ELIZA is an early 🏷️#natural_language_processing program, written at the MIT Artificial Intelligence Laboratory between 1964 and 1966 by Joseph Weizenbaum, that simulated conversation by matching patterns in a user's input and rephrasing their own words back at them.

Its best-known script, DOCTOR, imitated a Rogerian psychotherapist — a choice that let the program deflect almost any input without needing to understand it. "I am unhappy" returns "Why do you say you are unhappy?" ELIZA held no memory of the conversation, no model of the world, and no representation of the subject under discussion; it ranked keywords against a ruleset and applied substitution templates.

Weizenbaum built it to demonstrate how superficial human-machine communication was, and was unsettled by what happened instead. Users attributed genuine understanding to it and confided in it readily — his own secretary reportedly asked him to leave the room. He spent much of his later career, most directly in Computer Power and Human Reason (1976), arguing against delegating human judgment to machines.

The ELIZA effect — the tendency to read far more comprehension into a program's output than its mechanism supports — is the term's most durable contribution. It applies more sharply to modern 📝Chatbot systems and large language models than it ever did to the script that named it.