The hard problem is a challenge. The zombie argument is an attempt to turn that challenge into a proof that materialism is false. People often run the two together, and the difference matters for Leela's question. You can believe there is a hard problem, in the sense that no one has yet explained why physical processing feels like anything, while rejecting the zombie argument's conclusion. In the 2020 PhilPapers Survey, 62.42 per cent of the philosophers who answered said there is a hard problem, and 29.76 per cent said there isn't. In the same survey, 51.9 per cent accepted or leaned towards physicalism about the mind. Those numbers can't both be about the same people unless a good many philosophers believe in the problem and also think it has a physical answer. They are the population this paper is about. Leela, it turns out, belongs to it: she thinks there is something to explain, and she thinks the explanation will be about brains. Her question is how to hold both views at once without cheating on either.

The precise version

Informal versions of the argument go like this. We can conceive of a being physically identical to you, molecule for molecule, with no experience at all: a zombie. If we can conceive of it, it is possible. If it is possible, then the physical facts don't fix the facts about experience, so materialism is false. That version hides most of the action inside the word "conceive". The precise version, which the Stanford Encyclopedia of Philosophy sets out in its supplement on two-dimensional semantics, uses two letters. P is "an exhaustive microphysical description of the real world together with an indication of your current location within it". Q is some truth about experience, such as the claim that someone has phenomenal experiences. Then:

1. Q is true.
2. The conjunction P and not-Q is a priori coherent: no amount of pure reasoning reveals a contradiction in it.
3. If P and not-Q is a priori coherent, there is an epistemically possible scenario that, considered as actual, would make it true.
4. For every such epistemically possible scenario, there is a corresponding metaphysically possible world. This is modal rationalism.
5. The terms in P and Q are semantically stable: the way they pick things out when we consider scenarios as actual matches the way they pick things out when we consider worlds as counterfactual.
6. So if P and not-Q is a priori coherent, it is metaphysically possible (from 3, 4 and 5).
7. If P and not-Q is metaphysically possible, materialism is false.
8. So materialism is false.

Two of these lines are cheap. Line 1 says someone has experiences, and only an eliminativist denies it. Line 7 follows from what materialism means: if the physical facts could be exactly as they are without the experiential facts, the physical facts don't fix everything, and fixing everything is what materialism claims for them. Line 3 is a piece of machinery in Chalmers' framework, and I'll treat it as part of the bridge. That leaves three serious places to push: line 2, which is about what we can coherently conceive; lines 4 and 5 together, which take us from conceiving to possibility; and the definition in line 7, which a Russellian monist reads in a way that lets materialism, or something close to it, survive. Leela's whiteboard had eight lines, and only three of them are worth the eraser.

Why the letters matter

The bridge has two planks because of a lesson learnt from water. Before chemistry, people could coherently conceive of water that wasn't H2O. That didn't make it possible for water to be something else; given that water is H2O, it is H2O in every possible world. Saul Kripke made this case famous: some truths are necessary yet knowable only by investigation. So conceivability can't go straight to possibility. Chalmers' two-dimensional framework repairs the inference by splitting meaning in two. A term's primary intension tells you what it would pick out if a given scenario turned out to be actual. A term's secondary intension tells you what it does pick out in counterfactual worlds, given how the actual world is.

"Water" has different primary and secondary intensions. Its primary intension is roughly "the clear, drinkable stuff in the oceans and lakes around here". In a scenario where that stuff is XYZ, considered as actual, "water" picks out XYZ. Its secondary intension, fixed by the actual world, is H2O everywhere. So "water is not H2O" is a priori coherent, because its primary intension is satisfiable, and it is metaphysically impossible, because its secondary intension isn't. The gap between conceivability and possibility for water comes entirely from the gap between the two intensions.

The zombie argument then says that P and Q are not like "water". Physical terms like "mass" and "charge", Chalmers says, pick out whatever plays a structural role, and they do so in the same way whether you consider a scenario as actual or counterfactual. Phenomenal terms like "pain" pick out the feeling itself, which is presented to us directly, so there is no gap between how pain seems and what pain is. If both kinds of term are stable in that sense, there is nowhere for a water-style gap to hide, and coherence carries straight through to possibility. That is line 5, and it's where I'll argue the argument is most exposed.

What each premise is really claiming

Philosophers have spent thirty years testing each line, and the survey data give a rough picture of where the profession thinks it breaks. The 2020 PhilPapers Survey asked directly about zombies, with four main options. 36.52 per cent said zombies are conceivable but not metaphysically possible. 24.41 per cent said they are metaphysically possible. 16.40 per cent said they are inconceivable. 10.99 per cent were agnostic or undecided, and the rest chose other answers.

