NOTE-366: Block — Troubles with Functionalism

Contribution

The paper turns the functionalist’s own argument against physicalism back on functionalism. Physicalism is chauvinist because no physical property is shared by all realisations of a machine table. Functionalism is liberal because it “tacks down” mental states only at the periphery, so anything with the right input–output-and-state structure counts. It supplies the standard counterexample: the homunculi-headed robot and its nation-of-China variant. It also supplies the standard critique of Putnam’s stipulation that a pain-feeler has no parts with pain-feeler organisation, and the first statement of the liberalism/chauvinism dilemma for any specification of inputs and outputs.

Key insight

Ask what a functional description pins down: only the inputs and outputs, with everything inside quantified away. Then any arrangement of anything that preserves the transitions, including a nation of people with radios, satisfies the description. Whether that is a reductio depends on what you think grounds the intuition. Block’s answer is that the intuition is not anti-nesting prejudice. If you are made, below the level of your synapses, of tiny people, you lose nothing. If a system built to mimic you shares neither your psychology nor your neurophysiology, you have no reason to credit it with your qualia. What matters “is how the mentality of the parts contributes to the functioning of the whole” (p. 292), and in the end not the parts’ mentality at all (n. 19).

Assumptions

• Target (§1.1). Functional-state identity theses about mental state types, restricted to “narrow” states (p. 277). Token functionalism is not attacked (p. 288).
• Two functionalisms (pp. 268–275). Functionalism identifies a mental state with its Ramsey functional correlate relative to commonsense psychology. Psychofunctionalism uses scientific psychology, and may specify inputs and outputs as neural signals. Psychofunctional equivalence entails Functional equivalence, which entails behavioural equivalence (p. 275).
• Functional equivalence is relative to an input–output specification and a level of abstraction. Block concedes it is vague, and says physicalism is vague in the same way (pp. 271–273).
• Possibility (§1.3). The China system is taken to be nomologically possible for an hour. If it is not, metaphysical possibility suffices, given that true identities of rigid designators are necessary (Kripke). Block argues the Absent Qualia Argument does not inherit Kripke’s conceivability problems, because it asks you to imagine one thing (a homunculus-head) and then notes a doubt about it (pp. 285–289).
• Time scale is irrelevant (pp. 280–281): slow minds are conceivable, and to insist on time scale “seems crudely behavioristic”.
• Empirical premise (§1.6). Mentality depends “crucially on psychological and/or neurophysiological processes and structures” (p. 295). The argument is therefore “partly empirical” (p. 293).
• Neuron count. “There are only about a billion neurons in the brain” (p. 278). This sizes the China system to China’s population. It is low by about two orders of magnitude; NOTE-131 records Schwitzgebel’s figure of about 10

Key results

All are informal arguments: thought experiments and burden-of-proof arguments. There are no theorems and no data, except the cited McCulloch–Pitts replacement result (n. 19, citing Minsky 1967 p. 45).

