2026-09-17 14:39 UTC
ELIZA, a pychoanalysis simulation, developed in 1967 by Joseph Weizenbaum, was the first AI that - according to Alex Hanna and Emily Bender in The AI Con - broke people's brains.
Weizenbaum intended the program as a method to explore communication between humans and machines. He was surprised that some people, including his secretary, attributed human-like feelings to the computer program
https://en.wikipedia.org/wiki/ELIZA
The trick was simple. Take whatever the user typed in and then reframe it back to them as a question that prods more info out of the user.
Prompt: "I've been having feelings of sadness lately."
ELIZA: "I see. Tell me more about your feelings of sadness."
When I was 11 in 1980, I ported ELIZA to my Atari 400 home computer. It took weeks to get running and then amused me for maybe a couple hours.
The illusion didn't work because I knew how it worked. Its responses were so algorithmic, and it was so easily tricked I couldn't imagine anybody thinking ELIZA was 'thinking' about anything.
1/2
Replies (1)
-
@zazzoo@mstdn.ca 2026-09-17 14:39
Modern LLMs work differently than ELIZA, of course. In fact they work so differently that not even the people who train them really understand why they work so well - but the result is the same. A computer interacting with human language fools our social brains into thinking we're dealing with another human. We find ourselves saying 'please' and 'thank you' at it. The illusion is even stronger if you give them 'personality' and name. But it's all just ELIZA in the end. The illusion is broken if you know how they work. 2/2