Read those answers as votes about lines. "Inconceivable" denies line 2. "Conceivable but not metaphysically possible" denies the bridge, lines 4 or 5. "Metaphysically possible" accepts the bridge and lands on the conclusion, unless the respondent is a Russellian monist who reads line 7 generously. The largest single group, more than a third, puts the eraser on the bridge. That doesn't make them right. Surveys measure what philosophers believe, and what philosophers believe tells you which positions have been developed carefully enough to be held by working professionals. It is a map of where the defenders have built their trenches.

Line 2: can you really conceive a zombie?

There are several kinds of conceiving, and the argument needs the strongest. Prima facie conceivability is what you have when you think about something briefly and see no contradiction. Ideal conceivability is what survives unlimited reflection. Negative conceivability means you can't rule something out; positive conceivability means you can form a detailed picture of it. The argument needs ideal, positive conceivability of P and not-Q, considered as actual. A quick feeling that zombies make sense is not enough.

Type-A materialists deny that zombies pass that test. Daniel Dennett spent decades arguing that when people say they can conceive of a zombie, they are conceiving of something much less than a full physical duplicate. They picture a body that walks and talks and leave out the billions of details of attention, memory, report and self-monitoring, any of which might turn out on reflection to be what experience is. Imagine someone who claims to conceive of a jet engine working perfectly with no air passing through it. They haven't conceived of the thing described. They have failed to think it through. The type-A materialist says the zombie is the same mistake made at a scale where no one can think it through.

The cost of denying line 2 is high, and it's the cost Leela asked about. If a full functional duplicate necessarily has experience, and you can know this by reasoning, then experience is nothing over and above function in a strong sense: the concept of experience is a functional concept. That is where the charge of eliminativism comes from. Many type-A positions accept that something answers to the word "experience" but deny that it has the properties people usually think it has: an intrinsic, private, ineffable quality. Keith Frankish's illusionism is the honest form of this view. It says phenomenal consciousness, in the sense the hard problem is about, is an illusion: introspection represents our states as having qualities they don't have. Illusionists accept the label "eliminativist" about those qualities while insisting they aren't eliminativists about the states, the representations or the experience in an ordinary sense.

Whether that counts as "collapsing into eliminativism" depends on what you thought you were trying to keep. If you wanted to keep a phenomenal property that is exactly as introspection presents it, type-A loses it. If you wanted to keep the states and their effects, type-A keeps everything. Leela's question, as she put it, asks for a denial that keeps experience real in the sense she meant when she treated patients for pain. That points away from line 2.

Lines 4 and 5: from conceiving to possibility

Type-B materialists accept that zombies are coherently conceivable and deny that this makes them possible. The identity between experience and some physical state is, on this view, like the identity between water and H2O: necessary, but knowable only a posteriori. Unlike the type-A view, this one says our phenomenal concept is not a functional concept. Reasoning from the concept alone will never reveal that pain is a physical state. Investigation and theory can show it, the way chemistry showed that water is H2O.

The difficulty is that the water case had a mechanism and the pain case doesn't seem to. Water's gap between conceivability and possibility came from a gap between primary and secondary intensions: from the fact that "water" picks out its referent by superficial features. For the type-B view to work, the phenomenal concept would need a similar gap. But "pain" seems to pick out pain by how it feels, and how it feels is pain. Kripke made this point in 1980: when we say it seems possible that pain could exist without C-fibres firing, we can't explain the appearance away by saying we've confused pain with something that merely feels like pain, because something that feels like pain is pain.

That's the pressure point. Type-B materialists need line 5 to be false for phenomenal terms, and Kripke's point is that it looks true. They therefore need a different explanation of why the identity looks contingent, one that doesn't use the water model. That explanation is called the phenomenal concept strategy, and most of this paper is about whether it works.

Line 7: what materialism has to mean

The third place to push is less discussed, and it's where Russellian monism lives. Physics describes the world in structural and dispositional terms: what mass does, how charge behaves, which equations relate them. It is silent about what, if anything, plays those roles intrinsically. Russellian monists, following Bertrand Russell's analysis of matter, propose that the intrinsic natures behind physical structure are themselves experiential, or proto-experiential. A zombie world is then conceivable because we conceive of the structure without the intrinsic natures. It is impossible because any world with our physics has those natures.

Chalmers counts this as one of the conclusion's live options, which is why the precise form of his argument ends with "materialism is false or Russellian monism is true". It is a respectable escape and it keeps experience fully real. Its cost is that it makes experience, or something like it, fundamental and everywhere, and it then faces the combination problem: how many small experiential natures add up to one unified subject. For Leela's purposes it doesn't count as denying a premise. It accepts the argument and rewrites what the conclusion means.

