Gigaom AI Minute – January 24

Transcript

There's a narrow AI, which is an artificial intelligence that can do a certain task. And then there's a general AI, which is one that can do things it hasn't been trained on, and that would be like what you see in the movies.

And I'd like to pose the question of: Can you just build 10,000 narrow AI's, or 20,000, or 30,000, or whatever, and essentially bolt them all together and effectively have a general intelligence? In essence, that's a lot of what we do now.

You take something like IBM Watson when it played Ken Jennings in Jeopardy. It was trained on all kinds of questions; it was, in essence, several different narrow AI's packaged together to do this one thing. So we will no doubt do that. We will no doubt, in domain area after domain area after domain area, master it and then add that to a large corpus. You could see a robot being able to vacuum, eventually being able to sweep, eventually being able to wash windows, and what not. But that never actually gets you to a general intelligence.

The thing about humans is that, once we know the 10,000 things, we can kind of do the 10,001st one without anybody teaching us how to do it. We know how to derive all of the information that we would need from all of those other actions and skills that we have. We don't know how humans do this, but it's clear that a general intelligence will have to be more than just a bunch of narrow intelligences bolted together.