Gigaom brings you our unique analysis and commentary on the present and future of AI.
I'm spending a few AI Minutes talking about common sense. Common sense is something that we humans do very well at a basic level, like you don't walk off a cliff. But it's something we don't really know how to teach computers to do. Now, if I gave you two sentence fragments, like "something something something swam to the bank" and "something something something robbed the bank," you wouldn't have any trouble understanding that in the context of those two sentences, "bank" means something different, or likely means something different in each of those sentences. But why is that? How do you know that right away? You associate the word "swam" in the sentence and you immediately start thinking, okay there are different kinds of banks, but you probably didn't have to go through that in your head. You probably instantly visualized something swimming to the bank or someone robbing a bank, and not have to go through a series of definitions and decide which definition is most likely to fit in this situation. That almost instinctual absent-minded selection of the context of the sentence, a way to understand it without having to say "can rob mean different things" or "can swam mean different things" and "can bank mean different things"? That is, again, something that humans are quite good at, but we're having a great deal of difficulty teaching it to artificial intelligence.