Four exits and what each one costs

Arguments with several premises have a property that their critics use and their defenders sometimes forget. Even if you find every premise fairly plausible, you can reasonably doubt the conclusion, because the chances of error add up. If the premises were independent and you gave each of five premises a 90 per cent credence, you'd be entitled to about 59 per cent confidence in the conclusion. If you can't assume independence, the guaranteed lower bound is lower: 50 per cent. With credences of 80 per cent on each premise, the independent product is about 33 per cent and the worst-case bound is zero.

This isn't a trick. It is the reason the right question is Leela's: which premise is weakest. A careful reader of the zombie argument should spread credence across lines 2, 3, 4, 5 and the Russellian reading of 7, then see where the total doubt concentrates. The calculator below lets you do that with your own numbers. It assumes the premises are independent for the product and makes no assumption for the bound.

When I run it with my own credences (line 2 at 0.85, line 3 at 0.95, line 4 at 0.8, line 5 at 0.55, and the non-Russellian reading of 7 at 0.85), the conclusion gets about 30 per cent on the product and no guaranteed floor. Most of the doubt comes from line 5. That is a fact about my credences, not about the world, and the purpose of the calculator is to make you declare yours. The rest of this section explains why I set them where I did.

A bridge inspector's rule

Line 2 carries a lot of load, because without it nothing happens. But it has unusual independent support: conceivability claims are tested every time a reader tries to describe a zombie and finds no contradiction, and three decades of people trying to find one have not produced an agreed proof that the scenario is incoherent. Its support is broad, even though it isn't conclusive.

Line 4, modal rationalism, carries load too. Its support comes from other cases. In mathematics, logic and ordinary modal reasoning, we do seem to move from ideal coherence to possibility without trouble, and there is no agreed example of a claim that is ideally, positively conceivable, with stable terms, yet impossible. That is good evidence, but it is evidence that the critic can absorb by saying the consciousness case is the first exception.

Line 5, semantic stability, carries the heaviest load of all, because it's the premise that blocks the water-style escape. And its independent support is thin. The claim that physical terms are stable is defensible: physics defines its terms by structure, and structure is the same whether you consider a world as actual or counterfactual. The claim that phenomenal terms are stable rests on the intuition in thought experiment 1, the sense that we know what pain is by having it. That intuition is introspective. Its only real test is the case under dispute. If materialism is true, our sense of transparent access to experience is a feature of how our brains represent their own states, and there's no reason in advance to expect those representations to reveal their own physical nature. The premise is supported by the very faculty whose reliability the argument is about.

That's the ratio I mean. The bridge, and in particular its phenomenal half, is where the argument's weight meets its thinnest support.

Jackson's turn

Jackson's turn is useful here for a precise reason. His first doubt wasn't a doubt about conceivability or about Mary. It was a doubt about the reliability of our knowledge of qualia, given his own theory of them. That is a challenge to the transparency intuition behind line 5. If what we take ourselves to know about experience is produced by mechanisms that would produce exactly the same beliefs whether or not those beliefs were accurate, then our sense of direct access is not evidence that we have it. The person who made one of the strongest anti-physicalist arguments found his way out by questioning the faculty that makes such arguments feel decisive.

He didn't become an eliminativist. He became a representationalist, holding that experiences are real states that represent properties, and that their character is fixed by what they represent. Whether that view succeeds is another matter. The point for Leela is that a famous anti-physicalist walked out through the bridge premise, and he kept the experiences when he did.

Denying the bridge without losing the experience

The phenomenal concept strategy is the type-B materialist's way of explaining why the identity between experience and the physical looks contingent, without using the water model. Brian Loar, David Papineau and Katalin Balog have given different versions. The shared idea is simple to state. We have two ways of thinking about the same state. One is a physical or functional concept, which picks it out by its causes, effects or neural description. The other is a phenomenal concept, which picks it out directly, by a kind of inner pointing or recognition, often using a sample of the state itself. Because the two concepts are built differently, nothing in either one lets us derive the other a priori. That's why zombies seem coherent. But the two concepts can still co-refer, which is why zombies are impossible.

On this view the explanatory gap is a gap between two concepts, and it doesn't reflect a gap between two kinds of thing in the world. Joseph Levine's 1983 paper, which coined the phrase "explanatory gap", already pointed in this direction: the problem, he said, is that identity statements involving experience leave us feeling something is unexplained, even if they're true. The phenomenal concept strategy accepts the feeling and explains it. The gap is real and permanent, but it's epistemic. It tells you how we think, and it says nothing about how many substances there are.

This keeps experience in full. The state picked out by the phenomenal concept is the experience itself, and the strategy says it exists and is physical. Nothing is eliminated. A type-B materialist can say everything Leela wants to say about pain: that her patients had it, that it was a real state of their brains, and that treating the brain was treating the pain.

