Crowdsourcing is a technique for farming out labor-intensive tasks over the Internet by splitting them into small chunks that dozens, hundreds or even thousands of people complete at their desks for a few cents each.
Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory are developing a new database system, called Qurk, that will automatically crowdsource tasks that are difficult or impossible to perform computationally. Images stored in a standard database system, for example, could be sorted according to date of creation or some other data tag, whether applied automatically or by hand. Images in a Qurk database, however, could be sorted according to the approximate age of the people depicted, or the appeal of the depicted locations as travel destinations, or any other attribute whose assessment would require human judgment.
In a pair of conference papers last year, the researchers described and demonstrated Qurk’s general computational framework. In a new paper they’re presenting this month at the 38th International Conference on Very Large Databases, they get into the nitty-gritty, describing a series of experiments on how best to crowdsource the common database operations “sort” and “join.” The researchers found that, using the most obvious implementation of the join operation, it cost $67 to combine two sets of images through Amazon’s Mechanical Turk crowdsourcing service. With an improved implementation that they arrived at experimentally, they could get the cost down to $3. Via Making crowdsourcing easier.