Gigaom brings you our unique analysis and commentary on the present and future of AI.
I'm spending some AI Minutes talking about common sense, how it is that humans know not to walk off the edge of a cliff. But that's hard to program a computer why that's bad. Let me talk about the Sally Ann test. It's something that you show children--you show them two dolls, Sally and Ann. And Ann has a basket and in the basket there's a marble. Both dolls, quote, "see" the marble put in Ann's basket. Then Sally leaves and goes for a walk. Sally is taken out of the room, and the child sees you take the marble out of Ann's basket and put it some other place. Then the Sally doll is brought back in, and the child is asked, "Where does Sally think the marble is?"
Well, an adult human has no problem answering that question--Sally still thinks it's in the basket that Ann has. But we know that it isn't, that it's some place else. We can maintain both of those perspectives. It turns out that children under the age of four have a hard time passing the test and children over four pass it easily. Why is that important? It could be a key aspect of consciousness. This doesn't explain why it is that we're conscious or how consciousness comes about, but it could be that a benefit of consciousness is that you are able to see yourself, and put yourself in someone else's shoes and see that they see the world in some way that's different.
You can imagine how useful that would be. I don't see anything behind the rock, but Og over there, standing beside the rock, can see something different so I should ask Og. You can see how that would be some kind of an advantage to people. It also might be a real key constituent to intelligence itself. Computers can be forced to do this. You can have things like, in programmed memory, Sally's knowledge and Ann's knowledge and your knowledge and there's the marble and its location changed and so forth. You can code that, but it isn't something that seems to be natural, obviously, to machines the way it comes naturally to humans. And that ability that we acquire around the age of four may be a key part of why we’re intelligent and a key understanding we need in order to build a general intelligence.