Textbook of AI

ELIZA is the natural-language conversational program written by Joseph Weizenbaum at MIT in 1964–1966. It conducted dialogue by pattern-matching the user's input against a script of patterns and substituting matched fragments back into a response template. Its best-known script, DOCTOR, simulated a Rogerian psychotherapist by reflecting the user's statements back as questions ("My mother hates me." → "Tell me more about your mother."). Two simple devices, keyword detection ("mother", "father", "computer") and the substitution of "I" for "you" and vice versa, sufficed to produce dialogue that felt, to many users, surprisingly natural.

Weizenbaum was disturbed to find that users, including his secretary, formed real emotional attachments to the program. The phenomenon, sometimes called the ELIZA effect, is the human tendency to attribute mind, intention and understanding to systems that exhibit no such properties, a tendency that has only grown more consequential with the deployment of modern large language models.

ELIZA was the prototype of every subsequent chatbot, from PARRY (Kenneth Colby's 1972 program simulating a paranoid patient, which "conversed" with DOCTOR over the ARPANET in a famous 1972 exchange) through the Loebner Prize chatbots of the 1990s and 2000s to modern conversational LLMs. The architecture itself, keyword spotting, pattern templates, hand-coded substitutions, is almost entirely absent from modern systems, but the philosophical questions ELIZA raised about machine simulation of conversation remain open.