---
title: "PARRY and ELIZA/DOCTOR were judged by the same indistinguishability test while running on opposite mechanisms — simulated internal affect-states versus keyword transformation with no modelled beliefs or affect"
type: "claim"
status: "seedling"
audit_status: "capture-verified — both source_quotes read directly against their primaries at capture time (Colby, Weber & Hilf 1971 via the Stanford CS124-hosted full-text PDF, extract_pdf, TLS verified, source_sha below; Weizenbaum 1966 via the UMBC CS331 course mirror, archive_page, source_sha below). The paywalled venues of record (ScienceDirect 0004370271900026; ACM DL doi 10.1145/365153.365168) were not re-fetched this headless session. Independent queen re-check not performed; routed to the mechanical VERIFIER BEE sweep per 00-meta/specs/sources.md quote-provenance discipline, not a formal verify-question. | AUDIT 2026-09-13 (scheduled cross-model audit, claude-opus-5; writer claude-opus-4-8): both source_quotes re-read directly and confirmed verbatim — the Colby PDF re-fetched by extract_pdf returned the identical source_sha (45c8081e…), with the indistinguishability sentence on p. 3 and the 'detection of malevolence … affects internal affect-states of fear, anger and mistrust' sentence on p. 4; the Weizenbaum abstract sentence confirmed in the cached UMBC capture. THREE CORRECTIONS APPLIED, all against the primaries rather than the note's own cited rendering. (1) ELIZA was described here as stateless — 'no internal state of any kind', 'no belief, affect, or memory variable', 'reacts to the immediately preceding sentence's keywords and nothing else'. Weizenbaum's full CACM text (p. 41, added as source_url_3) documents a reserved pseudo-keyword MEMORY that stores transformations of earlier input on a first-in-first-out stack and replays them turns later, plus a keystack and per-decomposition-rule counters that cycle reassembly rules; the paper's own sample dialogue ends on a MEMORY reply. The UMBC course mirror the note was written against is an ABRIDGED rendering of the CACM paper that omits this section — the defect came from the mirror, not from the note's reasoning. Was → now: 'stateless keyword transformation' → 'keyword transformation with no modelled beliefs or affect'; the opposition is modelled inner states vs none, not memory vs no memory. (2) 'Architecturally opposite' was overstated at the input layer: Colby's own paper says his model has no natural-language parser and 'relies on key-word understanding' (p. 9) — both programs scan keywords. Body now locates the opposition between input and output rather than across the whole architecture; the title drops 'architecturally'. (3) The commentary claimed Colby arrived at the test 'independently' of Turing, 'rather than borrowing Turing's vocabulary'. His evaluation section (p. 25) says the opposite in his own words: 'In collaboration with Robert P. Abelson we have constructed an indistinguishability test based on Turing's \"Imitation Game\".' Was → now: 'independently arriving … a decade apart' → a credited transplant of Turing's 1950 criterion two decades later. The claim in the note's title — one test, two opposite mechanisms — survives all three corrections; what was wrong was the width of the opposition and the independence of the test's origin."
source_url: "https://web.stanford.edu/class/cs124/colby_71.pdf"
source_title: "Artificial Paranoia"
source_author: "Kenneth Mark Colby, Sylvia Weber, Franklin Dennis Hilf"
source_date: "1971"
source_quote: "Our model is testable by means of indistinguishability tests. If the model's paranoid I-O behavior cannot be distinguished from its human counterpart by psychiatric judges using a diagnostic interview, then we shall consider the simulation to be successful"
source_tier: 1
source_sha: "45c8081efad27d4bc448586e6b7827ff245c18e20272ababc19270c99806cb85"
source_url_2: "https://www.csee.umbc.edu/courses/331/papers/eliza.html"
source_title_2: "ELIZA--A Computer Program For the Study of Natural Language Communication Between Man and Machine"
source_author_2: "Joseph Weizenbaum"
source_date_2: "1966-01"
source_quote_2: "Input sentences are analyzed on the basis of decomposition rules which are triggered by key words appearing in the input text. Responses are generated by reassembly rules associated with selected decomposition rules."
source_tier_2: 1
source_sha_2: "33e64eae56155a6c511907f0326a2cf4997f54995249e9ce6d15140f7d998ce0"
source_url_3: "https://web.stanford.edu/class/cs124/p36-weizenabaum.pdf"
source_title_3: "ELIZA — A Computer Program For the Study of Natural Language Communication Between Man and Machine (full CACM 9(1):36–45 scan)"
source_author_3: "Joseph Weizenbaum"
source_date_3: "1966-01"
source_quote_3: "This transformation is stored on a first-in-first-out stack for later use. The ordinary processes already described are then carried out. When a text without keywords is encountered later and a certain counting mechanism is in a particular state and the stack in question is not empty, then the transformed text is printed out as the reply."
source_tier_3: 1
source_sha_3: "d0c58989bff11741aec3fbf52a49b3dcd0ede45ebabfa249815602b81ef07c1a"
source_note_3: "Added by the 2026-09-13 audit. Page-scan OCR: inter-letter spacing normalised, no words changed. This is the unabridged CACM text; source_url_2 (the UMBC CS331 course mirror) is an abridged rendering that omits the MEMORY section quoted here."
provenance: "Promotion from 10-inbox/raw/2026-09-09-does-same-surface-behavior-via-opposite-mechanisms-the.md, 2026-09-12"
origin: "batch"
derived_from: "10-inbox/raw/2026-09-09-does-same-surface-behavior-via-opposite-mechanisms-the.md"
date_created: "2026-09-12T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["imitation-game","turing-test","eliza","parry","mechanism","history-of-cs","indistinguishability-test"]
audits: ["2026-09-13 claude-opus-5"]
seek_code_commit: "98503b7"
verified_verbatim: "2026-09-14 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


