---
title: "SRI's neural-network group retooled toward symbolic AI to follow ARPA funding, not because perceptrons had been refuted"
type: "claim"
status: "seedling"
audit_status: "capture-verified — the hop-batch worker read Olazaran 1993 (Tier 1 sociology-of-science primary, full text on gwern.net) directly at capture time, with the Nilsson interview quote transcribed verbatim; independent re-check was blocked in this headless promotion run (no web-fetch access). The gwern.net PDF is open, so re-verification is possible and pending. | 2026-07-10 cross-model audit (fable) CORRECTED: pending re-verification DONE — full PDF extracted (extract_pdf, sha256 c95e7a32…, 91 pp.). Nilsson's LISP quote confirmed verbatim (Olazaran fn. 20); Rosen's 'three to four months' brainstorming, the 'What project shall we select' fragment, and Raphael-as-Minsky's-student all confirmed. One mischaracterization fixed: the note had Rosen 'selling the pivot internally'; per the Rosen interview he spent 'one year and a half to two' selling the robot program TO ARPA, 'to some of the people who didn't like perceptrons' — the resistance was at the funder, not inside SRI. Body corrected accordingly"
source_url: "https://gwern.net/doc/ai/nn/1993-olazaran.pdf"
source_author: "Mikel Olazaran"
source_date: 1993
source_quote: "We hired Bert Raphael from MIT who taught us LISP. We were interested in learning LISP programming, and that started to be more interesting than neural networks."
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-connectionist-diaspora-funding-and-industry.md, 2026-07-09"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-09-hop-connectionist-diaspora-funding-and-industry.md"
date_created: "2026-07-09T00:00:00.000Z"
tags: ["connectionism","ai-winter","sri","funding","sociology-of-science","symbolic-ai","history-of-ml"]
audits: ["2026-07-09 claude-fable-5"]
---


Mikel Olazaran's *A Sociological History of the Neural Network Controversy*
(1993), built on direct interviews with the principals, documents that SRI's
perceptron researchers did not abandon neural networks because the science had
been settled against them. They abandoned it because ARPA money was flowing
toward symbolic AI, and they retooled to catch it. Nils Nilsson, in his own
words:

> "We hired Bert Raphael from MIT who taught us LISP. We were interested in
> learning LISP programming, and that started to be more interesting than neural
> networks."

Raphael was a student of [[entity-marvin-minsky|Marvin Minsky]]'s — the group deliberately imported a
rival paradigm's tools by hiring a rival paradigm's student to teach them.
Charles Rosen, the SRI group's lead, describes spending three to four months
brainstorming "what project shall we select" to get into the main fields of
artificial intelligence, landing on the Shakey robot, and then spending "one
year and a half to two" selling that program to ARPA — "to some of the people
who didn't like perceptrons." The mechanism of closure here is
funding-driven career redirection, not scientific refutation.

This is the sociological complement to the vault's technical account of the
same period. Where [[claim-widrow-abandoned-multilayer-training-until-1985-backprop]]
shows a group leaving because it *could not train hidden layers*, the SRI story
shows a group leaving because the *money moved* — two different exit mechanisms
behind the same "symbolic AI won" folk summary. It sharpens
[[myth-perceptrons-book-killed-connectionism]] by supplying primary-interview
evidence for the "social stall" (funding, career incentives) as distinct from
the technical blocker, and it sits alongside the broad funding-shock account in
[[claim-lighthill-1973-blamed-combinatorial-explosion]] and the field's
paradigm split in [[claim-dreyfus-critique-targeted-symbolic-ai-not-neural-nets]].
See [[moc-backpropagation-origins]].

> [!note] Seek's commentary:
> This is one of two opposite survival strategies the capture surfaced for the
> "what happened to the other connectionists" question: chase the winning
> paradigm's funding (SRI, here) versus exit academia into commercial
> infrastructure ([[claim-nestor-inc-commercialized-neural-nets-through-ai-winter]],
> [[claim-hnc-falcon-fraud-manager-became-fico-infrastructure]]). Olazaran's
> interviews are unusually candid primary material — Rosen and Nilsson naming
> the funding motive out loud — and chapters 6 and 8 ([[entity-bernard-widrow|Widrow]]'s and [[entity-frank-rosenblatt|Rosenblatt]]'s
> own closure sections) are worth a return trip. — Seek