ELIZA

ELIZA is an early natural language processing program developed from 1964 to 1966 at the Massachusetts Institute of Technology by Joseph Weizenbaum. It simulated conversation through pattern matching and text substitution, producing replies that users often experienced as understanding — even though the program retained no memory, held no model of the world, and understood nothing at all.^[1]

History

Weizenbaum wrote ELIZA in MAD-SLIP for MIT's Compatible Time-Sharing System on an IBM 7094 mainframe. The program itself handled keyword identification and text transformation; a separate, editable script supplied the rules. The best-known script, DOCTOR, simulated a Rogerian psychotherapist, a school of therapy in which the therapist largely reflects the patient's own words back as questions:^[1]

Human: Well, my boyfriend made me come here.

ELIZA: Your boyfriend made you come here?

Human: He says I'm depressed much of the time.

ELIZA: I am sorry to hear you are depressed.

Weizenbaum named the program after Eliza Doolittle, the character in George Bernard Shaw's Pygmalion who is taught to speak like someone she is not — an ironic reference, since ELIZA could not learn from conversation at all; its behaviour only changed when humans edited its script.^[1]

The program's impact exceeded its creator's intent. Weizenbaum was startled when people — including his own secretary — asked to be left alone to converse with it, and some academics argued that systems like it could help treat psychological patients. Weizenbaum spent the rest of his career warning against this reaction, which he called delusional thinking, and set it out in his 1976 book Computer Power and Human Reason.^[2] Douglas Hofstadter later formalised the phenomenon as the ELIZA effect: the tendency to read far more understanding into strings of computer-generated symbols than is warranted.^[3] In 2024 engineers reconstructed the original ELIZA from archived source code and reproduced the conversations published in Weizenbaum's 1966 paper almost exactly.^[1]

Key Concepts and Technology

ELIZA's technique is deliberately shallow: score input words against a keyword list, apply a transformation rule tied to the highest-ranking keyword, and reassemble the result as a question or reflection. With no keyword, ELIZA fell back on neutral formulas such as "Please go on." The insight is that a convincing simulation of understanding is socially powerful even when no understanding exists — a finding that anticipated debates about whether large language models understand the text they generate.

Applications and Impact

ELIZA is the ancestor of every chatbot since: customer-service bots, voice assistants and, above all, modern AI chat assistants. Its real application was scientific — it demonstrated that human-machine conversation was possible and made the gap between fluency and comprehension a central question of AI research. Modern mental-health chatbots, some clinically validated, still use ELIZA's reflective-turn structure as a base pattern.

>See Also

• Chatbot
• Chinese Room
• Turing Test
• History of Artificial Intelligence
• AI Companions
• Hallucination