Mechanical Turk

There was an article in today's New York Times on the search for Steve Fossett, which unfortunately has not yet yielded positive results.  A similar search for Jim Gray had also turned up negative, and both are indeed very tragic cases of missing pioneers.

The article's focus was Mechanical Turk, a fascinating service by Amazon (I continue to be be very positively impressed by Amazon's innovations outside their core retailing operations, including the whole slew of Amazon Web Services) that uses the power of human beings to solve problems (such as finding wrecks in images) that are computationally more difficult for machines.  Their approach, "I will pay $.50 if someone does x" (called HITs, for Human Intensive Tasks) is a great way of combining the best of man and machines.  In some ways, it is a logical extension to the whole Web 2.0 phenomenon, where social capabilities enhance machine capabilities (e.g. tag cloud enhances (replaces?) formal, stuffy, taxonomies).

Pinos Ipeirotis (whose bogs I folow closely) has written a very informed piece on his experience on the quality of submissions (responses) to HITs requests.  As a computer scientist, he ran controlled experiments and summarizes some of the results there.  Go check it out.

What other such capabilities exist out there?

Comments

[Joscha Feth]

Have you seen reCaptcha?
Also a very nice thing to solve turing-resitant tasks by letting humans do it:
http://recaptcha.net/learnmore.html

[Joscha Feth]

Oh, and of course the Google Image Labeler: http://images.google.com/imagelabeler/

By the way also from Luis von Ahn.