NOTE-354: Chalmers — The Meta-Problem of Consciousness

Contribution

The paper isolates a problem every side of the consciousness debate has to solve: explaining our problem reports and intuitions without mentioning consciousness. It argues that this problem is empirically tractable. It organises a scattered literature of thirteen proposals under one problem, with one recurring test, “the problem of belief”. It then uses the problem in two ways. Against realist theories, it poses a challenge: the basis of consciousness must explain the reports. Against illusionism, it argues that only the strong form dissolves the hard problem, and that the strong form is false.

Key insight

Problem reports are behaviour, so the meta-problem is an easy problem in the 1995 sense (LIT-406). But it is the easy problem nearest to the hard one. If consciousness is real, whatever is its basis should also be what makes us say it is puzzling. If the reports can be explained completely without it, then either consciousness is an illusion, or its connection to our talk about it looks like luck. Every theory has to say which.

Assumptions

• Physical explicability of behaviour (intro). “At least if we accept that all human behaviour can be explained in physical and functional terms”, problem reports can be. §1.2 widens this to topic-neutral explanations (representational, rational, historical, structural), so that interactionists and panpsychists are covered.
• Wide sharing (§1.1). Problem intuitions are taken to be widely shared, or at least to have a widely shared basis. This is called an “empirically defeasible assumption”. The evidence cited is thin: work on intuitive dualism (Bloom; Chudek et al.), Gottlieb and Lombrozo on what is hard to explain, and the PhilPapers Survey (fn 8: 56% physicalism, 27% non-physicalism; 36% say zombies are conceivable but not metaphysically possible).
• Explananda are dispositions (§1.1). The targets are dispositions to report and judge. They are reconstrued topic-neutrally as “quasi-phenomenal judgments”.
• A design stance (§2). Mechanisms are assessed by asking what a well-designed self-monitoring robot would need.

Key results

All are conceptual arguments and surveys. No data are produced and no model is built; one built model by others is reported (fn 26 and p. 34).

• Taxonomy of problem intuitions (§1.1). Explanatory intuitions (gap, anti-functionalist) and metaphysical ones (dualist, fundamentality) are most central. Knowledge intuitions (first-person knowledge, third-person ignorance) and modal ones (zombies, inversion) complete the central four. There are also value, distribution (“robots are not conscious”), self, quality and presentation intuitions. Footnote 7 argues that Sytsma and Machery’s robot-sees-red versus robot-feels-pain finding was predicted by The Conscious Mind’s dual readings of mental terms.

• Candidate solutions (§2). Each is tested against the belief case. Beliefs are introspected, opaque and immediately known too, and they raise no hard problem.

  1. Introspective models: necessary, not sufficient.
  2. Phenomenal concepts: indexical and recognitional accounts are “too thin”.
  3. Independent roles (Nagel; Hill and McLaughlin): belief concepts are independent too.
  4. Introspective opacity (Armstrong’s headless woman): the move from “not seen as physical” to “seen as non-physical” is unexplained.
  5. Immediate knowledge, sharpened to acquaintance, meaning presentation and revelation.
  6. Primitive quality attribution (Pereboom; Hall’s dummy properties): works for qualia-style views. It must explain why illusionism about experience meets more resistance than illusionism about colour (Kammerer’s problem, renamed “the resistance problem”).
  7. Primitive relation attribution, the attention-schema idea generalised: “consciousness is a model of representation”.
  8. Introjection (Place’s phenomenological fallacy; Jackson; Rosenthal): the hard problem survives rejecting sense-data.
  9. The user illusion (Dennett): a slogan collecting 4, 6 and 8.
  10. Use–mention fallacy (Loar, Tye, Papineau): uncharitable, and it over-generates.
  11. Underestimating the physical (Stoljar, Strawson): does not touch the anti-functionalist intuitions.
  12. Historical and cultural: evolutionary design explanations are favoured.
  13. Others: Humphrey, Fiala et al., Molyneux, Drescher. These are dismissed “for fairly predictable reasons” that are not given.

• Summary picture (p. 34). Opaque introspective concepts, independent of physical ones, attribute primitive mental relations to primitive perceptual qualities, with a sense of immediate knowledge and acquaintance. It is “neutral between realism and illusionism”. The only computational test known is Muehlhauser and Shlegeris’s theorem-prover agent.

• The meta-problem challenge (§3). “If a theory says that mechanism M is the basis of consciousness, then it needs to explain how mechanism M plays a central role in bringing about judgments about consciousness.” It descends from the 1990 “coherence test” (fn 28).

