From CrowdsourcingSeminar

Assignment 1 - Create and Perform HITs on Mechanical Turk

Worker

This was so much worse than I had anticipated. I completed a wide variety of tasks, 11 in total. These included surveys (.50USD), Spamming a blog (.01USD), image tagging (0.03USD), as well as others. I originally started by taking the obvious scammy tasks, one of which was a "download test" for 4USD which seemed to be an attempt to install spyware on my machine. The payment was rejected, as I run linux. What surprised me was how many of these tasks were obviously scams. Things like "become my twitter friend and retweet my post", "like this youtube video" and such. Scams really seem to dominate the platform. The only remaining tasks are from large-form players like delores labs.

The tasks themselves were simple, and there was a strangely addictive element to the whole thing. There was a hope, somewhere, that some gold was available. It's not there, not by any sorting algorithm Amazon was providing. Honestly, after an hour of completing tasks (acquiring around $1.50) I was physically drained. Doing MTurk full time sounds soul-sucking in the worst way, less about wasting your time and more about failing. Perhaps it's the competitive aspect of the platform?

Requester

I'm sorta cheating here, as I've not added any new tasks. Instead, I'm going to talk a little about the tasks I have put on turk. Over the last few months I've placed around 100 different transcription tasks onto the engine. These were transcriptions of semistructured interviews about wireless and mobile. In general, I'm very positive about the service. Most of the transcriptions (probably more than 90%) were good, and most rejections were obviously wrong in a dramatic way. Still, there were some transcriptions which were low quality and discovering those was difficult. Quite a few sneaked through. I received answers within the day, usually within a few hours.

The most interesting part of the requesting side was the random messages I received. An alum from tufts called me a "bastard" because he didn't follow instructions, asking for me to "let him fix it instead of rejecting the HIT". I don't think MTurk lets me do that. The separation between workers and requesters is the core of the problem, I think. Everything is so hurried, we cannot negotiate any part of the arrangement. No questions are asked, and intentions mean little.

Lastly, the interface seemed great from this side, but terrible from the worker side. Design choices?