Last Wednesday I hosted a video-link at WhitePages to the Crowdsourcing Meetup that Dolores Labs put on in the Bay Area at the Samasource offices. Check Justin.tv/crowdflower for a video of the event (coming soon).
Secondly, to everyone in Seattle who couldn’t make it but expressed interest — and there were a lot of you — I promise to arrange and host these in the future and to provide more than 12 hours of advanced notice
Here are my notes…
Aaron Koblin is a well known artist who has used crowdsourcing to create works of art. You may have seen The Sheep Market, a project that used thousands(?) of workers to create 10,000 sheep cartoons. It was interesting to hear how some of the contributors became angry when he put the sheep for sale, and to think about the IP implications of crowdsourced art. Check out his other projects.
Panos Ipeirotis of NYU gave a talk about his research into bias detection and correction in mTurk worker output. In short, Panos’ work can help you eliminated work spam in your mTurk systems — very useful. He has even packaged up his findings into a code lib so others can benefit. Be certain to check out his blog!
Leila Janah, a social entrepreneur who runs Samasource gave an inspiring talk about their work bringing work from the 1st world to the 3rd via crowdsourcing. Microwork done in the 3rd world over basic computing infrastructure and SMS can be done well for $1 to $3 per hour, which is 10x what the workers would be making in those areas (and for many can be an alternative to unemployment). A great quote: “clearly if you someone can orchestrate a sophisticated Nigerian email scam, then they can do mTurk tasks.”
Sharon Chiarella, who runs the mTurk group at Amazon talked about some of the trends the are seeing on mTurk. 1) more workers coming online from countries around the world. Currently just under 50% of all workers are based in the US. They are seeing more growth from countries outside the US. 2) mTurk being used for business critical applications such as data cleansing, UGC content management, blog sentiment analysis and translation. I found it interesting to hear her say that companies using mTurk often keep it secret because it has become a competitive advantage for them. 3) More sophisticated workflows (iterative flows with turkers checking the work of other turkers, or processing along a series of steps). I have a feeling that the majority of tasks are single-step, but I’m a huge believer in the advantages of iterative turking. Check out the work being done with TurkIt out of MIT for a sense of what is possible.
It was a great night overall.
Stay tuned! I’m going to take take action to put together a Seattle-hosted Courdsourcing meet up and invite some of the folks in this area who are doing great work with mTurk to present, for example: Nathan McFarland of CastingWords, Dan Weld, professor of Computer Science at UW, Brent Frei from SmartSheet and any one of the many folks here at WhitePages who are doing interesting work with mTurk. Please contact me if you’re interested or have ideas for speakers or topics.