  • IIT: “for every system with high integrated information there will be a computationally isomorphic simulated system with zero integrated information” that reports alike. So “high integrated information plays no essential role in explaining phenomenal intuitions”. Not a “knockdown”.
  • Biological theories (Block 2009, LIT-632): non-conscious silicon duplicates share the intuitions.
  • Global workspace and first-order representationalism: they can begin to answer, since only globally available information is reported, but problem reports are unexplained.
  • Higher-order thought: Goldman’s regress.
  • Quantum collapse: unselected branches also report.
  • Panpsychism: the combination problem gains a reporting aspect.
  • Analytic functionalism: fails outright, because by Nagel’s point it makes the hard problem as unpuzzling as the easy ones.
  • The author’s lean: acquaintance, developed as a first-order property F that disposes states to be objects of primitive higher-order representation. Footnote 31 notes that his 1996 candidate, any informational state, faces the belief objection.

• Six reactions (§4). Non-reductive: no solution (“meta-problem nihilism”), correlationism, realizationism. Reductive, each with some illusionism: strong illusionism, consciousness as a lower-order meta-problem state, consciousness as a higher-order one. Realizationism and strong illusionism are judged the important two.

• Debunking (§5). Argument A: there is a consciousness-independent explanation of our beliefs, so they are unjustified. Three replies:

  • meta-problem solutions explain intuitions, but phenomenal beliefs are partly constituted by consciousness (the 2003 paper);
  • justification is by acquaintance, not by causal history;
  • “independent” equivocates: descriptive or modal independence makes premise 1 true, but premise 2 needs causal and constitutive independence. Compare explanations of table beliefs that do not mention tables.

  Argument B, the coincidence argument, survives better. With strong modal independence “it can seem lucky” that the meta-problem processes and consciousness coincide. Mørch’s phenomenal-powers view is offered as a way out.

• Illusionism (§§6–7). Weak illusionism (consciousness exists but is not intrinsic, non-physical, primitive, and so on) leaves the hard problem intact, because it needs only “something it is like”. A functionalised “phenomenal consciousness” is in effect strong illusionism about the ordinary concept. The lower-order version is type-B materialism. The distinctive illusionist view is type-A. Strong illusionism is coherent, since an illusion can be a non-experiential judgment, as in the “grand illusion” of visual detail. But it is false: “People sometimes feel pain.” A sample dialogue ends with the concession that “dialectically, the illusionist side is much more interesting”.

• Conclusion (§8). Neither side has “a truly satisfactory response to the charge of absurdity”. The illusionist needs to explain how a mind without consciousness could be like this. The realist needs to show why the meta-problem processes are “by their nature grounded in consciousness”.

Claims

id	claim	strength	support
C1	The meta-problem is an easy problem, solvable in topic-neutral terms, and amenable to experiment and computational modelling	moderate	intro, §1.2: follows from the physical explicability of behaviour; widened to topic-neutral explanation
C2	Problem intuitions are widely shared across people and cultures	weak; flagged by the author as empirically defeasible	§1.1, fn 8, fn 10: little direct evidence on core intuitions
C3	Introspective models, phenomenal concepts, independent roles, opacity and immediate knowledge each fail alone to explain what is distinctive about phenomenal states, because they apply to beliefs too	moderate	§2: the belief test applied case by case
C4	The most promising account combines opaque introspective models, primitive quality and relation attribution, and a sense of acquaintance	weak–moderate; “closer to a fleshed-out explanandum”	§2 summary, p. 34
C5	A theory’s basis of consciousness should play a central role in producing judgments about consciousness, and IIT and biological theories have a prima facie problem meeting this	moderate as a challenge; “not a knockdown argument”	§3: the zero-Φ isomorph and the silicon duplicate
C6	Debunking arguments for illusionism can be blocked by distinguishing descriptive or modal independence from causal and constitutive independence	moderate	§5: the table analogy; the constitutive role of consciousness in phenomenal belief
C7	A residual sense of coincidence between consciousness and the meta-problem processes remains on all realist views	the author’s concession	§5
C8	Only strong illusionism can dissolve the hard problem	moderate	§6: the hard problem needs only “something it is like”
C9	Strong illusionism is coherent but false, because people feel pain	assertion, by a Moorean argument	§7; the author grants the illusionist predicts the realist’s conviction
C10	Simple systems can mimic some phenomenal reports, so puzzlement reports are weak evidence of machine consciousness	weak–moderate; one reported system	fn 26

