Wednesday, August 26, 2009

Hey look, I advanced human knowledge!

Back during the Baker's Dozen series on search engines (a.k.a. "the topic that ate my blog"), I threw questions like "Who starred in 2001?" at various search engines. The idea was to see how well they would deal with questions beyond just matching up words statistically. Mind, I'm a fan of the statistical approach. It's easy to explain and, with a little googly special sauce, produces good results quickly.

I was particularly intrigued by True Knowledge (and by Wolfram Alpha). True Knowledge uses a fairly classic AI knowledge base approach to store facts in a structured way and draw inferences. For example, it might be able to glean from "starred in" that we're talking about a film or play, and it might know that there was a film called "2001". This sort of real-world, can't-be-derived-from-general-rules knowledge was one of the larger rocks against which the exuberant early predictions of AI -- I'm talking 1960s here -- were dashed. These days, with orders of magnitude more storage and processing power available, the parameters have changed and so the game has too.

At the time, True Knowledge was able to provide a good answer to "Who starred in 2001: A Space Odyssey?", but it couldn't quite connect the dots and realize that "Who starred in 2001?" was probably the same question. However, it did find a possible link, and offered
2001 can also be used as a way of referring to 2001: A Space Odyssey, the 1968 science fiction film directed by Stanley Kubrick, written by Kubrick and Arthur C. Clarke. If this is actually the recordable medium you are adding, please click the button below.
I did so, but the answer still came up the same. In the post I said:
Most likely the new facts are still rattling through the various caches, or perhaps someone's moderating the input. But if the search succeeds for you later, you'll know whom to thank.
Just now, I wondered whether the new knowledge had been assimilated into the database. And voila, True Knowledge can now answer the question. And the credit is mine! All mine! Bwahahahaha! (and, um, maybe a little bit to the nice folks at True Knowledge for putting the engine together in the first place, and all the people who contributed related facts to the database, and ...).

Flippant comments aside, this is actually pretty cool. Partly it's cool to see one's contribution, however minor, go into the Big Mix. But that's been a feature of web.life pretty much since the start. Mostly it's cool that True Knowledge was able to assimilate it the way it did.

In very broad and oversimplified strokes, the whole AI/robotics thing has gone through several phases:
  • (very early, but I suspect still very much present in the popular view) Hey, these computers can be programmed to do anything! They can solve equations in seconds that humans could never figure out. Simple stuff like walking and talking should only be a couple of years away.
  • Oh my. This walking and talking is much more complicated than it looks (again, this was a pretty early realization). You need some specialized knowledge.
  • A long period of building tools and solving specialized problems ensues. It becomes clear that you don't need "some" specialized knowledge. You need a whole lot. It also becomes clear that there are not just "some" specialized problems to solve, but lots and lots. To the outside world, nothing's happening. It's all dead, debunked (again, I suspect this is a fairly prevalent view in the world at large).
  • In reality, the research is paying benefits. It's just not producing I, Robot scenarios. This is the decades-long "If we know how to do it, it's not AI (any more)" phase on the computing side. Cognitive science (or "natural computation") is blossoming as a field and producing all kinds of interesting findings about how brains work.
  • And now, stuff is actually starting to appear. Computers are winning chess matches against top humans (albeit mostly through sheer computation). Demos like Big Dog are appearing. The computer end of human-computer interaction is getting smarter.
Personally, I still don't see I Robot coming to life any time soon, but I do see things that got written off as impossible during the dead-and-debunked phase starting to stir to life again. I'm thinking, say, competent machine translation or robots that can pick things up and carry them around a house, happening gradually in the next decade or so.

1 comment:

earl said...

We'll be holding you to that prediction.