Poll: What experiment to run on AWS mechanical turk?

After our experience using the mechanical turk we've become a bit obsessed with trying to find a really awesome use case for the service - even if its just experimental.

We have some downtime right now and we are narrowing down the experiment between the following:

1. Some sort of cooperative art project - inspired by http://www.tenthousandcents.com 2. A massive chess game that forks every time a move is made (probably a poor idea) 3. Some implementation of the Prisoner's dilemma game. 4. Any time of game to get a sense of how risk seeking or risk averse the Turks are - possibly a tie in with the Monty Hall Problem.

Most of these ideas are centered around game theory but we are pretty much open to anything (within reason). Any votes or opinions on what would be the most interesting/entertaining?

-Thanks

I'd love to see a prisoner's dilemma experiment via MTurk. These tend to be financial in nature, though; can you set variable payment rates per task, depending on external factors?

A friend last week told me about a business he's working with. They do Natural Language Processing to scan thousands of articles about a specific topic/product to see if they are positive, negative or neutral. They are then hoping to turn around and sell the business intelligence.

I was wondering how it would compare on a cost/effectiveness basis with the Turk.

Let us know what your experiment is and how you fared.

How about a dollar auction?

we used the Turk service to basically do OCR on about 3k images - http://shout.setfive.com/2008/11/20/artificial-artificial-in...