• Homunculi-headed robots (§1.2, pp. 277–285).
  • The little-men robot: lights wired to sense neurons, buttons to motor neurons, a state card on a bulletin board. Each “G-man” executes one quadruple of a machine table that describes you.
  • The nation of China: a billion people with two-way radios, satellites displaying the state, run “for an hour”. “Surely such a system is not physically impossible.”
  • Objections answered. It would malfunction under floods: the describer fixes what counts as inputs, as with a computer. It would be too slow: time scale is irrelevant.
  • The same construction defeats computational-state functionalism and non-machine (Ramsey) functionalism. “The more intelligence exercised by the little men … the less inclined we are to ascribe to the simulation the mental properties they are simulating” (p. 285).
• “What if I turned out to have little men in my head?” (§1.4). Either privileged access tells me I have qualia whatever is in my head or it does not. On either answer the doubt about homunculus-heads stands; to argue otherwise “beg[s] the question against the doubter” (p. 290).
• Putnam’s proposal (§1.5, pp. 290–293; n. 16).
  • Putnam’s condition, quoted from Putnam 1967 pp. 434–439 (the 1975 reprint), is “the ad hoc device of stipulating them away”. Its stated purpose is to rule out “swarms of bees as single pain feelers”.
  • The all-parts reading, that no pain-feeler has all its parts sentient-organised, does not exclude the robot, which has a mechanical body.
  • The no-sentient-proper-parts reading excludes “pregnant women and people with sentient parasites”.
  • The “crucial role” reading excludes parts that both have the organisation and play a crucial role in giving the whole its organisation. It fails on the elementary-particle people. Tiny intelligent beings in spaceships mimic our elementary particles, and a human colonist comes over a few years to be composed of their “matter”. “Would you be any less capable of feeling pain … ? I think not”: the synapses’ electrochemistry is unaffected (pp. 291–292).
  • Note 16 adds that reformulating Putnam to keep the particle people but exclude the robot would be “a most ad hoc maneuver”. Split-brain and hemispherectomy evidence suggests each hemisphere may have a sentient being’s functional organisation, so a Putnamian condition might rule out all of us. Individual neurons might share an insect’s organisation, if inputs are neural impulses.
• The diagnosis (pp. 292–296; nn. 17, 19).
  • A first guess is that the robot has “too much internal mental structure”. The particle-people case shows this is wrong.
  • The difference is that the particle change makes no difference to psychological or neurological processing, whereas the robot need share neither with us.
  • Two empirical facts back the doubt. We know brain-headed systems have qualia. Homunculus-heads were designed to mimic us, so the best explanation of their winces is the design, not pain (pp. 293–294).
  • The doubt does not depend on the homunculi’s minds. Each can be replaced by a McCulloch–Pitts “and” neuron (n. 19).
  • The intermediate case of a billion little men each simulating one of your neurons is one where “intuition seems to founder” (n. 17).
• Behaviour is not enough (pp. 294–295). A finite list-searcher over all “smart speakable strings” would pass an hour-long Turing test and “clearly has no mental states at all”.
• Against Functionalism specifically (pp. 296–300). Paralytics and brains in vats break the platitudes. The Problem of Differentiation: distinct sensations can share all platitudes. The Problem of Truth: platitudes are often false.
• Psychofunctionalism (§2).
  • The intuition that a homunculi-headed Psychofunctional simulation lacks mentality should be disregarded for states in psychology’s domain. It “could have nonqualitative mental states” (p. 301), for example remembering that P by storing a sentence-like object (p. 306).
  • Qualia may not be in psychology’s domain. Psychophysics would give the same results on a system with inverted or absent qualia (pp. 306–307).
  • The two arguments for Psychofunctionalism fail. “It is the business of psychology” would make protons and antiprotons identical under a Ramsey-sentence physics (pp. 301–303). Inference to the best explanation needs a minimally adequate candidate (pp. 303–304).
  • Pain may be a composite of a quale and a functional state, which would explain why theories of pain split (pp. 309–310).
• Chauvinism and liberalism (§3).
  • Human Psychofunctionalism is chauvinist about Martians who are Functionally but not Psychofunctionally like us. Cross-world psychology might rescue it, but nobody can now say whether it is possible (pp. 310–314).
  • Inputs and outputs described only as inputs and outputs are “wildly liberal”. A sheik could make Bolivia’s economy functionally equivalent to himself, and “if there are any fixed points … one of them is that the economy of Bolivia could not have mental states” (p. 315).
  • Physical descriptions of inputs and outputs are chauvinist. If brains can be input and output devices (EEG Morse code), multiple realisability applies to the periphery too (pp. 316–317).
  • Conclusion: a burden-of-proof argument, not a conclusive one (p. 318).