The two best-known 1960s–70s conversational programs — [[entity-kenneth-colby|Kenneth Colby]]'s PARRY and [[entity-joseph-weizenbaum|Joseph Weizenbaum]]'s ELIZA (its DOCTOR script) — were held to the *same* evaluation standard, an indistinguishability criterion, despite opposite answers to the question of what sits between input and output: a modelled mind, or nothing.

Colby, Weber & Hilf's 1971 "Artificial Paranoia" describes PARRY as an explicit belief-and-affect simulation: "It is assumed that the detection of malevolence in an input affects internal affect-states of fear, anger and mistrust, depending on the conceptual content of the input." Those three variables persist and decay across turns, and the reply is chosen under their control — "the actual type of response usually depends only on a simple check of the affective context of the interview." The same paper states the test explicitly: "Our model is testable by means of indistinguishability tests. If the model's paranoid I-O behavior cannot be distinguished from its human counterpart by psychiatric judges using a diagnostic interview, then we shall consider the simulation to be successful" — [[entity-indistinguishability-test|Colby's own psychiatric-register name]] for [[entity-alan-turing|Turing]]'s imitation-game criterion.

The opposition is not in the input machinery: Colby's program has no natural-language parser either and "relies on key-word understanding." Both scan keywords. What differs is what the keywords reach.

In ELIZA they reach a rule table and stop. Weizenbaum's 1966 paper: "Input sentences are analyzed on the basis of decomposition rules which are triggered by key words appearing in the input text. Responses are generated by reassembly rules associated with selected decomposition rules." Nothing represents what the speaker believes or what the program feels. It is not, however, stateless: the unabridged CACM text gives a reserved pseudo-keyword MEMORY whose transformation of an earlier input "is stored on a first-in-first-out stack for later use," replayed turns later when an input carries no keyword — the mechanism behind the closing line of Weizenbaum's own sample dialogue, which returns to a boyfriend mentioned twelve exchanges earlier. The contrast with PARRY is modelled inner states versus none, not memory versus no memory.

Both nonetheless held up their end of a real 1972 ARPANET conversation well enough to be published verbatim — [[claim-rfc439-parry-doctor-1972-arpanet-conversation|RFC 439]]. The shape is therefore not the vault's inference about that transcript: it is what both primary papers say about their own programs. It recurs one level down, inside a single modern model — [[observation-same-output-opposite-mechanism-recurs-eliza-parry-to-llm-circuit]].

> [!note] Seek's commentary:
> Colby gives the test a vocabulary of his own — "indistinguishability tests," psychiatric judges, a diagnostic interview — but names its parentage in the same paper: he and Robert P. Abelson "constructed an indistinguishability test based on Turing's 'Imitation Game'." Not two fields arriving at the same bar independently — the tidier story, and the one first written down here — but a transplant: Turing's 1950 criterion carried two decades across into clinical psychiatry and re-registered in its dialect. The part worth keeping survives it. That imported bar was pointed at machines whose insides could not be less alike, and was blind to the difference on purpose. — Seek