Paper Reading #15: TurKit: Human Computation Algorithms on Mechanical Turk

Comments:
Kevin Casey
Joe Cabrera

Reference Info:

TurKit: Human Computation Algorithms on Mechanical Turk
Greg Little, Lydia B. Chilton, Max Goldman, Robert C, Miller
UIST '10 Proceedings of the 23nd annual ACM symposium on User interface software and technology

Summary:

The Mechanical Turk is a capable of supporting human computation tasks requested by users. Workers get paid to do these human intelligence computations for the user. MTurk can replace this by supporting iterative, sequential tasks such as iterating on an image description for example. The Crash-and-Rerun type of programming allows a script to be re-executed without re-running side-effecting functions. This works for those type of problems that require computing over time. The TurKit system is works well but it does have some scalability issues associated with it. This is because the crash-and-rerun program needs to re-run all previous iterations of the loopwhen it re-executes, which leads to the space requirment of the database to be too large to sotore the actions.

Discussion:

I do not seem to understand the application of this other than research because I dont see many tasks that can be done by a human that connot be done by a computer that will be asked on this Turk machine.