Daniel C. Dennett (1978), “Artificial Intelligence as Philosophy and as Psychology”, ch. 7 of Brainstorms: Philosophical Essays on Mind and Psychology, Montgomery, VT: Bradford Books, 1978 (Library of Congress 78013723). Also in M. Ringle (ed.), Philosophical Perspectives in Artificial Intelligence, Humanities Press and Harvester Press, 1979. Reprinted in the MIT Press edition of Brainstorms (1981; DOI 10.7551/mitpress/1664.003.0011) and its 40th-anniversary edition (2017, pp. 119–138).
On the citation. The brief gave Ringle (1979) as the venue and Brainstorms (1978) as a reprint. By date the order is the other way round: Brainstorms is catalogued 1978 and Ringle’s volume 1979, so the essay’s first appearance in print is Brainstorms. Whether it was written for Ringle’s volume and printed first in Brainstorms only because that book came out sooner, I could not tell, since neither book’s acknowledgements could be read. Pages:
- Brainstorms (1978): the passages Stich quotes are on pp. 123–124. The chapter’s full range was not confirmed.
- Ringle (1979): a search-engine summary of the PhilPapers record gives pp. 57–80; PhilPapers itself was blocked, so this is unconfirmed.
- 2017 edition: pp. 119–138 (Project MUSE).
The identifier is the 1981 MIT Press chapter’s DOI, as for “Where Am I?”
(LIT-405) from the same book; the 1978 Bradford printing has no DOI.
published: is the 1978 book, by year only.
Key takeaways
Registered from a contemporary summary, not a reading. Everything below is what Stich (1980) reports, with his page references to the 1978 printing. Dennett’s reply begins: “I have no quarrel with Stich’s summary of my views.”
- The worry. Stich calls the essay “a travel guide for philosophers” to AI, which also states and answers what Dennett labels Hume’s Problem: a representation is useless unless someone reads it, so a psychology of internal representations seems to need a homunculus to interpret them, and so to be circular or regressive.
- The answer. Homunculi are harmless if they are stupider than the whole. “If one can get a team or committee of relatively ignorant, narrow-minded, blind homunculi to produce the intelligent behavior of the whole, this is progress” (p. 123). Each can be explained by “teams of still stupider ones” (Stich’s paraphrase), and “eventually this nesting of boxes within boxes lands you with homunculi so stupid … that they can be, as one says, ‘replaced by a machine’” (p. 124). The ellipsis is Stich’s.
- AI as the method. AI’s bottom-level homunculi are and-gates and or-gates, and AI theorists organise “armies of such idiots” (p. 124, as quoted by Stich) to produce intelligent behaviour. No homunculus may “duplicate entire the talents they are rung in to explain” (p. 123). “Homunculus talk is ubiquitous in AI” (p. 123).
The brief’s “army of idiots”. The phrase Stich quotes is “armies of such idiots”, and he quotes it as a fragment. I could not read the sentence it comes from, so it is not quoted in full here.
Standing in the record
Filed on 2026-10-02 at the owner’s request for Dennett, as the essay in which the debt that “Intentional Systems” (LIT-440, read in NOTE-350) names in 1971, the “loan of intelligence” that every content-ascribing theory takes out, is said to be repayable.
How it relates to Minsky and Block, on Stich’s summary:
- Minsky (LIT-399, the 1977 memo, read in NOTE-344; The Society of Mind, LIT-429, unread). Both describe a mind as an organisation of agents none of which has the whole’s competence. They differ in direction and in what the agents are for. Dennett’s homunculi are an explanatory device: boxes in a top-down flow chart, each borrowing intelligence, posited and then discharged by decomposition until nothing borrowed remains. Minsky’s agents are a design: simple from the start, built upward, and their interactions are the mind. So Dennett’s argument is the licence for Minsky’s picture rather than a version of it: it says why a mind of agents is not a regress, as long as the agents are stupider than what they compose. Whether either text cites the other is not known here.
- Block (LIT-407, read in NOTE-366). Block’s homunculi-heads run the other way. His homunculi are whole people, the citizens of China, each doing one table entry, so they are smarter than the role they fill. Block’s note 19 replaces each one with a mindless gate and says his doubt about qualia survives. That replacement is the bottom rung of Dennett’s discharge, reached in one step. So discharge does not answer Block: it shows that homunculi cost a theory of intelligence nothing, which Block concedes (he grants a homunculi-head beliefs); Block’s doubt is about qualia, which discharge does not address. The two essays appeared in the same year, 1978.
For THEORY-043 this is the downward direction’s warrant. A society of mindless agents explains a mind exactly when it discharges its homunculi, and Schwitzgebel’s upward case (LIT-159) is the one where the parts are as smart as, or smarter than, the whole, the case discharge was never meant to license. Selfridge’s demons (LIT-427) sit on Dennett’s side of that line, and so, as THEORY-043 describes it, does Hofstadter’s ant colony in GEB (LIT-437, unread).
A reading should check, against the essay itself: the full “armies of such idiots” sentence; whether the essay cites Minsky, Selfridge or Newell and Simon for the top-down method; and whether it says anything about consciousness or only about intelligence. The Mind’s I (LIT-413) and Searle’s Chinese room (LIT-400, whose systems reply is a homunculus question) are the nearest filed neighbours.
It carries no instruction for machine-learning practice. It is a philosopher’s account of 1970s AI method, and the anthology does not hold it.