Blog #6: TurKit

Reference

Authors:
Greg Little     Massachusettes Institute of Technology, Cambridge, MA, USA
Lydia B. Chilton     University of Washington, Seattle, WA, USA
Max Goldman     Massachusettes Institute of Technology, Cambridge, MA, USA
Robert C. Miller     Massachusettes Institute of Technology, Cambridge, MA, USA


Published in:
UIST '10 Proceedings of the 23nd annual ACM symposium on User interface software and technology

Summary


This paper is mainly about a concept called, "Mechanical Turk" or (MTurk); a concept where computer programs use humans to do complex mechanical computations that are easy for humans.  Examples of this include labelling pictures, reviewing things, and rating things.  All of these tasks are either impossible or very difficult to program into a computer but very simple for humans to do.  The authors list the features that their program implements: an API for using MTurk with standard programming techniques, a web GUI for managing MTurk scripts, and a new model of computer programming called, "crash and re-run" programming.  This  new model of programming takes into account the long delay in waiting for human input.  A common example of MTurk the authors cite as a success is the online encyclopaedia: Wikipedia.  This website builds instead completely from user input and interaction. 

The script the authors implement to aid in MTurk programming is entitled, "TurKit Script."  It is based on the popular scripting language Javascript but is lightly modified to allow for crash and re-run programming and other MTurk features.  Although TurKit is single threaded, it has functions to mimic multi-threaded programs.  This is useful because MTurk is inherently  a multi-threaded concept since many people will be entering in input at the same time.


Discussion

Yet again, I am completely fascinated about the topic in this paper.  The ability to correctly identify images or write acturate reviews of certain items is extremely difficult in terms of computer programming.  Farming out difficult work like this to humans who can do this kind of work easily is very smart.  I already see this kind of technique used in the ever popular "recaptcha" and I could foresee this being used elsewhere also.