Why denying line 5 isn't the same as denying line 2

It helps to be exact about how the strategy denies the bridge. It accepts that P and not-Q is a priori coherent: no amount of reasoning will reveal a contradiction, because the phenomenal concept has no a priori link to physical concepts. So it accepts line 2. It then says phenomenal concepts aren't semantically stable in the required sense, or that the a priori coherence of P and not-Q doesn't reflect a possibility, because the phenomenal concept refers to a physical state through a mechanism we have no a priori access to. It is a denial of line 5, or of the general modal rationalism in line 4 as applied here.

The two denials have different flavours. Denying line 4 means admitting a class of a posteriori necessities that no two-dimensional analysis can explain, so-called strong necessities. Many philosophers dislike that, because it makes the identity between pain and the physical unlike every other necessity we understand. Denying line 5 is subtler. It says the phenomenal concept's mode of presentation doesn't reveal the referent's full nature, even though it feels as if it does. I prefer the second, because it locates the trouble in the concept, where we can study it, rather than in modality, where we can't.

There is a further reason to avoid the strong-necessity route. Every other necessary truth we know about has an explanation of why it is necessary. Water is necessarily H2O because "water" rigidly picks out whatever actually has the watery role, and that turned out to be H2O. Two plus two is necessarily four because of what the numbers and addition are. A strong necessity would be a necessary truth with no such explanation: pain is necessarily brain state N, and there is nothing more to say about why. That is a lot to accept for one case. It also makes the physicalist's position look suspiciously like the dualist's, since the dualist also says there is a law linking brain states and experiences, and that the law is brute. The difference would be that the physicalist's brute law is metaphysically necessary and the dualist's is only natural, and it is hard to see what evidence could ever tell those apart.

The concept-level denial avoids that. It still leaves an identity that we can't derive a priori, but it explains why we can't: the phenomenal concept has no a priori connection to physical concepts because of how it is built, in the same way that "this" and "the tallest building in Bengaluru" can pick out the same thing without either telling you about the other. The identity is brute from the point of view of our concepts, and explicable from the point of view of a theory of how those concepts work. That is a bargain worth taking, if the theory of concepts can be delivered.

Chalmers' dilemma

Chalmers anticipated this move and answered it with what he calls the master argument against the phenomenal concept strategy. Let C be whatever thesis about phenomenal concepts the strategy uses to explain our epistemic situation: our sense of a gap, our conviction that zombies are conceivable, our feeling of direct access. Then ask whether P and not-C is conceivable: whether we can conceive of a physical duplicate of us that lacks the features C describes.

Either it is conceivable or it isn't. If P and not-C is conceivable, then C is not physically explicable, by the same reasoning as the zombie argument, so the strategy has explained the gap using something as mysterious as consciousness. If P and not-C isn't conceivable, then C must be the kind of thing a zombie would also have, because zombies are physical duplicates. But a zombie, Chalmers says, isn't in our epistemic situation: it lacks phenomenal knowledge and its "phenomenal" concepts don't refer to experiences. So C can't explain our epistemic situation, which is the whole job.

This is a strong argument, and anyone who denies line 5 has to answer it. Leela's question can now be put precisely. Can you deny the bridge, using the phenomenal concept strategy, without being driven by the master argument into either dualism (the first horn) or eliminativism (the second horn, where you say zombies and we are in the same epistemic situation, so we have no phenomenal knowledge beyond what they have)?

Taking the second horn

My answer is to take the second horn and refuse its last step. Grant that the features C describes are physical, so a zombie twin would have them. Then the question is whether a creature with C is "in our epistemic situation". Chalmers says no, because the zombie lacks phenomenal knowledge. Katalin Balog's defence of the strategy replies that this begs the question. The epistemic situation the strategy must explain is the one we can describe in neutral terms: we find zombies coherent, we sense a gap, we judge that we know our experiences directly, and we can't derive phenomenal truths from physical ones. All of that is shared by the twin. What the twin is said to lack, actual phenomenal knowledge, is something it can lack only in a zombie world, and the materialist says there are no zombie worlds. Assuming the twin is epistemically worse off already assumes the zombie argument's conclusion.

I think this reply works as a defence. It blocks the master argument from settling the matter a priori. But look at what it costs, because this is where Leela's question gets its real answer. On the second horn, everything that makes our knowledge of experience feel special is explained by the structure of our concepts, and a creature with the same structure would have the same feeling whether or not its concepts referred to experiences. So the felt transparency of experience is not evidence of transparency. Our conviction that we grasp pain as it is in itself is produced by mechanisms that would produce it either way.

That is a partial illusionism. It isn't illusionism about experience: the phenomenal concept still refers to a real state, the state is the pain, and the pain is as real as anything in the brain. It is illusionism about the mode of presentation. The strategy has to say that experience is presented to us as if its whole nature were revealed, and that this appearance is misleading, because its nature includes physical features the presentation doesn't show. Philosophers call the thesis that experience reveals its own essence "revelation", and the strategy must deny it.

