Source
- Chalmers (1995), “Absent Qualia, Fading Qualia, Dancing Qualia”, in Metzinger (ed.), Conscious Experience, LIT-392, read in full in NOTE-347 from the author’s text (sections, no page numbers).
- Chalmers (1995), “Facing Up to the Problem of Consciousness”, LIT-406, §7, read in NOTE-336: the same principle as one of two nonbasic psychophysical laws, argued in one page.
What was actually shown
The principle (§1). “Given any system that has conscious experiences, then any system that has the same functional organization at a fine enough grain will have qualitatively identical conscious experiences.” A functional organisation is a set of abstract components, their possible states, and the dependencies among them and the inputs and outputs; “fine enough” means fine enough to fix behavioural dispositions, and the two systems must be in corresponding states.
Fading qualia (§3). If a silicon isomorph lacked experience, then along a one-neuron-at-a-time replacement series experience either vanishes at one step, which would need a brute discontinuity in laws that compound only continuous dependencies, or fades. Halfway, “Joe” says “bright red” while experiencing “tepid pink”, and is “utterly out of touch with his conscious experience”. Everyday fading, in drowsiness, comes with a matching change in function; Joe’s does not. Searle’s variant, with true but impotent new beliefs, is rejected: an unchanged organisation has “no room” for them.
Dancing qualia (§4). If an isomorph had inverted qualia, some two systems ten percent apart along the series would differ noticeably. Install the second’s circuit beside the first’s with a switch. Flipping it changes experience and no function, so the subject is “not suddenly disposed to say ‘Hmm! Something strange is going on!’”. The same construction serves against absent qualia.
Populations (§3). The series runs equally with neurons replaced by tiny homunculi, or by normal-sized ones linked by radio, “ending up with a network of homunculi that is essentially equivalent to the population controlling a robot”, and on to one homunculus working from written records. The system has the experiences; the homunculus “is merely acting as a sort of causal facilitator”.
The modality (§5). Fading and dancing qualia are intelligible, and the premise that some brains are conscious is empirical, so the conclusion is natural, not logical, supervenience: “nonreductive functionalism”, “just as compatible with certain forms of property dualism … as with certain forms of physicalism”.
What could have come out otherwise. The conditionals (C2, C4 in NOTE-347) could not, given the series. What carries the weight is the rejection of their consequents, and that is a judgment, not a result: “these arguments do not constitute conclusive proof” (§2).
What this does not say
- That any actual AI system, animal or group is conscious. The principle quantifies over fine-grained isomorphs of something already conscious. The paper leaves open “just what sort of organization gives rise to experience” (§5). No current AI system is an isomorph of a human, and the United States is an isomorph of nothing, so the principle does not license Schwitzgebel’s step to it by similarity of kind (NOTE-347, Bearing). Schwitzgebel’s Leo case (LIT-459) applies the invariance style of argument to a group without an isomorph, but only against Kammerer’s principle on a loose reading.
- That experience is functional organisation. The supervenience is natural, not metaphysical, and the view is offered as compatible with property dualism. It sits inside the naturalistic dualism of THEORY-694, not against it.
- That coarser realisations are covered. The series is built neuron by neuron, and “the arguments generalize” is asserted. Block’s main China system, a machine table with a line per person, is reached only by that assertion; his note-17 case, a billion people each simulating a neuron, is reached directly.
- That the loopholes are closed. Speed (a much faster or slower isomorph cannot be swapped in), history, very mild inversions and unattended qualia are conceded as coherent but called unmotivated (§4).
- That self-knowledge is reliable outside the systems it was observed in. An opponent can hold that it is, which is the question at issue (NOTE-347, Limitations).
Connections
- Anti-nesting (THEORY-043, QUESTION-046). If invariance holds, a whole whose parts are minded people can have experiences of its own while the parts keep theirs, so Putnam’s stipulation and IIT’s exclusion, where it would deny the whole, fail for isomorphs. Kammerer’s functional principle survives, since a homunculus in a neuron’s role is one a chip could fill. The dispute over a modular mind and a nation becomes wholly architectural, which invariance cannot decide because neither is an isomorph of the other or of us. That is the partial answer to QUESTION-046: a constraint on any criterion, not a criterion.
- IIT holds that a functional equivalent can differ in Φ and lack experience (LIT-445, NOTE-365, C7), which denies this principle for isomorphs. The two cannot both be true.
- Opposite conclusions from one premise. Dennett’s “Quining Qualia” (THEORY-415) also turns on a change in experience the subject could not notice, and concludes that qualia separable from reactions are incoherent. Chalmers concludes they are fixed by organisation.
- THEORY-252 (Gidon, Aru & Larkum) argues that on computational functionalism an idle clamp that changes no actual activity can change consciousness. It reads itself as the converse of fading qualia: this argument holds organisation fixed and changes the substrate, the clamp holds substrate and activity fixed and removes unused organisation. THEORY-095 (Block’s Inverted Earth) denies that long-arm functional role fixes phenomenal character, and says it must not be cited against this principle, which is about internal organisation.
- THEORY-023: the principle is the conditional the functionalist side of the AI dispute assumes, and it reaches no system that is not an isomorph.