This is similar to a cocktail party game I once played.
You get everyone in a circle, and then you choose someone across the circle to look at. Basically, everyone is looking at someone else, and they, in turn are looking at someone else (not you).
Then, you're asked to mimic as exactly as you can the person you are looking at. Everyone starts out standing still.
But soon, funny stuff happens. Little inevitable movements by people are mimicked and magnified as the movements ripple down the chain. Eventually, you get ridiculous huge movements that ripple.
It's a fun game, and I suggest everyone try it out with a group. It's a real life experiment of signal and error amplifications.
For a whopping 2 cents, I imagine there are people who participate only out of sheer boredom and trolling it gives them more than 2 cents of value. It would be interesting to repeat the experiment with varying payouts and see how that impacts accuracy. If possible, pay each worker a bonus based on programatically calculating their accuracy and see how that turns out too.
I recommend checking out the other pieces on his site---all very beautiful. Several others also build on the theme of turkers interacting with each other, albeit through a limited channel.
This reminds me a little bit of Sketch-a-bit, though it's not the same premise (users modifying other user's sketches, rather than trying to trace them): http://superfiretruck.com/sketchabit/
You get everyone in a circle, and then you choose someone across the circle to look at. Basically, everyone is looking at someone else, and they, in turn are looking at someone else (not you).
Then, you're asked to mimic as exactly as you can the person you are looking at. Everyone starts out standing still.
But soon, funny stuff happens. Little inevitable movements by people are mimicked and magnified as the movements ripple down the chain. Eventually, you get ridiculous huge movements that ripple.
It's a fun game, and I suggest everyone try it out with a group. It's a real life experiment of signal and error amplifications.