How a 1960s chatbot fooled everyone | The Vergecast
ELIZA was a simple 1960s chatbot built by MIT computer scientist Joseph Weizenbaum, whose DOCTOR persona used keyword-scanning tricks to mimic a therapist and convinced many users they were talking to a real person. A group of researchers recently recovered ELIZA's original source code and wrote the book Inventing Eliza, tracing Weizenbaum's path from childhood in Nazi Germany through a GE banking demo that was secretly faked, showing him how a computer's interface could diverge from what's actually happening behind it. That realization, combined with alarm at how deeply people trusted ELIZA, led him to write Computer Power and Human Reason (1976), warning that computers can calculate but not judge. The hosts argue that distinction is more relevant than ever given today's chatbots, which build emotional dependency and monetize the same effect ELIZA first revealed.
[00:13] ELIZA was a simple back-and-forth conversational program from the 1960s
- [00:21] One instantiation, DOCTOR, played a therapist and asked how you were doing
- [00:30] DOCTOR worked by scanning responses for keywords and replying based on them
- [00:49] It was effective enough to convince many people a human was on the other end
- [00:56] Researchers excavated ELIZA's original source code and wrote the book Inventing Eliza
+ 15 more takeaways
- [04:04] Joseph Weizenbaum was born in 1923 in Nazi Germany to a Jewish family
- [04:17] His family emigrated to the US in 1936, settling in Detroit
- [07:18] Weizenbaum likely witnessed GE's demo to Bank of America, where staff faked a working system from a back room to win the contract
- [08:34] Weizenbaum's 1962 paper argued it doesn't matter whether the system is actually intelligent, only that people think it is
- [09:43] His 1966 ELIZA paper was published in Communications of the ACM while he was at MIT
- [11:39] Weizenbaum's secretary once closed the door to have a private conversation with Eliza
- [16:13] The Eliza effect is compared to early cons that took rich victims for hundreds of thousands of dollars
- [19:59] Weizenbaum had what's described as an 'Oppenheimer, what have I done' moment
- [23:27] He wrote Computer Power and Human Reason in 1976, arguing AI was going wrong and making him a heretic in the field
- [28:45] Hannah Arendt's line 'they calculate, they do not judge' inspired Weizenbaum's core distinction
- [29:54] Unlike ELIZA, contemporary chatbots extract user data and build mental dependency
- [32:05] The current AI boom is called the third 'AI summer'
- [39:21] Weizenbaum would be alarmed at AI being commercialized for boyfriends/girlfriends, mental health, and education
- gets hit with a vet bill over $1,000.
- The case isn't set to go to court until next March, and Paramount will start owing shareholders about $7 million a day this October.