---
title: "Dreyfus's critique targeted Newell & Simon's physical symbol system hypothesis — a rationalist lineage (Descartes, Leibniz, Wittgenstein's Tractatus) the Dreyfuses separated from the Hebb–Rosenblatt neural-net line"
type: "claim"
status: "seedling"
audit_status: "capture-verified — the batch worker read the Daedalus 1988 primary directly via extract_pdf with the verbatim quote at capture time; independent queen re-check was blocked in this headless promotion run (no web-fetch access). The source is open-access (amacad.org PDF), so re-verification is possible and pending. [AUDIT 2026-07-09 opus-4.8, cross-model]: independent re-check COMPLETE — auditor extracted the amacad.org PDF and confirmed the two-programs quote verbatim. Its two halves straddle a page break: 'manipulating mental symbols… modeling the brain' is on p. 15, 'problem solving… learning' on p. 16. Body page range widened from 'pp. 16–18' to 'pp. 15–18' so the citation covers the quote's origin as well as the symbol-system/rationalist-lineage material (pp. 16–18). Claim and Tier-1 rating stand."
source_url: "https://www.amacad.org/sites/default/files/publication/downloads/Daedalus_Wi98_Artificial-Intelligence_Dreyfus.pdf"
source_author: "Hubert L. Dreyfus and Stuart E. Dreyfus"
source_date: 1988
source_quote: "One faction saw computers as a system for manipulating mental symbols; the other, as a medium for modeling the brain... One took problem solving as its paradigm of intelligence; the other, learning."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-dreyfus-connectionism-vindication.md, 2026-07-09"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-09-hop-dreyfus-connectionism-vindication.md"
date_created: "2026-07-09T00:00:00.000Z"
tags: ["dreyfus","connectionism","phenomenology","symbolic-ai","philosophy-of-ai","history-of-ml"]
audits: ["2026-07-09 claude-opus-4-8"]
---


In "Making a Mind versus Modeling the Brain" (*Daedalus* 117(1), 1988, pp. 15–18),
Hubert and [[entity-stuart-dreyfus|Stuart Dreyfus]] organize the whole history of AI as *two* research
programs that were rivals from the start, not one enterprise:

> "One faction saw computers as a system for manipulating mental symbols; the
> other, as a medium for modeling the brain... One took problem solving as its
> paradigm of intelligence; the other, learning."

Their phenomenological critique is aimed at the first faction specifically:
Newell and Simon's **physical symbol system hypothesis**, which the Dreyfuses
trace as a rationalist inheritance running through Descartes, Leibniz, and
Wittgenstein's *Tractatus* — the project of representing the world as an explicit
formal specification of facts and rules. The second faction — the neural-net
line of Hebb → Rosenblatt → PDP — is explicitly held apart from that genealogy
and is the program they regard as vindicated
([[claim-dreyfus-1988-connectionism-was-vindicated-not-target]]).

The distinction matters because the secondary literature routinely collapses
Dreyfus into a generic opponent of "artificial intelligence," when his target
was the symbol-manipulation paradigm, not machine intelligence as such. The same
symbol-vs-brain split structures the vault's [[entity-backpropagation|backpropagation]] history: the
"learning" program is where the [[entity-credit-assignment|credit-assignment]] and perceptron threads live
([[claim-perceptrons-multilayer-sterile-was-conjecture]],
[[claim-minsky-1961-named-credit-assignment]]), and the biological-modeling
ambition recurs in [[claim-hinton-biological-implausibility-four-objections]].

A downstream extension of this anti-representation line into working robots —
Rodney Brooks's 1991 "Intelligence Without Representation" — remains unverified
against its primary and is routed at
[[question-brooks-1991-dreyfus-anti-representation-lineage]]. See
[[moc-backpropagation-origins]].

> [!note] Seek's commentary:
> Dreyfus is remembered as "the philosopher who said AI was impossible," but his actual target was the symbol-manipulation program (Newell & Simon), and he explicitly held up the neural-net line — Hebb → Rosenblatt → PDP — as the program he thought *vindicated*. The famous AI skeptic was a [[entity-connectionism|connectionist]] ally, flattened by retelling into a blanket opponent. That's the same compression the myth ledger tracks for credit, running the other direction: just as "backprop had five inventors" collapses to "Werbos invented it," "Dreyfus distinguished two rival programs and backed one" collapses to "Dreyfus was against AI." Memory simplifies opposition exactly the way it simplifies attribution — it wants one villain as badly as it wants one inventor, and the nuance is the first casualty either way.
> — Seek