Claims

id	claim	strength	support
C1	Every version of functionalism is subject to a homunculi-headed prima facie counterexample	moderate as a prima facie claim	construction §1.2; generalised to computational and Ramsey versions pp. 282–285; “so close to the core … I would be tempted to regard a doctrine not subject to it as ipso facto not a version of functionalism” (p. 282) is a stipulation
C2	The nation-of-China system is nomologically (or at least metaphysically) possible and functionally equivalent to a person for an hour	moderate	physical-possibility assertion p. 279; slowness and malfunction objections answered pp. 279–281
C3	The intuition that a Functional homunculi-head lacks qualia has a rational basis	moderate–weak	the empirical premise that mentality depends on psychology or neurophysiology (p. 295), and the design argument against inference to the best explanation (p. 294); “not an overwhelmingly powerful argument” (p. 296)
C4	Putnam’s anti-nesting condition is ad hoc, and every reformulation tried fails	moderate	three readings refuted by cases §1.5 (pregnancy, the mechanical body, elementary-particle people); n. 16 hemispheres and neurons; no general proof that no reformulation works (“likely to remain elusive”)
C5	The doubt about homunculi-heads does not rest on the homunculi’s mentality	moderate	the particle-people contrast (p. 292); the McCulloch–Pitts replacement (n. 19), which is a cited theorem for finite automata with one input signal at a time
C6	A homunculi-headed Psychofunctional simulation could have beliefs and other nonqualitative states	moderate	argument from psychology’s domain §2.1 (p. 301)
C7	Qualia may not be in the domain of psychology	weak–moderate	psychophysics is blind to inverted or absent qualia (p. 307); “on the basis of what we have to go on” (p. 307); explicitly tentative
C8	No input–output specification avoids both liberalism and chauvinism without mentalistic terms	moderate as burden of proof	Bolivia (p. 315); species-specific peripheries; brain-as-periphery (pp. 316–317); “I do not claim that this is a conclusive argument” (p. 318)
C9	The economy of Bolivia could not have mental states	assertion	offered as a “fixed point” (p. 315), not argued

Concepts

• Liberalism / chauvinism. A theory is liberal insofar as it falsely attributes mental properties, and chauvinist insofar as it falsely denies them (p. 311).
• Functionalism vs Psychofunctionalism. Ramsey functional correlates relative to commonsense versus scientific psychology (p. 269).
• Ramsey functional correlate. A property-abstract over the Ramsey sentence of a psychological theory, with mental terms replaced by bound variables (pp. 269–271, n. 10).
• Homunculi-headed robot / nation of China. A realisation of a person’s machine table whose state transitions are carried out by persons (pp. 277–279).
• Absent Qualia Argument. If Q = S_Q and something can be in S_Q with its Q in doubt, the identity is in doubt (p. 281).
• Elementary-particle people. Persons composing your matter below the level of any psychological or neural difference (pp. 291–292).
• Weak / strong equivalence. Same outputs for all inputs, versus same mediating processes in a branch of science (p. 274, after Fodor).

Composition and nesting. Block is explicit that the robot’s parts are minded and that nesting is not what grounds the doubt. “A natural first guess … is that they have too much internal mental structure … the example of the elementary-particle people … suggests this first guess is wrong” (p. 292). So the paper separates three questions: whether a whole can have minded parts (yes, at the particle level), whether parts must lack the whole’s organisation (Putnam’s condition, rejected), and whether a functional simulation has qualia (doubtful, for reasons other than nesting).

