Souls of a new machine

ONE OF THE great broken promises of the 20th-century view of the future, right up there with personal jet-packs, was the promise of artificial intelligence. AI was supposed to lead to computers that wouldn't just calculate and organize, but reason and analyze; computers that could really think, like HAL in "2001" or KITT on the 1980s TV show "Knight Rider." (Of course, HAL turned out to be a homicidal psychopath and KITT was a smug know-it-all, but still, it seemed like a good idea.)

Recent efforts to realize the promise of AI have centered on teaching computers to better deduce meaning from the vast content of the Web, but there's still a long way to go. In the meantime, however, there's an alternative type of computerized system that is actually making big strides toward getting computers to think like humans. Publisher Tim O'Reilly calls it intelligence augmentation (IA for short), and it uses a very clever technique. It cheats.

IA is a catch-all term for a wide variety of methods that use actual human beings, with actual human brains, as part of computer programs. The idea is that by having a human deal with the specific parts of a problem that are difficult or impossible for a computer, but trivial for you or me, you can have a program that seems to possess real human intelligence. In the last few months a number of examples have appeared:

ª Mozes Mob (mozes.com/mob.php) is a free, cellphone-based service designed to answer questions. To see how it works, text message a question that a human could answer easily but a computer program would have a hard time figuring out (such as "Is the weather nicer in Miami or Buffalo?" or "What was Carlton Fisk's most famous home run?") to 66937. Behind the scenes, your question is sent to a swarm of volunteers. One of them answers, and their response is bounced back to your cellphone, usually in just a few seconds.

ª The Google image labeler (images.google .com/imagelabeler) is an addictive online game that takes advantage of the fact that it's very easy for a human to recognize the subject matter of an image ("That's a puppy!" "That's two airplanes and a bird!") but virtually impossible for a computer. The game teams up pairs of strangers to assign keywords to images as quickly as possible. The more images you can label in 90 seconds, the more points you get. Meanwhile, Google gets hundreds of people to label their images with keywords, something that would be impossible with just computer analysis.

ª While Mozes and Google image labeler depend upon the kindness of strangers for their intelligence augmentation, Amazon has a service driven by cold hard cash. Amazon's Mechanical Turk (mturk.com) is a system that lets people with specific types of tasks hook up with those willing to perform those tasks, at an agreed upon price. Recently on the Mechanical Turk someone offered $1.53 to anyone who would transcribe a 9-minute podcast and someone else offered a nickel for identifying the title of a snippet of music. Think of it as a kind of eBay for buying and selling brainpower.

Intelligence augmentation has a lot in common with what O'Reilly calls the online "architecture of participation": Thousands of people coming together to create Wikipedia, for example. But IA takes things one step further. Here the human knowledge isn't front and center the way it is in Wikipedia; it's automated away behind the face of the computer program, a tiny human cog, deep inside the machine.

Chris Spurgeon lives in Los Angeles, where he works as a Web developer and designer, writes his blog spurgeonworld.com, and studies obsolete technologies.