People get very excited every time there’s a tiny advance, but the tiny advances aren’t getting us closer. There was a Google captioning thing that got a lot of press. I think it was the front page of The Times. You could show it some pictures and it looked like it was great. You’d show it a picture of a dog, a person, and a Frisbee and it might be able to say, that’s a dog catching a Frisbee. It gives the illusion of understanding the language. But it’s very easy to break these systems. You’d show it a picture of a street sign with some stickers on it and it said, that’s a refrigerator with food in it. This is the kind of bizarre answer that used to send you to Oliver Sacks. It’s almost like a neurological deficit. The systems will be right on the cases that they have a lot of data for, and fall apart on the cases where they don’t have much data.

You can contrast this with a human being. You’ve never heard any of the sentences that I’ve said today—maybe one or two—and yet you can understand them. We’re very far from that.