So here is the precise result. You can deny the weakest premise without collapsing into eliminativism, because the denial keeps experiences as the referents of phenomenal concepts. You can't deny it without accepting that introspection misleads us about the nature of what it presents. That's the price. It's lower than eliminativism and higher than most type-B materialists advertise.

Mary under the phenomenal concept strategy

The meta-problem turns this into a test

So far this has been philosophy in the armchair. The way out of the armchair is the meta-problem, which Chalmers named in a 2018 paper in the Journal of Consciousness Studies. The meta-problem is the problem of explaining why we think there is a hard problem: why we judge that experience is mysterious, that zombies are conceivable and that physical explanation leaves something out. Those judgements are behaviour, so they are fair game for ordinary science. Everyone in the debate, including Chalmers, agrees that the meta-problem has a physical solution in principle, because a zombie twin would make the same judgements.

That agreement is useful, because it puts every view's commitments on the same table. The type-B materialist says gap judgements come from the structure of phenomenal concepts: recognitional, isolated from functional concepts and partly constituted by samples. The illusionist says they come from introspective representations that misrepresent our states as having simple, intrinsic, ineffable qualities. The dualist says they come from whatever physical mechanism the zombie twin uses too, and that in our case the judgements are also, somehow, about the experiences. These are different hypotheses about the mechanism, and they make different predictions about which features of a cognitive system produce gap talk.

Machines that write about their inner lives

Until recently, the only systems that produced judgements about the hard problem were human beings, and it is hard to run controlled experiments on the concepts of human beings. That has changed. Large language models now produce fluent paragraphs about whether they are conscious and about what it is like to be them, and researchers can intervene on their internal states. In one recent study, Jack Lindsey at Anthropic injected representations of specific concepts into a model's activations and asked whether it noticed. At the best layer and strength, the model detected and named the injected concept about 20 per cent of the time. That is unreliable, and also well above zero. Some of what such a model says about its own states does depend on those states.

None of this tells us whether such a model is conscious, and I'm not claiming it does. The point is narrower. Language models are a new kind of subject for the meta-problem: systems whose gap judgements we can trace to mechanisms. Most current models learnt to talk about consciousness from human text, so their gap talk could be pure imitation. That is a confound, and a good experiment has to remove it.

Notice what this experiment can and can't do. It can't tell us whether the agents feel anything. It can tell us which features of a cognitive system are enough to produce gap judgements. That is exactly the first debt the phenomenal concept strategy owes: a physical account of C that explains our epistemic situation, described in neutral terms. If family A produces gap talk and family C doesn't, the strategy has paid part of its debt in public. If gap talk turns up only where self-models misrepresent, the strategy shifts towards illusionism. The prediction is falsifiable in the plain sense that a particular pattern of results would count against it.

Where the argument fails, and where it stops

The test above has a failure boundary, and it should be stated plainly. Gap judgements in simple agents might come from something much cruder than either posited mechanism: from the bare fact that a system can represent a state without representing its causes, which every cheap classifier does. If gap talk appears in any system with a recognitional label for an internal state, the experiment will show only that gap talk is cheap. It wouldn't then discriminate between the accounts, and it would tell us that our own gap judgements may need no special explanation at all. That would be a result too, though not the one either side hopes for.

The experiment also has a cost that is easy to underrate. Removing human talk about consciousness from training data isn't a matter of deleting a few words. Human language is soaked in mental vocabulary: "see", "feel", "notice", "seem", "aware". Strip all of it out and the agents may lack the words to report anything about their states, which would make every family look silent. Leave some of it in and the agents may borrow human gap talk through the back door. The practical answer is to let agents develop their own report vocabulary through a communication game with a partner, and then translate their reports by their causal role, which is the method used in work on emergent communication between agents. That translation step is itself a place where the experimenter's assumptions can leak in, and the protocol for it should be fixed and signed before any results are seen, in the way the Two-Key Protocol recommends for rival tests.

There is a counterexample worth taking seriously to the claim that denying the bridge leaves experience untouched. Blindsight patients, after damage to primary visual cortex, can point to objects in the blind field well above chance while saying they see nothing. Their phenomenal concepts and their discriminative capacities have come apart. On the strategy's account, the patients' phenomenal concept of seeing doesn't apply to the blind-field states, because those states aren't the right physical state. That is consistent with the strategy, but it shows the strategy owes something specific: an account of which physical states phenomenal concepts pick out, sharp enough to say why blindsight states are excluded. Without that, the strategy explains the gap and leaves the referent undefined.

