Thursday, September 07, 2006

Testing Turing.

An eventful summer is over. Labor Day was last Monday, so the kids are back in school and the white shoes are back in the closet. It's time to get to work.

The most interesting thing to cross my radar this week was an article by Tim over on the O'Reilly Radar. It's about Google's new Image Labeller project. When I saw the Labeller, I thought it looked a lot like the Mechanical Turk, but Tim -- thanks to Luis von Ahn at CMU -- had a better insight.

Like the Turk, the Labeller is a way for computers to use humans for computation. Tim points out other places where this happens -- CAPTCHAs, for example.

Then he reminds us of the Turing Test, and makes this fascinating observation:

This is an interesting variation on the Turing test, in which humans generate and grade tests that most humans can pass, but current computer programs cannot pass. Is there another variation in the future, in which computers generate and grade tests that computers can pass, but humans cannot pass?

I've been thinking about it, and I've decided that the fourth variation -- human testers, humans fail, computers pass -- is isomorphic to the regular Turing Test. It would, however, allow us to recognize the androids reliably in Blade Runner.

The idea of computers using humans for computation reminded me of George Dyson's talk at Google.

Interesting stuff!


Post a Comment

<< Home