Category: artificial-intelligence
Type: thought experiment
Associated thinker(s): Ned Block
Original publication or approximate period: Block, “Psychologism and Behaviorism” (The Philosophical Review, 1981)
Difficulty: Intermediate
Themes: behaviourism, Turing test, intelligence
Related experiments: The Turing Test; Chinese Room; China Brain
Imagine a machine designed for a one-hour typed conversation. Block asks us to suppose that programmers prepare a finite catalogue of typable conversational strings they judge sensible. The machine searches for strings matching the conversation so far and chooses a continuation. A tree-search version can store shared branches instead of repeating whole strings.
On Block’s account, the machine’s sensible responses come from this prepared catalogue, not from a response-producing process he regards as the machine’s own intelligence.
If a system’s behaviour is indistinguishable from an intelligent being’s, is that enough to say it is intelligent?
Block argues that the capacity to produce sensible conversational behaviour is not, by itself, sufficient for intelligence: the process producing it also matters. He presents the machine as a counterexample to the neo-Turing and related behaviourist conceptions he discusses, not as proof that every behavioural test has no evidential value.
Block calls his view “psychologism”: whether behaviour is intelligent can depend on the internal information processing that produces it. His conclusion concerns this machine’s particular process; he does not require all intelligent systems to share one kind of processing.
The Stanford Encyclopedia of Philosophy describes two possible replies: deny that Blockhead is logically possible, or deny that it lacks intelligence. It also notes that the judgment is contested. Schwaninger (2022) separately describes the table as physically infeasible and interprets passing the Turing test as evidence rather than a necessary condition. These are named positions, not a consensus. Block also connects the case to a homunculi-head challenge to functionalism and presents a possible “bite the bullet” reply.
Block considers an objection that the table would require more strings than there are particles in the universe. He replies that the counterexample needs logical possibility, not practical feasibility, and discusses ways to reduce storage. The size comparison is part of the stipulated illustration, not a verified physical estimate.
Block does not argue that machines cannot think. He argues that behavioural capacity alone is insufficient under the behaviorist conception he targets.
Block discusses a fixed test duration, Basic English, and a tree-search version. As teaching extensions, one could shorten the test, restrict its topics, or imagine a chess-playing lookup strategy; these are not historical variants attributed to Block in the sources checked.
Block’s full 1981 article was read in a posted copy, especially §§1–2 and the objections and replies on pp. 17–21; publication details were checked against its DOI record. The Stanford Encyclopedia of Philosophy entry “The Turing Test” was consulted in §4.2. Relevant sections of Schwaninger (2022) were read. The earlier expansion record logs page requests and keyword counts, not passage-level review. The term “Blockhead” is the SEP’s label for Block’s machine thought experiment. This is an editorial source check, not expert review.
Research check: 2026-11-01. Further comparisons and the chess example are teaching prompts, not established historical variants.