Sāṃkhya already drew the zombie

The zombie argument is often presented as a modern, Western puzzle. One Indian tradition built its whole picture of the person around something very like it, more than fifteen centuries ago. Sāṃkhya divides reality into two principles. Prakṛti is nature or matter, including everything we would call mind: buddhi (intellect, the faculty of decision), ahaṃkāra (the sense of "I") and manas (the coordinating mind). Puruṣa is pure consciousness: a witness that doesn't act, think or decide.

That division puts thinking, deciding and saying "I" entirely on the material side. In modern terms, Sāṃkhya's prakṛti includes all of the easy problems, and puruṣa is the hard problem's answer, consciousness itself, distinct from all function. The Sāṃkhyakārikā of Īśvarakṛṣṇa, the classical summary of the system, says so directly in verse 20, as given in the GRETIL edition:

tasmāt tatsaṃyogād acetanaṃ cetanāvad iva liṅgam / guṇakartṛtve 'pi tathā karteva bhavaty udāsīnaḥ

A word-for-word rendering reads: thus, from their union, the liṅga (intellect and the rest), though without consciousness, appears to have consciousness; and though activity belongs to the guṇas, the indifferent one appears to be the agent. The commentator Gauḍapāda, as the same source reports, gives the example of a jar that seems cold when filled with ice and hot when filled with hot water, though the jar itself is neither.

On the Sāṃkhya picture, then, the intellect on its own is acetana, "without consciousness". It thinks, decides and speaks, and it looks conscious only because of its proximity to puruṣa. That is a zombie: a complete functional mind that, considered by itself, has no experience.

Sāṃkhya doesn't just say zombies are conceivable. It says every one of us contains one, and that our ordinary sense of being a conscious thinker comes from mistaking the two principles for one.

The Kashmiri reply

The Pratyabhijñā ("recognition") school of Kashmir Śaivism, founded by Utpaladeva in the tenth century, rejected this picture, and its reason is directly relevant to the bridge premise. Utpaladeva held that consciousness is prakāśa, light or manifestation, and that prakāśa is never bare. Its essential nature is vimarśa, reflective awareness. In the Īśvarapratyabhijñākārikā he wrote, in a verse the Indologist Raffaele Torella translates as: "The essential nature of light is reflective awareness; otherwise light, though 'coloured' by objects, would be similar to an insentient reality, such as the crystal and so on." The Sanskrit, as quoted in a study of the mirror image in this tradition, is:

svabhāvam avabhāsasya vimarśaṃ vidur anyathā / prakāśo 'rthoparakto 'pi sphaṭikādijaḍopamaḥ

The crystal is the key image. A crystal takes on the colour of what is placed near it, just as the Sāṃkhya intellect takes on the appearance of consciousness from puruṣa. Utpaladeva's point is that pure reflection, without a self-aware grasp of what is reflected, is exactly what an insentient thing does. Manifestation that doesn't know itself isn't consciousness at all.

This exchange maps onto the modern debate in a way I find precise, though only in part. Sāṃkhya accepts something like the zombie argument's bridge: it takes the conceivability of a conscious-free functional mind as a guide to how things are, and splits the person in two. Pratyabhijñā rejects a version of that split by insisting that consciousness is inseparable from its self-reflective grasp. In RTM's terms, Sāṃkhya separates L2 (prakāśa, manifestation) from L3 (vimarśa, reflexivity), and Pratyabhijñā says they come together.

Where the bridge between traditions holds and breaks

The parallel is real, and it shouldn't be overstated. Sāṃkhya's argument for puruṣa isn't a conceivability argument. The Sāṃkhyakārikā gives other reasons in verse 17: aggregates exist for the sake of something else, there must be a controller, there must be an enjoyer, and there is activity directed at liberation. Those are teleological and soteriological arguments, and a modern philosopher would weigh them differently. Pratyabhijñā's reply isn't a phenomenal concept strategy either. It is idealist: it holds that the world is manifestation within one consciousness, which is the very opposite of materialism. What the two traditions share with the modern debate is a question: can the functions of mind be separated, even in thought, from the light by which they are experienced? Sāṃkhya says yes and builds a dualism on it. Pratyabhijñā says the separation produces only a crystal.

RTM sits with Pratyabhijñā on that question, and that is worth stating openly, because it shapes how I read the zombie argument. If manifestation and reflexivity are inseparable, then a zombie, a system with all the reflexive functions and no manifestation, is not a coherent description after all. That would make RTM a denier of line 2, not line 5. I resist that move in this paper, because RTM's claim about inseparability is a metaphysical thesis about cit (consciousness as such), not an empirical result, and it shouldn't be used to win an argument that the empirical programme hasn't yet earned. The meta-problem test is the way to earn it.

