Why the singularity is greatly exaggerated

Ken Goldberg, Professor of Industrial Engineering and Operations at the University of California, interviewed by Jeanne Carstensen.

In 1968, Marvin Minsky said, “Within a generation we will have intelligent computers like HAL in the film, 2001.” What made him and other early AI proponents think machines would think like humans?

Even before Moore’s law there was the idea that computers are going to get faster and their clumsy behavior is going to get a thousand times better. It’s what Ray Kurzweil now claims. He says, “OK, we’re moving up this curve in terms of the number of neurons, number of processing units, so by this projection we’re going to be at super-human levels of intelligence.” But that’s deceptive. It’s a fallacy. Just adding more speed or neurons or processing units doesn’t mean you end up with a smarter or more capable system. What you need are new algorithms, new ways of understanding a problem. In the area of creativity, it’s not at all clear that a faster computer is going to get you there. You’re just going to come up with more bad, bland, boring things. That ability to distinguish, to filter out what’s interesting, that’s still elusive.

Today’s computers, though, can generate an awful lot of connections in split seconds.

But generating is fairly easy and testing pretty hard. In Robert Altman’s movie, The Player, they try to combine two movies to make a better one. You can imagine a computer that just takes all movie titles and tries every combination of pairs, like Reservoir Dogs meets Casablanca. I could write that program right now on my laptop and just let it run. It would instantly generate all possible combinations of movies and there will be some good ones. But recognizing them, that’s the hard part.

That’s the part you need humans for.

Right, the Tim Robbins movie exec character says, “I listen to stories and decide if they’ll make good movies or not.” The great majority of combinations won’t work, but every once in a while there’s one that is both new and interesting. In early AI it seemed like the testing was going to be easy. But we haven’t been able to figure out the filtering.

Can’t you write a creativity algorithm?

If you want to do variations on a theme, like Thomas Kinkade, sure. Take our movie machine. Let’s say there have been 10,000 movies — that’s 10,000 squared, or 100 million combinations of pairs of movies. We can build a classifier that would look at lots of pairs of successful movies and do some kind of inference on it so that it could learn what would be successful again. But it would be looking for patterns that are already existent. It wouldn’t be able to find that new thing that was totally out of left field. That’s what I think of as creativity — somebody comes up with something really new and clever. [Continue reading…]

Print Friendly, PDF & Email