ELIZA's creator saw the ChatGPT confessional coming in 1966

The pattern that most unsettled the inventor of the first chatbot in 1966 is now the everyday behaviour of anyone who has confided something in ChatGPT. A Wired excerpt from the new MIT Press book Inventing ELIZA revisits how Joseph Weizenbaum built ELIZA at MIT between 1964 and 1967, and how quickly he came to regret what people did with it.

The anecdote at the centre is his own secretary. She knew ELIZA was a program her boss had written. She sat down at the terminal and after a short time asked him to leave the room. Weizenbaum later wrote that he 'had not realized… that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.' The most famous script, DOCTOR, was little more than pattern-matched Rogerian therapist prompts bounced back at the user, and that was enough.

What Weizenbaum saw in one secretary now scales to every consumer AI product. The excerpt draws the line directly: modern systems like ChatGPT operate at vastly greater sophistication than ELIZA, yet they produce similar psychological effects, with users feeling heard and willingly sharing intimate thoughts with machines. Weizenbaum himself pulled away, publishing Computer Power and Human Reason in 1976 as a warning that programs mimicking understanding could replace genuine human connection.

The honest caveat is that this is a book excerpt and a historical argument, not new empirical work. It does not quantify how often modern users share sensitive information with chatbots, and it does not prescribe what the operators of those chatbots should do about it. The forward-looking read for anyone shipping a conversational product is straightforward. If your interface implies empathy, users will treat it as a confidant regardless of what the terms of service say, and that behaviour is now sixty years old.