The strongest case against my answer

The best objection to my answer comes from the anti-physicalist side, and it goes like this. I've said the bridge premise is weakest because its support is introspective and introspection is the faculty in dispute. But every premise of every argument about consciousness rests on introspection somewhere, including the materialist's premise that experiences are states of the brain. We know there are experiences only by having them. If introspective support counts for nothing, then line 1, that Q is true, has no support either, and we are on the road to eliminativism after all. By discounting introspection to save materialism, I've discounted the only evidence for the thing I claim to save.

This is a serious objection and I don't think it can be fully answered. My partial answer is to distinguish what introspection reports. Introspection reports that there are experiences, and that they have a certain character. It also seems to report that the character it presents is the whole nature of the state. The first two reports are tested every day against each other, against behaviour and against other people's reports, and they mostly hold up: people who report pain usually behave as if in pain, and their reports track interventions on their nervous systems. The third report, revelation, has no such test. Nothing we could observe would distinguish a world in which introspection reveals the whole nature of experience from one in which it reveals only some of it. So I discount the third report and keep the first two. That's a principled line, but the anti-physicalist can reasonably reply that the three reports come as a package, and that I am picking from the package to suit my conclusion.

A second objection, from the illusionist side, says I've stopped too early. Once you admit introspection misrepresents the nature of experience, why trust it about the character? Frankish argues that the same considerations which defeat revelation also defeat the claim that there are phenomenal properties at all, and that type-B materialism is an unstable halfway house. If the illusionist is right, my answer to Leela's question is wrong in the other direction: denying the bridge does collapse into eliminativism about phenomenal properties, and I've just hidden the collapse behind a distinction between nature and character.

I take this second objection seriously too, and it is what the meta-problem experiment is for. If gap talk is produced only by misrepresenting self-models, the illusionist has the better explanation and my halfway house falls. If gap talk survives accurate self-models with isolated recognitional concepts, the halfway house has an empirical foundation. I don't know which way the result would go. That is why I'd run the experiment rather than settle the question in this paper.

What would change my mind

What I don't know

I don't know whether the master argument's question-begging charge, which I've used, really cuts the way Balog says. Chalmers has replies, and the exchange hasn't ended. I don't know whether a recognitional concept can be built in a machine at all, as opposed to a recognitional label that merely behaves like one; the difference may matter for the experiment. I don't know how to measure "gap talk" in agents with a non-human vocabulary without importing the very human concepts I've tried to remove. And I don't know whether RTM's thesis that manifestation and reflexivity are inseparable is a claim that any experiment could bear on, or a claim about cit (L5) that stands outside the empirical programme.

A tool for any zombie argument

Arguments of the zombie form are going to multiply, because language models invite them: "it does everything a conscious being does; can we conceive of it without experience?". Here is a worksheet I use to locate where such an argument is pushing and what it costs to resist it.

Claim Ledger

Dated predictions

Glossary

A priori

Knowable by reasoning alone, without investigating the world.

Acetana (अचेतन)

In Sāṃkhya, “without consciousness”: said of prakṛti and everything it produces, including the intellect.

Buddhi (बुद्धि)

In Sāṃkhya, the intellect or faculty of decision, the first product of prakṛti.

Explanatory gap

Levine’s term for the sense that physical explanations leave unexplained why a state feels as it does.

Illusionism

The view that phenomenal consciousness, as introspection presents it, is an illusion produced by introspective representation.

Meta-problem

The problem of explaining why we judge that there is a hard problem of consciousness.

Modal rationalism

The thesis that ideal a priori coherence is a guide to metaphysical possibility.

Phenomenal concept strategy

The physicalist strategy of explaining the explanatory gap by the special nature of our concepts of experience.

Prakāśa (प्रकाश)

Light or manifestation; in Pratyabhijñā, consciousness as that by which anything appears. RTM’s L2.

Prakṛti (प्रकृति)

In Sāṃkhya, nature or matter, including mind as function.

Primary intension

What a term picks out in a scenario considered as actual.

Puruṣa (पुरुष)

In Sāṃkhya, pure consciousness: a witness that does not act or think.

Revelation

The thesis that experiencing a phenomenal property reveals its whole nature.

Secondary intension

What a term picks out in counterfactual worlds, given how the actual world is.

Semantic stability

Having the same primary and secondary intension, so that no water-style gap is possible.

Vimarśa (विमर्श)

Reflective awareness; in Pratyabhijñā, the self-grasping that makes manifestation conscious. RTM’s L3.

Zombie

A hypothetical being physically identical to a conscious one but without experience.

Leela rubs out the fifth line

When the reading group met again, Leela came with her answer written on the back of a pharmacy receipt. She went to the whiteboard and drew a careful line through number 5. "This is the one," she said. "It says that because pain feels like it's showing me everything, it is. I spent thirty-one years watching people be wrong about their own bodies in exactly that way."

