This is a draft.
This page describes a failed project.
I late January 2017, unimpressed with the pace of content creation work, I decided to look into Amazon Mechanical Turk as a way to crowdsource the writing of articles. After researching MTurk for several hours, I decided to run my own HIT (human intelligence task). I allowed the task to run for a month, and decided the experiment was a total failure.
If I were to run a second experiment, I would look into whether raising the task payment is important, trying to identify higher quality workers, and figuring out how to enable automated supervision of work.
In order to understand my excitement and willingness to try out MTurk, it is necessary to understand both the tedium of content creation work as well as my potential solution for automating the most repetitive parts.
- 2017-01-31: I seriously consider trying to use Mechanical Turk, and begin reading about it a lot.
- 2017-02-01: around noon, I purchase $20 worth of prepaid HITs.
- 2017-03-01: I finally cancel my Mechanical Turk batch. By this point, I have 11 approved tasks, but end up using none of it.