The people in the machine, the humans in the app, and the metaphorical import of the Mechanical Turk look to be one of the big new buzzes. Felipe Cabrera of Amazon demonstrated their Mechanical Turk– a human API. Just as API libraries allow programs to make calls for data, their API allows programs to make calls for real-life human processing.

CastingWords.com, for example, uses the API to hand-off podcast transcription jobs to the human processors and receive the information back, taking human management out of the process except for quality assurance purposes. This new ability brings people to podcasting by making it available in a format more amenable to them. Interesting side-effect: a chemistry prof discovered that his students were performing the transcriptions of his podcasts through CastingWords… essentially being paid to study.

He showed how the same tech was used to cleanup hundreds of thousands of images, handing them off to human workers for touching-up and evaluation.

Again, it’s about pattern recognition and capabilities. Computers are great at some tasks, horrible at others. Felipe showed a picture of a chair and asked: is it a chair or a table? We can tell easily– it’s incredibly hard to instruct a computer how to do this (and not fiscally productive). The Amazon Turk enables “massively parallel human computing.”

Ultimately this means developers can solve problems that need human intelligence, businesses can harness a distributed workforce, and anyone can make money with their skills on their schedule.

[ruminate etech etech06 aai mechanicalturk]