The majority do, but no all. We track accuracy obsessively, and tasks requiring fluent English won't go to workers with the weakest skills.
Our crowd is surprisingly diverse – there are medical students, engineers, doctors and housewives.
For our best-tested category of work to date, optical character recogition, even workers with the weakest English skills do well at recognizing English text.
In fact, I can guarantee they're not solving CAPTCHAs for hire. Our review process for new software applications keeps the quality of our tasks high and keeps spam work out.
CAPTCHA-breaking jobs generally rely on sweatshop labor and don't pay the fair wages we do, so they're not even capable of hiring our workers.
One of our core motivations as a team and a company is to use crowdsourcing as a force for good: happy workers, better technology, no spam. If you're interested in cracking CAPTCHAs, we'll be happy to refer you to our better known, difficult-to-use competitor.
Handwritten and printed OCR tasks, on the other hand, are fantastic for our system.
That is one of the issues with Mechanical Turk. In fact up to 40% of the tasks hosted at turk are spam tasks. Our goal is to be entirely spam free. We train our workers to report all tasks that ask for email addresses, creation of accounts or tasks that look similar to CAPTCHAs.
Right now it's kept internal for most purposes, but there's no reason not to open it up. The idea behind managed crowdsourcing is that end users shouldn't need to deal with reputation and crowd management themselves.
The next version of the API will support the ability to assign work only to individuals with certain skillsets - very powerful! You'll get exactly the crowd you need for the job you have.
Our crowd is surprisingly diverse – there are medical students, engineers, doctors and housewives.
For our best-tested category of work to date, optical character recogition, even workers with the weakest English skills do well at recognizing English text.