Amazon Mechanical Turk

Ideas for Amazon Mechanical Turk

This is IMHO, a groundbreaking capability that I ranted and raved about in the Web 1.0 days, noting that (1) it is a pretty big project to do, but (2) it basically is second to government in being able to tax elements of work at each step in the ‘value creation chain’. In Amazon’s case, they get 10% of any work effort. In any case, if this is done right and they don’t blow it, it’s part of the “bigger picture” and a fundamental change. Anyway, here are some ideas for the operations/”chunks” of work (feel free to add to):

• Book Chapter Audio Transcription: Read out loud and record this chapter.


• Blog transcription. Text to Podcast


• Does this image contain x? imageContains(“vehicle”, imageUri)


• Convert this outline/xml/yaml/json to a textual description


• paraphraseThis(text)Code Formatting/IV&V/Review:


• Does this code follow these rules. Here, the checks are broken into chunks and distributed, as is the set of code. It is broken up like a text message over UDP, distributed to lots of people (who get it into their queue and check it in one-five minutes for a few cents. It’s then returned, possibly packaged up as another workitem and returned.


• If language people are signed up, we could easily ask “what’s wrong with this bit of code” and get the hint that get us past it. (This one could also work more informally, with a ‘tip jar’ concept.
  


• Blog comment verification!

For a general description:
Amazon's Mechanical Turk