Concepts

• Meta-problem — explain why we think there is a problem of consciousness; strictly, explain problem intuitions in topic-neutral terms, or explain why that is impossible (§1).
• Problem intuitions / problem reports — dispositions to judge and say that consciousness is puzzling (§1.1).
• Topic-neutral explanation — an explanation that does not mention consciousness or its cognates (§1.2).
• Quasi-phenomenal judgments — problem judgments reconstrued up to, but not including, their being about consciousness (§1.2).
• The problem of belief — the recurring test that a mechanism must not equally generate problem intuitions about beliefs (§2).
• Resistance problem — Kammerer’s “illusion meta-problem”, renamed: why illusionism meets so much intuitive resistance (§2).
• Meta-problem challenge — the basis of consciousness must explain our judgments about it (§3).
• Meta-problem processes; lower-order and higher-order meta-problem states — what the topic-neutral explanation invokes, split into the first-order states modelled and the introspective states that do the modelling (§4).
• Realizationism / correlationism — consciousness realises, or merely accompanies, the meta-problem processes (§4).
• Strong / weak illusionism — consciousness does not exist / exists but lacks properties it seems to have (§6).

Composition and nesting. The paper does not discuss groups or wholes with minded parts. It touches composition only through panpsychism. The meta-problem challenge adds a reporting aspect to the combination problem: explain “how these combination states play a central role in bringing about reports of consciousness” (§3).

Connections

• The 1995 paper (LIT-406) supplies the easy/hard distinction the whole paper depends on. The Conscious Mind (LIT-441) is where the author’s first solution appeared (fn 5: pp. 184–8, 289–92) and where he wrote that if he were a materialist he would be an illusionist (p. 189).
• Frankish (LIT-103) named the view “illusionism” (Frankish 2016), and the “illusion problem” is his term, though Chalmers says he suggested the name (fn 6). §6’s argument agrees with Frankish that dissolving illusionists must be strong illusionists, for different reasons.
• IIT (LIT-445, LIT-185). The §3 challenge rests on IIT’s own claim that a zero-Φ computational isomorph is possible (NOTE-365, C7).
• Organizational invariance (LIT-392). Not cited. But the §3 argument against IIT and biological theories has the same shape as the 1995 dancing-qualia argument: a functional duplicate makes the same judgments. Here the conclusion drawn is only that M plays “no essential role” in the judgments, not that the duplicate is conscious.
• Hofstadter (LIT-423, unread), cited in fn 13 for self-model illusionism about the self.
• Kammerer is cited for the resistance problem and as a possible illusionist (Kammerer 2016, 2018a). The record’s Kammerer is the 2015 reply to Schwitzgebel (LIT-409), a different paper.
• Seth (LIT-135). His real-problem approach is among the strategies that try to solve easy problems and expect the hard one to dissolve. The paper does not cite him.

Bearing on the record

• THEORY-023. C10 (fn 26) is a direct, if small, support for the account’s mimicry point. Even puzzled phenomenal reports, the behaviour most often proposed as hard to fake, can be produced by a trivially simple agent. C5 adds a second consideration: indicator properties taken from theories whose basis does not explain the reports (IIT, biological theories) carry that weakness into any indicator list (LIT-056). The second point is my application.
• THEORY-043. (1) C5 gives a reason to doubt IIT as the basis of consciousness that is independent of IIT’s verdict on groups. That is relevant to the account’s promote_when, which asks for a criterion motivated independently of the verdict it gives. It weakens IIT as the architectural criterion, rather than supplying one. (2) The challenge suggests a test nobody in the bridge has applied: does a candidate group mind produce problem reports of its own, over and above its members’ reports? My inference; the paper is silent on groups.
• LIT-097. No bearing.
• No ML instruction. Its proposed computational models of phenomenal reports are a research programme for cognitive science, not a practice instruction. It does not belong in the anthology.

Limitations

• The data the programme needs mostly do not exist yet, by the author’s own account (§1.1).
• The favoured solution is an explanandum sketch, not a mechanism (p. 34).
• Several rejections are brief: proposal 13 is dismissed without reasons.
• The anti-illusionist conclusion rests on a Moorean premise that the illusionist explicitly predicts the realist will find compelling. The paper concedes the dialectic does not settle it.
• The IIT challenge assumes that a zero-Φ simulation has the same intuitions, on a functional reading of intuition. An IIT theorist who denies that the simulation judges anything would resist it. The paper notes that “various ways” of responding exist and does not pursue them.

Open questions

• What experimental results would discriminate between primitive-quality and primitive-relation accounts, or between realizationism and strong illusionism?
• Does any group, AI system or modular architecture produce problem reports that are its own? And would that count, by the §3 challenge, as evidence bearing on its consciousness?
• Can realizationism be made to remove the coincidence (C7)? The paper points to Mørch’s phenomenal powers as one route and leaves it there.

Corrections

• None to a seeded skim (there was no seed). The brief’s citation is correct.