Ned Block (1995), “The Mind as the Software of the Brain”, in Edward E. Smith and Daniel N. Osherson (eds.), Thinking: An Invitation to Cognitive Science, 2nd edition, Volume 3, Cambridge, MA: MIT Press, chapter 11, pp. 377–425 — DOI-10.7551/mitpress/3966.003.0016.
The brief’s citation (in An Invitation to Cognitive Science, 1995) is correct; it is volume 3, Thinking, of the second edition.
Key takeaways
- The Turing test fails as a definition (§11.1.1). The judge is the gap: in the First Turing Test (Boston Computer Museum, 1991) an ELIZA variant was classified as human by five of ten judges, and Block was one of the referees who failed to enforce the restriction to non-tricky questions. The “Aunt Bubbles machine” (the Blockhead of LIT-642, n. 1) refutes even the claim that the mental supervenes on the behavioural.
- Functional analysis and primitive processors (§§11.1.3–11.1.4). Intelligent capacities are explained by decomposition into less intelligent ones, ending in primitive processors whose working is a question for electronics or physiology. “Primitive processors are the only computational devices for which behaviorism is true.” Gates can be electrical or a cat and mice; hardware is irrelevant to computational description, which is why the computer model is “profoundly unbiological” (§11.1.5).
- Intelligence vs intentionality (§11.2). A swamp-brain duplicate is arguably intelligent but has no intentional states: “intelligence is future oriented”, intentionality “has a past-oriented requirement”. Functional decomposition explains intelligence, not intentionality, since parts can be as intentional as the whole.
- Syntactic engine, semantic engine (§§11.2.1–11.2.2). An adder works because a symbol function is isomorphic to a semantic function; at the primitive level, subject matter shifts from numbers to numerals. Searle’s wall computes nothing, because the isomorphism must cover “all the computations that the machine could have performed”: “A rock … is not an X-OR gate.”
- Language of thought and functional role (§§11.3–11.4). Content derives from the meaning of internal symbols, and functional role, including input and output relations, fixes meaning. Dennett’s three objections (infinite beliefs, queen out early, sister in Cleveland) are answered with a protoscientific notion of explicit, causally active belief. The systematicity argument is sound for conscious thought but “shows little about encapsulated modules and other unconscious systems”.
- Levels and the Chinese room (§§11.5–11.6). The “Reductionist Cruncher”: if a syntactic level undermined the content level by being more general and finer-grained, physics would undermine syntax in turn. The systems reply is right that a person can implement a Chinese-understanding system without understanding Chinese. But “I have my doubts as to whether there is anything ‘it is like’ to be the Chinese system”, since “consciousness is more a matter of implementation of symbol processing than of symbol processing itself”.
Standing in the record
Filed on 2026-10-03 at the owner’s request, as the short statement of the computational theory of mind by its most persistent critic of functionalism about experience. No anthology topic holds a philosophical account of the computer model of mind, and it carries no instruction for machine-learning practice. It is a textbook chapter; read it for its formulations, and the primary papers for the arguments.
No relation is declared: it restates Block’s own earlier arguments and surveys others’.
Where it bears.
- Chalmers on implementation (LIT-404, NOTE-342). §11.2.2 answers Searle’s and Putnam’s triviality claim the way Chalmers does, by counterfactual transitions, and cites Chalmers’s “On implementing a computation” (1994) and O’Rourke and Shattuck’s critique of Putnam’s theorem.
- Searle (LIT-400, NOTE-349). The systems reply is defended against the internalisation move with a nine-to-five Chinese simulator who quits at 5 p.m. The concession on phenomenal consciousness is the point where Block and Searle agree.
- AI consciousness (LIT-056, NOTE-052; THEORY-023). Butlin et al. adopt computational functionalism as a working hypothesis. Block grants a computational account of intelligence and of intentionality but doubts it for phenomenal consciousness, on the ground that consciousness may depend on implementation. That is THEORY-023’s second deadlock, stated in 1995 by a functionalist about thought.
- NOTE-117 (RTM, metasemantics). The orthodoxy Block describes, Fodor’s Belief Box with meanings fixed by functional role including perception and action, is the form of RTM NOTE-117 defends against Toon, and n. 4 notes that the reference of ‘Aristotle’ in the language of thought derives from the public word, a point for a socially grounded metasemantics.
- Connectionism. Backpropagation models are “firmly within the computationalist paradigm” because they keep connections that flip sign (§11.1.5). López-Rubio’s computational functionalism for deep learning (LIT-148) is a later neighbour.