Contribution
It separates functionalism from its first formal statement. Putnam’s machine-state theory said psychological states are machine-table states of an automaton; the paper shows that identity fails on several independent counts, and it shows which counts leave broad functionalism standing (all but the qualia argument). It also states, in 1972, the productivity and structure arguments that later carry the representational theory of mind.
Key insight
Ask what a machine table individuates. A machine-table state is fixed by its outputs and successor states, so it is a single, occurrent, unstructured item on a finite list. Psychological states are none of these: some are never occurrent, several co-occur, there are indefinitely many, and they contain one another. So whatever functional story is true of minds, its states are computational states of the system (“has just proved Fermat’s last theorem”), not entries in its table.
Assumptions
- Target (§II). FSIT: every organism that satisfies psychological predicates has a unique best description as a probabilistic automaton, whose inputs are sensory transducers and outputs motor instructions, and each psychological state is one of its machine-table states (p. 165).
- Type identity means the biconditionals “O is in psychological state type X at t iff O is in machine-table state type Y at t” (p. 165). The paper sidesteps the identity conditions of abstract objects.
- Physicalism here is type physicalism (n. 2).
- Empirical premises. Lashleyan equipotentiality, convergent evolution of psychology, and the buildability of psychologically similar machines (pp. 160–161); the productivity of human belief is “an empirical claim, albeit an overwhelmingly plausible one” (p. 176).
- Methodological conservatism. One should resist models requiring changes “not independently motivated” (p. 170), used to keep the everyday dispositional/occurrent distinction.
Key results
All are informal arguments.
- Behaviourism (pp. 160, 162–164). Behaviour depends on the interaction of beliefs, desires, inputs and memories, so the one-to-one pairing fails; the disjunctive (many-to-one) rescue fails three ways: no distinct disjunctions, no lawlikeness (Davidson), and contradiction when one behaviour expresses different states at different times.
- Physicalism (pp. 160–161). Equipotentiality, convergent evolution of psychology, and machines that are psychologically like an organism but physiologically unlike it.
- FSIT is not trivial (pp. 165–166). Every organism is simulable by some Turing machine and realises the “null” Turing machine (Putnam, in conversation); FSIT needs a unique best description whose states correspond to psychological states.
- Argument 1 (pp. 168–170). Every machine-table state is a possible occurrent state; “believes that P” has no occurrent “is believing that P”. Psychological states may correspond to “some abstract property of the organization of machine tables”, but that is “very much weaker”.
- Argument 2 (pp. 170–172). An automaton is in one state at a time, so FSIT models sequential but not simultaneous interaction; organisms need “a set of intercommunicating automata operating in parallel”. An equivalent single automaton exists, but it and the parallel set cannot both be best descriptions.
- Argument 3 (pp. 172–174). Inverted qualia: verificationist counterarguments “are not persuasive”. Two replies are open: nomological impossibility of functionally identical, qualitatively distinct states; or ruling qualia irrelevant to type identity. The second risks saying “an organism might be in pain even though it is feeling nothing at all”. This argument alone threatens functionalism in the broad sense.
- Argument 4 (pp. 174–175). “Damn” versus “darn” makes your pain type-distinct from mine; with paths between all states, “any two automata which have less than all their states in common will have none of their states in common”.
- Argument 5 (pp. 175–177). Machine-table states form a list; a person’s possible psychological states are productive, specifiable “at best … by finite axiomatization”. They may still correspond to computational states, for which countability suffices.
- Argument 6 (pp. 177–179). Believing P is “somehow” a constituent of believing P & Q (n. 8: “Very much ‘somehow’”); lists have no notion of constituency, and productivity and structure are “two aspects of the same phenomenon”.
- §IV (pp. 179–181). “The basic condition upon type identity in science is that it makes possible the articulation of the domain of laws.” Behaviourism and physicalism type states by a subset of the relevant laws; FSIT adds the machine-table constraint, which “cannot be satisfied”.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Behaviourism and type physicalism are probably false on empirical, not conceptual, grounds | moderate | §I, three empirical considerations from Putnam; no data presented |
| C2 | Many-to-one (disjunctive) correlations do not rescue them | moderate | three arguments, pp. 162–164 |
| C3 | Psychological states are not one-to-one with machine-table states | strong as an argument against FSIT’s stated form | arguments 1, 2, 4, 5, 6, each with a stated escape that abandons FSIT |
| C4 | Inverted or absent qualia are a prima facie objection to any functional definition | weak–moderate, as stated | p. 172: “only a prima-facie objection”; two replies left open |
| C5 | Psychological states may be computational states of an automaton | moderate | arguments 5 and 6 leave it open; n. 7 |
| C6 | Type identity at any level is fixed by the laws at that level | assertion with an analogy | §IV, the North Pole and charge example |
Concepts
- Machine-table state — a state “specified by columns in its machine table” (p. 177).
- Computational state — “any state of the machine which is characterizable in terms of its inputs, outputs, and/or machine table states” (p. 177).
- Unique best description — the one automaton description whose states correspond to the organism’s psychological states (p. 165).
- Type identity of psychological states — being in states of “exactly the same type” (p. 159).
Connections
- Putnam, “Psychological Predicates” (LIT-396; read only as an excerpt in NOTE-335), is the source of the theory under test, and is quoted on the “normal form” difficulty (p. 175).
- Davidson’s “Mental Events” supplies the lawlikeness objection.
- Block’s “Troubles with Functionalism” (LIT-407, NOTE-366) turns argument 3 into the Absent Qualia Argument and adds the homunculi-headed robot. In NOTE-366 Block’s §1.1 restricts his target to type functionalism and leaves token functionalism alone; this paper already notes that FSIT “is not an ontological theory” and is compatible with token physicalism (pp. 166–167).
- Chalmers (LIT-404, NOTE-342) answers the triviality worry that FSIT’s “unique best description” clause is meant to block, with counterfactual conditions on implementation. Argument 2’s parallel automata are the shape of his combinatorial-state automata. That pairing is mine.
Bearing on the record
- THEORY-043 takes organisation to be what makes a mind. This paper says which organisation cannot be meant: not a machine table, whose states are too few, too fine and unstructured. That narrows THEORY-043’s premise to computational or Psychofunctional organisation, the level LIT-407 and NOTE-342 already discuss.
- It carries no instruction for machine-learning practice.
Limitations
- The arguments are aimed at one formulation; each comes with an escape that keeps functionalism (abstract properties of machine tables, computational states). They show FSIT’s machine-table clause fails, not that a functional theory of mind fails.
- The qualia argument is explicitly prima facie and not developed.
- Argument 1 rests on English grammar (“is believing” is not said), backed by a conservatism principle; a theory might posit occurrent belief states.
- The empirical premises are stated, not evidenced.
Open questions
- Which “abstract property” of an automaton could a disposition such as belief correspond to (argument 1)? The paper names the option and leaves it.
- Is argument 4’s fineness objection answered by Psychofunctional typing at a level of abstraction, as LIT-407 later suggests, or does it recur for any functional individuation?