mechanicalturk

Mechanical Turk

ETech has been over for a week, and one presentation is still nagging at me on a regular basis. Amazon has a Web Service called Mechanical Turk (named after this Mechanical Turk), and Felipe Cabrera from Amazon spent 15 minutes or so talking about MTurk during one of the ETech morning talks.

The talk focused on the idea that artificial intelligence hasn't materialized, and there are still some tasks that are easy for humans but impossible for computers. For example, a human can look at a picture of a chair and answer the question: Is this a picture of a chair or a table? A computer would have a tough time with that.

MTurk farms out these sorts of questions to real live humans and wraps their decisions (or HITs in MTurk parlance) into a Web Services API so they can be used in computer programs. Cabrera called this process of tapping humans to make decisions for machines Intelligence Augmentation (IA) as apposed to Artificial Intelligence (AI). The talk was good, and MTurk is definitely a clever hack, but the idea has been bothering me.

I can imagine a world where my computer can organize my time in front of the screen better than I can. In fact, I bet MTurk will eventually gather data about how many HITs someone can perform at peak accuracy in a 10 hour period. Once my HIT-level is known, the computer could divide all of my work into a series of decisions. Instead of lunging about from task to task, getting distracted by blogs, following paths that end up leading nowhere, the computer could have everything planned out for me. (It could even throw in a distraction or two if that actually increased my HIT performance.) If I could be more efficient and get more accomplished by turning decisions about how I work over to my computer, I'd be foolish not to.

I guess this idea of people being managed and controlled by machines is nothing new, and it was the bread and butter of science fiction books I read as a kid. But MTurk puts this dystopia in a new, immediate context. Machines are smarter than ever, and control of human decision-making could be highly organized.

MTurk is only a few months old, and there's nothing inherently wrong with it. But I can't stop projecting the ideas behind the system ahead a few years, and that's what's bothering me. I can't even fully articulate why it's bothering me. I don't have any conclusions, or even concrete hypotheticals of MTurk gone awry—so I'm just using my blog as therapy. Obviously my computer didn't ask me to write this.