Man and machine

Following Tuesday’s presentation on the Future of Interfaces at the O’Reilly Emerging Technology 2006, discussion became a bit more concrete and immediate with a session about the relationship of human and computer intelligence. Felipe Cabrera of Amazon Web Services gave an interesting presentation on what he called “Artificial Artificial Intelligence,” which translates as human intelligence.

Basically, this allows for programmers to use regular people to answer questions about data that may be near-impossible or very complex for computers to answer. An example of this would be to look at a photo and determine if there was a wooden chair present or not. Amazon has called this Mechanical Turk, named for the 18th century hoax that purported to be a chess-playing mechanical device.

Just as the original European version had an actual man inside the box that was a “thinking machine,” Amazon uses actual people to perform intelligence gathering. In order to take advantage of these people, you just go to Amazon’s MTurk Web site where you can view available Human Intelligence Tasks. At the time of this writing, there were eight available, ranging from French translation to taking photographs of real estate in Brooklyn. Regular people can go to the MTurk Web site and be paid for their work.

Cabrera said that the Mechanical Turk system would be advantageous to programmers, in the sense that they could simply write a variable (like the one below) to retrieve the MTurk data, automatically adding it into their program. Once that data was collected, the program would simply continue.

	read (photo);
	photoContainsHuman = callMechanicalTurk(photo);
	if (photoContainsHuman == TRUE) {
	   acceptPhoto;
	}
	 else {
	   rejectPhoto;
	}
	
However, at this early stage, there remains one looming problem to the Mechanical Turk. While asking a person to figure out whether or not there is a human present in a photo is a trivial task and can probably be achieved in a minute or less, translating a document into another language is not. Currently, the listings on the Web site are fairly complex and don’t have payment greater than $5. Given that there is such a low payment, it surprises me that MTurk is getting any users at all. Cabrera offered CastingWords.com as an example of a company that seems to rely on Mechanical Turkers. CastingWords is a new company that offers to provide transcripts of podcasts within 24 hours at 42 cents per minute of audio. How CastingWords is making any money remains to be seen.

Nonetheless, perhaps Mechanical Turk simply needs time to mature. Or maybe it will be exposed as bogus, as the original one was.

  
Shop Tech Products at Amazon