I’ve been reading a lot of social science and human computer interaction papers lately. Mechanical Turk has been a boon to researchers in these areas. With the click of a button, you can recruit a large sample size to test an inconceivable array of theories about society: everything from validating the assumptions of economic models, to testing methods of persuasion, to exploring deception in dating profiles. Papers are published and win acclaim using interview studies, surveys, and behavioral experiments conducted on Mechanical Turkers.
The platform is cheap and it delivers results. It has the economic properties that Fox News was the first to discover on cable.
Now what if you had a budget and an agenda? What if the reality of the social, economic, or psychological world was undesirable to you? What kind of budget would be required to systematically take advantage of the fact that today’s social science has come to depend so much on Mechanical Turk. Sure there’s a threat model of sorts for using MTurk for research. Plenty of best practices to weed out pathologies like a Turker optimizing their income by participating randomly.
But there is no model for an advanced persistent threat who wants to be a shadow editor of the production of scientific truth.