Comment: On Crowdsourcing and Human Computation

Is it possible to build a computer in which the processing units are human brains rather than silicon chips?

That was the big picture behind the Workshop on Crowdsourcing and Human Computation, held yesterday here as part of the CHI Conference on Human Factors in Computing Systems.

The thinking goes something like this: Outsourcing platforms like Mechanical Turk have made it easy to tap into an online workforce. These platforms have interfaces that allow computers to post jobs and collect the results. So why not write software that uses online workers to perform specific tasks?

The first applications are already under development. Michael Bernstein, a computer scientist at the Massachusetts Institute of Technology, has developed an editing plug-in for Microsoft Word that looks a little like Word's spell checker - except that it is powered by humans. The plug-in, known as Soylent, can be used to shorten or proof-read documents. It works by breaking the documents up into chunks, which are sent to Amazon Mechanical Turk for editing and then automatically reassembled back in Word.

Researchers at the meeting batted around ideas for other human-powered apps. Suitable tasks include things that humans can do online, but which computers find difficult. One participant mentioned a search engine that ranks images by cuteness. A researcher from Adobe wondered about building a plug-in for Photoshop that crops people or objects from an image; it would be difficult to write software to do so, but humans with a little experience of Photoshop find such tasks easy.

The field is both new and old. The idea of building algorithms powered by crowdsourced labour dates back only a few years, but organised teams of humans were used to calculate ballistic trajectories for projectiles during war time around a century ago. In fact, the term "computer" originally referred to humans who did calculations for a living.

Internet-based human computers promise to do much more, but, like most new fields, the emerging science of human computation is currently somewhat disorganised. There are many different software platforms that can be used to coordinate the workers and it's not clear which is best, for example.

"Today everyone is trying to build their own computer," says Panos Ipeirotis of New York University, who argued that researchers need to find a common operating system for human computation.

Jim Giles, contributor to New Scientist