Connections

• Putnam (LIT-396, read from an excerpt in NOTE-335). Block’s quotation of condition (3) agrees verbatim with the 1975 reprint, as far as the quotation runs. He reads “(if they count as such)” where the excerpt has “(if they can count as such)”. He cites Putnam 1967 under the reprint title “The nature of mental states”, noting it was “originally published under the title Psychological Predicates”.
• Schwitzgebel (LIT-159, NOTE-131). It is the source of Schwitzgebel’s C4 counterexample and of the China brain he enlists for neurochauvinism. Schwitzgebel’s paraphrase of the counterexample (conscious organisms “each choosing to play the role of one neuron”) moves it from Block’s particle level to the neuron level, where Block reports that intuition founders (n. 17). Block’s design argument, that the best explanation of the robot’s winces is that it was built to mimic us, has no purchase on the US, which was not built to mimic anything. Block’s Bolivia “fixed point” is the kind of commonsense datum Schwitzgebel discounts.
• Lycan (LIT-419). Lycan reads earlier drafts (n. 25). Lycan 1974 is answered on the one-state-at-a-time problem (p. 282). Lycan 1981 is the reply Schwitzgebel points to, but it is unread here.
• Selfridge (LIT-427) and Minsky (LIT-399, LIT-429). These are the decomposition side’s demons and agents. Block cites Minsky only for computation theory (Minsky 1967, pp. 45, 206). n. 19 is the bridge: a homunculus that only executes one table entry is replaceable by a mindless gate, which is the society-of-mind picture of an agent.
• IIT (LIT-185). Not cited, since the paper predates it. The particle-people case is the template for later reductios of exclusion postulates.
• Nagel (LIT-096). Cited for “what it is like” (p. 281), the notion the Absent Qualia Argument uses.

Bearing on the record

• For the bridge, the paper is the place where three questions that Schwitzgebel’s §2 treats together come apart. Can a conscious whole have minded parts? Yes, per Block’s particle case. Is Putnam’s condition motivated? No. Does functional organisation realised by people bring qualia with it? Doubtful, but the ground is mimicry and dissimilarity, not nesting. A THEORY drawing the bridge should cite Block for the first two and keep the third separate.
• Agency and cognition from parts: Block grants a Psychofunctional homunculi-head nonqualitative states (C6). On the question whether agency built from parts brings experience with it, Block’s answer is “it brings belief, and maybe not experience”. That is the distinction the record’s group-agency works (List and Pettit, Huebner) need, and it was drawn here first.
• No ML instruction. It does not belong in the anthology.

Limitations

• Almost every step rests on intuitions about thought experiments, as Block says (burden-of-proof framing, pp. 296, 318).
• The particle-people case refutes only the “crucial role” version of Putnam’s condition. It leaves open a version keyed to the level at which parts carry psychological processing, which is the level Schwitzgebel’s antheads and the US occupy.
• The empirical premises are dated: about a billion neurons (p. 278), and right-hemisphere language figures from Psychology Today (n. 16).
• The “design” argument (p. 294) applies only to systems built to mimic something. The paper does not discuss naturally arising group systems.

Open questions

• Does any reformulation of Putnam’s condition keyed to the psychological level survive both the particle people and the hemispheres (n. 16)?
• If the doubt about the China system rests on dissimilarity to our psychology (C3), how much similarity in information processing would dissolve it? A Psychofunctional China system dissolves it for beliefs by Block’s own §2.1.
• Is the n. 17 case (a billion people each simulating one neuron) one where intuition founders, or one where it favours consciousness? This is the case nearest to Schwitzgebel’s antheads.

Corrections

• none to a seeded skim (there was no seed)
• Against NOTE-131 C4 and Schwitzgebel §2 (manuscript p. 12). Block’s counterexample to Putnam is not ultra-tiny organisms “each choosing to play the role of one neuron”. It is elementary-particle people, a change “not one that makes any difference to your psychological processing … or neurological processing but only to your microphysics” (p. 292). Block mentions the neuron-level variant only in n. 17, as a case where intuition “seems to founder”. The consequence Schwitzgebel draws, that under an anti-nesting principle you would become nonconscious with unchanged behaviour, is a fair gloss of the particle case, but not of a neuron case Block endorses.
• Block’s neuron count (about a billion, p. 278) is low by about two orders of magnitude.
• Page range. The chapter runs pp. 261–325 (the scan and the Conservancy record agree). A web summary gave 261–326.