Will work for books–Amazon Mechanical Turk

Amazon’s developed an extremely clever–but kinda strange–interface to match humans with little mundance tasks they say machines can’t do. Called Mechanical Turk, the service asks you to sign up and then get paid–pennies, mostly–to do all sorts of things, like correct A9 mapping photos.
A typical task right now is for A9 and reads as follows:
“You are presented with the name and address of a business as well as a set of photos taken along the street where the business is supposed to be located. Your task is to identify the best photo of the business that is listed.”
For that, you get .03; if you do it 179 times, you get 5.37.
Amazon’s estimates work so that if you have spent an hour to fulfill 1170 of these requests, you get $35.00, which means your time is worth $35 an hour if you can fulfill about 1 per second–if it takes you 5 seconds to fulfull each task, your time is only worth $6-7.00.
Amazon says that they take a slice of each transaction and that workers can transferr money to their U.S. personal bank account or to their Amazon.com account.
The most amazing thing about this, IMHO, is that the Amazon guys let the Web Services team put this up–which means they had to run some use case scenarios on what it would save them in terms of hiring freelancers or outsourcing company wide.
Yeesh.

Latest Comments

  1. bleh says:

    Most discussion seems to be focussed on the viability of the slave [non judgemental term for worker] in this case here.
    Wondering about the issues for the buyer [master] here – first off, the buyer needs to manually approve the submission on each HIT – otherwise it would be pretty simple to automate the process of sending fraudulent submissions.
    And the above requirement pretty much kills the possibility of automating the process of submission of HITs at the buyer’s end.
    Its tough to buy the idea that human beings can now be harnessed to perform monotonous tasks – monotonous tasks are really tasks that the buyer doesnt want to do by themselves – but if they have to approve the submissions, thats pretty much the same thing. Unless of course they come up with more sophisticated ideas (random checking of submissions to HITs for instance)
    Some speculation ahead –
    Another possibility that we’re looking at is perhaps more specific – connecting people who have some specific talent/knowhow to people who require that talent/knowhow – especially when that talent/knowhow isnt worth that much ($0.20 being cited as an example price).

  2. susan mernit says:

    Bleh–spot on–I was thinking this could be conceptualized as a form of elance.com–a platform for a labor marketplace for micro-tasks–but these are tasks many people would have a student do locally, if possible (not possible for A9)–I am curious if this Amazon thing will go anywhere-or what it will evolve into (a labor market for tasks worth .03 cents each?)

  3. bleh says:

    I didnt give this further thought until you brought up the above question…but a few interesting possibilities occured to me…
    1) Labour like the third world manual labour (packing matchboxes for instance) – this is a unique way to give them opportunities, the incentive to the buyer being the low skilled repetitive labour, the skillset still being marginally above those of an automated system.
    2) Spam! Imagine google’s blog systems using people to weed out spam blogs for instance. This would depend heavily on a component I just thought of…below at the ***
    3) Optimizing a system which adapts to user feedback. Rather than just using user feedback per se, you use worker feedback. For instance, a search engine might use worker feedback to modify the results it provides – this is useful since normal users lack the incentive to provide feedback, they just move to a competitive service or try a different set of keywords.
    *** This component could be applied to any such system!
    You could match up the results from 2 independent workers to eliminate errors. And to verify quality, do a random check of 3-4 in a 100. As workers keep their quality levels over time, the random checks could be done less and less often.
    And you know what? You could have the workers with the highest trust levels doing the random checks..that way the system is self sustaining!
    With a large enough set of workers, it becomes tough to *game* the system, and those who mess up would be very quickly weeded out of the system..they’d lose trust pretty rapidly. All that it takes is to inspect a workers total output when she/he makes a mistake, and you’ll see very quickly whether the quality of that output is good or bad.
    Lots of possibilities…the idea in its current shape may not go very far, but then I’m pretty sure it’ll break out of its current shape soon :)

  4. philippines peter says:

    I saw this article and right away signed up. The jobs didn’t seem interesting enough for me, but I also did an experiment with hiring someone to do a few trivial tasks. I have no idea who did it, but it was well-done. Although to be completely honest, I’m short amazon stock, I think this service is a good one and has a bright future. Here in the Philippines it’s unbelievable but educated people sometimes are working for just five dollars a day – people who can speak English and use a computer@! Perfect for the Philippines

Latest Comments

Comments are closed.