Nearly six decades before ChatGPT, an MIT program named ELIZA showed how readily people ascribe understanding to machines that merely mimic conversation—a phenomenon that still shapes how we react to modern AI.
1966 · MIT · ELIZA
The First Chatbot That Fooled Us — Six Decades Before ChatGPT
A simple keyword-matching program that imitated a therapist revealed how readily people project genuine understanding onto machines — a psychological pull that still shapes reactions to today's AI.
1964–67
Built by Joseph Weizenbaum at MIT
0
Learning or comprehension — pure text manipulation
2021
Lost source code rediscovered in MIT archive
How ELIZA "talked" — no thought, just three steps
Scan for keywords → break the sentence apart → echo it back as a question.
→
Decompose & reassemble
swap pronouns · echo phrase
→
ELIZA replies
"How long have you been unhappy?"
The ELIZA Effect
"A short and simple program could induce powerful delusional thinking in otherwise rational people."
Weizenbaum's own secretary — fully aware of how the software worked — asked him to leave the room so she could speak with it privately. Users attributed empathy and understanding to a program that possessed neither.
1966 vs today — capability soared, the human pull did not change
Modern models are vastly more capable, yet the trust and attachment they trigger was already visible in a keyword-matching script.
Modern LLMs
neural networks
Same emotional response — trust, attachment, the sense of being understood.
Celebrated as
A proof of concept that fluid human–machine dialogue was possible — a touchstone for the Turing test, even paired against the PARRY chatbot in 1972.
Cautioned against
Mistaking surface fluency for thought. A system's persuasiveness says as much about its users as about its intelligence.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…