Mechanical Turk

I love Amazon Mechanical Turk, but I'm coming to find that it's really only good for surface data.  That is, something you can easily snatch up only if it's right in front of you.  For more complicated tasks, you may just end up increasing the required approve-percentage to absurd degrees and still get something akin to crap.  My current issue with having Turk workers write things is inevitably you get one who does a ton of them by simply copy and pasting relevant passages from Wikipedia.  These aren't tough to spot, because they still have the endnote markers!  I just blocked another person doing this after requiring 95% approval rating.  That means this person gets approved 19/20 times and yet still ignores the instructions.  Cripes!

There are diamonds in the rough, for sure, but sorting through all the garbage to find them is really difficult.  Plus you can only give them so much as incentive and the rest is random chance.  Based on how much quality I'm getting out of a writer I hired of eLance, I'm starting to rethink this whole Turk thing.  It seems to be causing me undue amounts of work when the entire purpose was a form of delegation!

Here's another example.  I ask for videos of certain people and when videos of those people are scarce, search results show either A.) a video for a person with a similar name (different last or first), or B.) absolute garbage.  Also, trying to train specific people is difficult enough, but writing succinct instructions to accomplish the same is nearly impossible!

I think I'll stick to Turk only for non-search HIT's, because people try to game those too much.  Ugh, maybe I'll change my mind again too, hard to say!