I'm doing a Master's thesis in the intersection between Cognitive Science and Machine Learning. The group I work with looooves Mechanical Turk. It's such a cheap way to do small validation studies.
Once you realize these large groups of people exist, you can still get useful work done by accounting for obvious biases.
Disclaimer: I've used MTurk both for surveys as a requester and also used it as a Turker for a month two years ago, whenever I got bored.
I saw so many Kahneman inspired studies it was not even funny at the end. (Monty Hall problems, priming, $1.10 bat+ball problem, etc etc) . I ended up making around $6 an hour.
The best problems were porno classification problems.
I suppose that is standard stuff you do in psychology these days.