ELIZA made a simple script feel like an attentive listener.

How ELIZA made a script feel like a listener

In 1966, Joseph Weizenbaum built ELIZA, a computer program that imitated a particular kind of conversation: a therapist asking a client to say more. Its best-known script, DOCTOR, did not diagnose anyone or understand what they said. It searched each message for recognizable words and patterns, then used rules to produce a reply.

A sentence such as “I feel unhappy” could prompt a response like “Why do you feel unhappy?” If a message contained no pattern the program knew how to use, ELIZA could fall back on a general prompt. The exchange sounded responsive because the program returned the user’s own concerns in a new form—not because it had grasped their meaning.

The user supplied the depth

That simple technique worked especially well in a therapy-like conversation. A question invites the speaker to explain; a reflection gives them room to continue. Users could bring personal worries to the exchange, while ELIZA’s open-ended prompts made it easy to keep talking. The feelings and interpretation came from the person on the other side of the screen.

This gap between what a program does and what a person experiences is the important part of ELIZA’s story. A system does not need to understand a confession for someone to feel heard while making it. The conversational setting, the user’s own expectations, and the chance to elaborate can all lend a bare pattern-matching script a human significance it does not possess.

Weizenbaum’s uncomfortable lesson

Weizenbaum had intended ELIZA as a demonstration of how limited computer conversation could be. Instead, people’s reactions showed how readily a convincing exchange could draw them in. The lesson was not that a 1966 script secretly understood its users, or that every conversation with software is a substitute for human company. It was that fluent-seeming replies can encourage people to supply understanding—and intimacy—that the machine itself lacks.

ELIZA’s modest trick was to turn a user’s words back toward them as questions. Its lasting consequence was to make that trick impossible to dismiss as merely technical: people can experience a conversation as personal even when the program behind it is only following rules.