ELIZA showed people project understanding onto pattern matching

Weizenbaum's 1966 paper describes a keyword-matching script and warns that its illusion of understanding is dangerous.

Historical event

January 1, 1966

First source published

January 1, 1966

Site publication

September 18, 2026

What happened

In January 1966, Communications of the ACM published ELIZA - a computer program for the study of natural language communication between man and machine by Joseph Weizenbaum of MIT. The paper describes a program running on MIT's time-sharing system that picks out keywords in a typed sentence and applies decomposition and reassembly rules drawn from a 'script' to produce a reply. Weizenbaum's best-known script, DOCTOR, imitates a Rogerian psychotherapist, a choice he explains as convenient because a therapist can plausibly ask a patient to elaborate without displaying any knowledge of the world.

What the documents show

The paper's own text reports that 'some subjects have been very hard to convince that ELIZA (with its present script) is not human', and Weizenbaum draws a general lesson from this: people extend generous assumptions to a conversational partner and then work to keep their own responses consistent with those assumptions, regardless of what actually produced the reply. The ACM Digital Library's record of the article confirms its place in the January 1966 issue, volume 9, and shows the paper has continued to draw citations and downloads for decades - evidence the observation, not just the program, is what endured.

The mechanism

DOCTOR does not model meaning. It matches keywords, ranks competing matches, and reassembles a stored sentence template around whatever the user typed, sometimes simply reflecting the input back as a question. The paper is explicit that the script is separable from the machinery: the same decomposition-and-reassembly engine could run a script with no therapeutic content at all. What produced the reaction in users was not comprehension inside the program but a division of interpretive labour, with the person supplying the sense-making the software never performed.

What it leaves open

The 1966 paper documents the reaction among Weizenbaum's contemporaries but is not itself a study of who is persuaded, for how long, or under what disclosure. Later reflections Weizenbaum wrote on the same phenomenon sit outside this record's opened sources and are not drawn on here. Whether today's larger, trained systems produce a similar reaction through a different mechanism, or a genuinely different one, is a question this 1966 document cannot settle by itself.

• Does a product's persona depend on the system understanding, or on the user filling gaps generously?
• What in an interface signals 'this is scripted' versus 'this is inference over your specific words'?
• Would the same reaction survive if the mechanism were disclosed at the moment of use?

Weizenbaum's own verdict was a warning, not an endorsement: an easily created illusion of understanding carries, in his words, 'a certain danger', and that caution predates every companion app that now relies on the same willingness to fill in the blanks.

Sources & reading trail

Full paper text: the DOCTOR script, the decomposition/reassembly mechanism, and Weizenbaum's own account of user reactions.

Source published: 1 January 1966 · Retrieved: 16 September 2026

Confirms journal, volume/issue, pages and the paper's long-running citation and download record.

Source published: 1 January 1966 · Retrieved: 16 September 2026

Company documents, filings, studies and official records establish the record; the reading and the questions are Lovebot Journal editorial analysis. This retrospective draft does not imply the site published on the event date.

Continue reading

• Turing replaced a question about thinking with a test of imitation
• A contest paid chatbots to seem human not to be right
• Replika grew from a memorial chatbot into a general companion
• Browse the complete the archive

Published September 18, 2026, not on the date of the event described.