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, that way that we know not to walk off a cliff's edge, which is strangely hard to teach a computer. Whenever I see a system that purports to try to pass the Turing test, to try to trick you into not knowing if you are talking to a person or a machine, I always ask the question, "What's bigger, a nickel or the sun?" and I haven't found a system to answer that question.
Now, if I ask you that question, you know the answer immediately. And yet, how do you know that? For instance, you don't know if I said the word sun, s-u-n, or the word son, s-o-n. But then at some level in your head, the fact that I said "the" sun, a sun of which there is only one. In addition, the nickel and the sun both have this property of being round, which at some level instantly kind of puts you in a place of knowing that's the comparison we're drawing. Comparisons are often drawn between things that have both like and dissimilar characteristics.
So the question I want to leave you with is how did you know what kind of sun I was talking about, and for that matter what kind of nickel. Was I talking about nickel the metal or a nickel as in five cents? The fact that I said "a nickel" instead of nickel without the word "a" is a clue. But you probably didn't have to parse all of that out. You probably instinctively knew when I said a nickel or the sun your mind was immediately able to categorize those two things correctly and come up with the answer, but we can't teach computers to do that yet. And that's an impediment to being able to build a general intelligence.