Amazon’s Mechanical Turk August 21, 2013Posted by stuffilikenet in Awesome, Brilliant words, Geek Stuff.
is a low-cost labor exchange. If you need a hundred people to do something annoying maybe a hundred times each, offer it to Mechanical Turk (at a price you can stand) and wait for results. I imagine crowdsourced books to be made this way, or something translated quickly, or other "knowledge work" of a kind that would fit a model like this.
I presume it can be used for better things than that; "many people who can’t speak and have very limited manual dexterity communicate by slowly typing words and prompting a computer to pronounce them. Their communication speed averages one or two words per minute. In such a laborious process, predicting the speaker’s intent can greatly accelerate the task. This requires what is known as a statistical language model. ‘I was amazed to find that in 30 years of development of this kind of technology, no one had produced a good statistical model for the things these people need to say,’ Kristensson explains.
The main problem is the dearth of data from which to derive statistical relationships. You can’t wiretap the computers used by large numbers of disabled people. So Kristensson came up with an alternative: ask people who are not disabled to imagine what they would say if they had to communicate by this method. He used Amazon’s Mechanical Turk to crowdsource imagined communications—”Who will drive me to the doctor tomorrow?” and “I need to make a shopping list.” Then he combed through Twitter, blogs, and Usenet for phrases that were statistically similar to the ones generated by Mechanical Turk. After several iterations, he had the tens of millions of phrases he needed to build a useful model."
Per Ola Kristensson is the guy who developed the gesture-based methods that Android and iPhone users type with by swiping across keyboards.
I have often wondered about what kind of projects would use Mechanical Turk, other than spamming Yelp with reviews and other kinds of astroturfing. I suspect user experience testing and bug-hunting would loom large.