Jeff Atwood has an post worth reading about what he views as the failure of Amazon’s Mechanical Turk service, a service that’s baffled me ever since I saw it spring up. (For those who haven’t heard of it, Mechanical Turk is a clearinghouse set up by Amazon for organizations to solicit assistance in completing rote tasks, paying people a certain amount per task. For more background and info, Wikipedia has a reasonably good article about the service.) To be clear, I understand the idea behind the service — there are certainly a bunch of things that pop up in everyday work life that are worth hiring someone (in effect, a short-term contractor) to help you finish — but every time I browse Turk, it seems that there’s a vast disconnect between the available tasks and the amount people are willing to pay to get them completed.

For example, as of this morning, the Missouri Department of Purchasing and Materials Management has a Mechanical Turk post requesting assistance extracting details from around 250 state purchasing contracts. To complete each of the 250 tasks, the user has to:

  • search a web-based Missouri contracts database for a specific contract number;
  • visually locate a few fields on the detail page for the contract and cut-and-paste the information into a Turk form;
  • download and open up the actual contract (sometimes a Word document, sometimes a PDF) using a link on the contract detail page;
  • manually search through the dozen-plus pages of the contract for a bunch of other details, and cut-and-paste them into the Turk form;
  • cut-and-paste all the various document links from the contract detail page over to the Turk form;
  • and finally, submit all the extracted data back to someone in Missouri.

Once all that is done, someone (ostensibly from the Missouri Department of Purchasing and Materials Management) then evaluates the submitted information before agreeing to pay the Turk user the piecework fee — which is a whopping ten cents. And the funny thing is, this isn’t some isolated case; browsing the available tasks page, most are asking for someone to do something reasonably time-consuming, and are willing to pay reasonably little… and the fact remains that payment of the worker is still at the sole discretion of the person requesting the work. It’s hard for me to understand who’d be willing to participate in the service, and I’d love to see someone take a more longitudinal view of the posted tasks and provide real stats on such things as how many tasks get completed, how many users end up getting paid, and what kind of money ends up moving through the service over a given increment of time.