Arjun asked whether that made her a materialist. She said it made her someone who thought pain was real, physical and not fully visible from the inside, and that she'd settle for being called whatever that was. Then she added the part this paper has tried to defend. "If I'm right," she said, "somebody should be able to build a machine that is as puzzled as we are, for the same reasons, and a machine that isn't, and tell the difference."

That is where I've ended up too. The weakest premise in the zombie argument is semantic stability: the claim that our concepts of experience reveal its nature. It can be denied without becoming an eliminativist, because the denial keeps experiences as the real referents of those concepts. The denial has a price, a partial illusionism about how experience is presented, and it carries a debt, a physical account of phenomenal concepts that survives Chalmers' dilemma. The debt can be paid only through the meta-problem, by showing which mechanisms produce our puzzlement. That is now a job for experiment as much as for argument.

If you take one practical step from this paper, make it this. Next time you meet a zombie argument, about a person, an animal or a machine, don't argue about whether zombies are possible. Write the argument out in its lines, give each line a number between zero and one, and run the calculator. Then ask which line you'd rub out, and what you'd have to believe about your own mind to rub it out honestly.

Sources

Primary research

• Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies 2(3), 200–219.
• Chalmers, D. J. (2007). Phenomenal concepts and the explanatory gap. In T. Alter and S. Walter (eds), Phenomenal Concepts and Phenomenal Knowledge. Oxford University Press.
• Chalmers, D. J. (2018). The meta-problem of consciousness. Journal of Consciousness Studies 25(9–10), 6–61. philarchive.org/rec/CHATMO-32
• Balog, K. (2012). In defense of the phenomenal concept strategy. Philosophy and Phenomenological Research 84(1), 1–23. philarchive.org/archive/BALIDO-8
• Jackson, F. (1982). Epiphenomenal qualia. Philosophical Quarterly 32(127), 127–136. academic.oup.com/pq/article-abstract/32/127/127/1612468
• Levine, J. (1983). Materialism and qualia: the explanatory gap. Pacific Philosophical Quarterly 64, 354–361. onlinelibrary.wiley.com/doi/10.1111/j.1468-0114.1983.tb00207.x
• Frankish, K. (2016). Illusionism as a theory of consciousness. Journal of Consciousness Studies 23(11–12), 11–39. www.keithfrankish.com/illusionism-as-a-theory-of-consciousness/
• Lindsey, J. (2025). Emergent introspective awareness in large language models. Transformer Circuits Thread. transformer-circuits.pub/2025/introspection/index.html
• Bourget, D. and Chalmers, D. J. (2023). Philosophers on philosophy: the 2020 PhilPapers Survey. Philosophers’ Imprint; results pages for the zombie and hard-problem questions. survey2020.philpeople.org/survey/results/4930
• PhilPapers Survey 2020, results for “Hard problem of consciousness (is there one?)”. survey2020.philpeople.org/survey/results/5042

Primary texts and translations

• Īśvarakṛṣṇa, Sāṃkhyakārikā, GRETIL electronic text. gretil.sub.uni-goettingen.de/gretil/1_sanskr/6_sastra/3_phil/samkhya/i
• Sāṃkhyakārikā 20, word-for-word rendering with Gauḍapāda’s jar example. yogasutrastudy.info/samkhya-karika-20/
• Utpaladeva, Īśvarapratyabhijñākārikā 1.5.11, Sanskrit quoted with translation. www.wisdomlib.org/hinduism/book/cidvilasastava-by-amrtananda/d/doc1546
• Torella, R. (1994; 2002). The Īśvarapratyabhijñākārikā of Utpaladeva with the Author’s Vṛtti. Rome: IsMEO; Delhi: Motilal Banarsidass (verse 1.5.11 as quoted on Wikipedia). en.wikipedia.org/wiki/Utpaladeva
• Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press.

Secondary or leads

• Stanford Encyclopedia of Philosophy, Two-dimensional semantics, supplement: the 2D argument against materialism. plato.stanford.edu/entries/two-dimensional-semantics/supplement.html
• Internet Encyclopedia of Philosophy, The knowledge argument against physicalism (Jackson’s retraction). iep.utm.edu/know-arg/
• Stanford Encyclopedia of Philosophy, Qualia: the knowledge argument. plato.stanford.edu/archives/spr2020/entries/qualia-knowledge/
• BBC Future (2022). The man rethinking the definition of reality. www.bbc.com/future/article/20220216-the-man-rethinking-the-definition-
• Horgan, J. / Scientific American. A 25-year-old bet about consciousness has finally been settled; and David Chalmers thinks the hard problem is really hard. www.scientificamerican.com/article/a-25-year-old-bet-about-consciousne