ELIZA

A program to talk with

ELIZA is a program that makes natural language conversation with a computer possible ^[2]. It was written in MAD-SLIP and ran on an IBM 7094 under CTSS ^[1]; the paper announcing it describes it as implemented on the MAC time-sharing system at MIT ^[2].

Where it came from

Weizenbaum developed the program at MIT between 1964 and 1967, and its release is listed as 1966 ^[1]. He intended it as a method to explore communication between humans and machines rather than as a product or a claim about machine intelligence ^[1]. The paper, "ELIZA - A Computer Program For the Study of Natural Language Communication Between Man and Machine," appeared in Communications of the ACM, Volume 9, Issue 1, pages 36 to 45, on 1 January 1966 ^[2]. The name was chosen to emphasise that the program may be incrementally improved by its users, since its language abilities may be continually improved by a teacher ^[2].

How it worked

ELIZA used pattern matching and substitution to create an illusion of understanding, but it gave no response that could be considered really understanding what was said ^[1]. It had no built-in contextual framework or universe of discourse, so separate scripts supplied most of its language capability ^[1]. Within a script, keywords were assigned a precedence number, or RANK, and the highest-ranked keyword on the keystack determined the transformation applied to the input ^[1]. In the terms of the paper's abstract, input sentences are analysed on the basis of decomposition rules triggered by key words appearing in the input text, and responses are generated by reassembly rules ^[2].

The DOCTOR script

The most famous script, DOCTOR, simulated a psychotherapist of the Rogerian school, which often reflects the patient's words back to them ^[1]. This is a style of therapy in which the therapist restates or turns around what the client has just said instead of steering the conversation. DOCTOR used rules dictated in the script to respond with non-directional questions to user inputs ^[1]. Weizenbaum chose the psychotherapeutic setting to sidestep the problem of giving the program a database of real-world knowledge ^[1]: a therapist who mostly asks questions back can keep talking without knowing anything about the world.

The reaction

Weizenbaum was surprised that some people, including his secretary, attributed human-like feelings to the program, a reaction later called the ELIZA effect ^[1]. He wrote that he had not realised that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people ^[1]. The lay response disturbed him and led him to write Computer Power and Human Reason: From Judgment to Calculation ^[1].

Its place in the history

ELIZA is an early demonstration that a conversation can be carried on fluently without the program holding any model of what the words refer to. Everything the user sees comes from rules written into a script — keyword ranks, decomposition rules, reassembly rules — and the same machinery can be re-pointed at a different subject by swapping the script ^[1][2]. The program itself contributes no context and no universe of discourse of its own ^[1]. Two things outlasted the particular code: the design habit of building a conversational system out of a small engine plus a body of scripted rules, and the name given to the way people read understanding into a machine that only reflects their words back ^[1].