Geoffrey Hinton left his position at Google last year so he could speak more freely about artificial intelligence, a field he helped to pioneer and is increasingly critical of. He spoke with The Globe about where he believes AI is headed.As researchers at the world’s most advanced AI companies prepared to sign a public letter last week warning of frightening new developments in artificial intelligence, Geoffrey Hinton, the godfather of AI, emailed to ask if I wanted to have lunch.
One of the tech bros planned to have his body cryogenically frozen at a cost of US$250,000, to be thawed and reanimated in a more technologically advanced future. Another entrepreneur was going the head-only route, at half the price. For emotions we have to distinguish between the cognitive and the physiological. Chatbots have different physiology from us.
But if you put the pieces in a pile, its vision system wasn’t good enough to see which pieces were which. So it smashed the car, and the pieces were scattered. Then it could do it. Obviously it didn’t experience the physiological aspects of being angry. But it was a very sensible emotional response. There’s something you don’t understand, so destroy it.I didn’t say it was incapable of that.
Right now, the main tools we have, other than just stopping doing this, are to curate the AI training data. It’s crazy what we’ve been doing, training AI using everything on the web. As a result these big language models have a huge number of possible personas. After they’ve read a little bit of a document, they sort of take on the character of that document. Then the bot starts thinking like that document, so it can predict what comes next.