The machine that listened too well

In 1966, a program of barely a few pages of code imitated a psychotherapist. Its creator was alarmed to discover that people were telling it their whole lives.

It was called ELIZA and was written by Joseph Weizenbaum at MIT. Its trick was almost a parlor game: it detected keywords and handed your own sentences back to you turned into questions. If you typed "I'm sad about my mother," it would answer something like "tell me more about your mother." It understood absolutely nothing.

And yet it worked all too well. Weizenbaum recounted that his own secretary, who knew perfectly well how the program was built, one day asked him to leave the office so she could talk to ELIZA alone. That changed him: he became one of the first great critics of artificial intelligence. Today we call our tendency to attribute understanding to machines that only imitate the "ELIZA effect." Sixty years later, it's worth remembering.

Source: MIT