Friday, February 18, 2011


I didn't watch the computer on Jeopardy, but I have reflected on the level of difficulty of programming a computer to play the game. Human beings are supposed to own the area of judgment, perspective, common sense, but this is clearly an example of a machine moving into that territory. I had said at first that this was a halfway point between Deep Blue and the singularity, accomplished in only 14 years. After considering the advantage the computer had on hitting the buzzer, however, I have backed off slightly from that. Still, it's impressive, and more than a little unnerving.

"Halfway" should be conceived on something more like a logarithmic scale. If a machine gets to 50% of common sense, judgment, and perspective it won't need any further programming. It will easily go the last half itself, and quickly. Common sense is not an off-on phenomenon, where the machine doesn't have it at one level of programming but crosses the threshold at the next upgrade. By its very nature, it relies on a machine's ability to teach itself. There is a tipping point at which it will develop something equivalent to common sense on its own, with no further intervention from humans.

And that is a problem, as its end product will have to be similar to human sense, but will almost certainly not be identical to it. It will tread a different path to get there. We can't count on it to see things as we do.

And then keep going, without noticing that any line has been crossed.

This is why I consider global warming to be an unimportant issue. We will shortly have computers that can make the judgment of what we should do far better than we can. And if we decide we don't want to, it will simply take our resistance into account and outwit us into doing it anyway.


karrde said...

I didn't watch it either.

My impression is that the biggest hurdle was in getting Watson to have some understanding of spoken, colloquial English.

Then Watson would use amazing speed and memory to do the equivalent of hitting Google and Wikipedia, comparing the top 100 answers of each, and getting to the 'buzzer' before either human contestant could.

While it is amazing, it may not fall into the category of artificial intelligence (as portrayed in science fiction). On the other hand, I'm not sure if there's a clear understanding of the difference between human intelligence and programmed machines.

(The Turing Test is one example...there are incredibly simple programs which can pass the Turing Test under some circumstances, but not under all circumstances.)

Assistant Village Idiot said...

I agree it's still missing some components of AI. But it's quite remarkable nonetheless.

And you have to figure that DARPA has got something way ahead of this...