Source
- Block (1981), Psychologism and Behaviorism, LIT-642, read in full in NOTE-506: pp. 5–6 (the doctrine), 14–21 (dispositions, capacities and the machine), 25–43 (the objections), nn. 13, 17 and 21. Page numbers are the journal’s.
- Block (1995), The Mind as the Software of the Brain, LIT-630, read in full in NOTE-490: §11.1.1 (pp. 381–384), the Aunt Bubbles machine and its n. 1.
What was actually shown
The doctrine. Block (1981), LIT-642, defines psychologism as “the doctrine that whether behavior is intelligent behavior depends on the character of the internal information processing that produces it”. Two systems “could be exactly alike in their actual and potential behavior, and in their behavioral dispositions and capacities and counterfactual behavioral properties”, and still differ so that “one is not at all intelligent while the other is fully intelligent” (p. 5).
The target is made as strong as it can be. Block grants the behaviourist a non-question-begging definition of “sensible”, “an enormous concession” (p. 11). He then moves from a disposition to a capacity, the neo-Turing Test conception: intelligence is “the capacity to produce a sensible sequence of verbal responses to a sequence of verbal stimuli, whatever they may be” (p. 18). Stated with internal states in the antecedents of its conditionals, it escapes the Chisholm–Geach, perfect actor and paralytic objections. Those objections, he argues, only ever told against behaviourist necessary conditions, so the sufficient condition had survived them (pp. 14–19).
The counterexample. Programmers list every typable string of an hour’s conversation in which one party is sensible. The machine finds the strings that begin with the conversation so far and types the next sentence; a tree-searching variant stores one reply per branch. It has the capacity the conception requires, “But actually, the machine has the intelligence of a toaster. All the intelligence it exhibits is that of its programmers” (p. 21). The length of the test is not essential: any finite length can be met the same way. Block concludes that “the capacity to emit sensible responses is not sufficient for intelligence”, that the neo-Turing conception is refuted “along with the older and cruder Turing Test conceptions”, and that “whether behavior is intelligent behavior is in part a matter of how it is produced” (p. 21). Note 21 says what disqualifies it: not that its replies are echoes, but that its processing “is of the most elementary sort”, which would hold as well for a machine produced by cosmic accident. The argument needs only logical possibility, because behaviourist analyses are conceptual (Objection 6, pp. 30–33).
Not chauvinism. Psychologism “requires only that intelligent behavior not be the product of a (at least one) certain kind of internal processing” (pp. 6, 21). It does not require our kind: a Martian with rich but alien processing still counts.
Wider than intelligence. Block (1995), LIT-630, restates the machine as the “Aunt Bubbles machine”, which “has the intelligence of a jukebox” (p. 383). Its n. 1 draws the wider conclusion: the machine “refutes something stronger than behaviorism, namely the claim that the mental ‘supervenes’ on the behavioral; that is, that there can be no mental difference without a behavioral difference”, a doctrine Block notes is held even by those who reject behaviourism. The chapter also names the point the 1981 paper’s capacity move turned on: the Turing test “focuses on performance rather than on competence”, and competence, idealised performance, needs a distinction among components of the mind that no behaviourist can draw (p. 384). The list-searcher was first sketched in “Troubles with Functionalism” (LIT-407, pp. 294–295), where it “clearly has no mental states at all”.
What could have come out otherwise: the machine could have been incoherent as specified, or the standard objections could have reached the sufficient condition. Block argues neither holds; the first is answered by tree-search and logical possibility, the second case by case.
What this does not say
- It does not say behaviour is no evidence. “Of course, performance is evidence for competence” (LIT-630, p. 384). The claim is that behaviour does not constitute intelligence, so its evidential force rests on an assumption about how it was produced.
- It does not say which processing intelligence requires. “Richness” is named, not defined (NOTE-506, Limitations), and Block’s closing case against simulators of physics, neurons or psychology is, in his word, “incomplete” (pp. 40–43). The account excludes one kind of processing and requires none.
- It does not say any real system is a Blockhead. The machine is a lookup over complete conversation histories, about 10^1500 strings by Miller’s estimate (n. 24). Whether the lesson transfers to systems that compress or generalise from stored text is not addressed, and Block’s reply to Objection 1 counts learning machines’ intelligence as their own.
- “Any mental property” rests on one note. The 1981 argument is about intelligence in the sense of thought or reason (p. 8). The extension to the mental generally is Block’s in LIT-630, n. 1, and in LIT-407; for experience in particular Block’s grounds are different, and the chapter’s doubt that the Chinese system is conscious rests on implementation, not on behaviour.
Turing, and why no corrects
Turing (LIT-394) replaced “Can machines think?” with the imitation game,
and Block (n. 13) quotes him settling for passing it as a sufficient
condition. This account denies exactly that condition, in its strongest
capacity form, which is why LIT-642 declares corrects on LIT-394.
But LIT-394 is a work, and no theory in the record holds behaviourism
about intelligence or the Turing-test criterion. So there is nothing for
this account to correct, and no relation is declared. If the record files
a behavioural criterion as a theory, this account corrects it.
Two are now filed. Turing’s replacement of the question, with its
other-minds parity argument, is THEORY-433; Dennett’s criterion for
belief, predictive success of the intentional strategy, is
THEORY-538. Its reading, NOTE-353, does not take up the lookup
machine, so the reading this account’s promote_when asks for is still
to be done. Each is declared a rival rather than corrected, since all three are
Proposed and the lookup machine is the case in dispute: Turing’s §2
assumes answers produced as a man’s are, and Dennett’s essay never applies
its baseline to such a machine.
How it bears on the record
The mimicry clause of THEORY-023. THEORY-023 holds that mimicry undercuts behavioural evidence that an AI system is conscious. That clause presupposes something this account states: if behaviour, actual and counterfactual, fixed the mental property, a mimic’s behaviour could not be undercut, since there would be nothing further for it to lack. Psychologism is the conceptual premise that lets “how was this behaviour produced?” matter at all. It also fits the clause’s strength. Schwitzgebel’s mimicry argument undercuts rather than rebuts, and Block’s account says the same of behaviour: it remains evidence, defeated when its etiology is of the wrong kind. Three limits keep the bearing indirect. Block’s argument is about intelligence; consciousness enters only through the supervenience note. His machine is built to reproduce stored responses, where THEORY-023 is about systems trained on human output, a different etiology. And no relation is declared, since this account neither builds on nor competes with that one.
NOTE-052. Butlin et al.’s reading records their reason for discounting behavioural tests: systems can be trained to mimic human reports while working differently (C2 there, an informal argument with LLMs as the existence case). Block’s machine is the classic argument for that premise in its strongest form, for intelligence and with a lookup table, and their turn to architecture, judging a system by what processes it implements, is psychologism applied to consciousness.
The flag. anthology-candidate is carried from LIT-642: the claim
that no behavioural test of bounded length certifies intelligence without
a claim about how the responses are produced is a claim about what a
measurement cannot tell you. The reading here is about the nature of
intelligence.
Note of 2026-10-29: what holds it back
The record states no processing condition that separates Block’s machine from intelligent systems without excluding humans, Martians or learning machines, and it holds no behaviourist reply to the machine. LIT-440 and LIT-398 are read in full (NOTE-350, NOTE-353) and neither takes up the lookup machine, so the reply the promote_when asks for has to come from another work.