Episode 362

Mechanical Turk was one of those services I thought sounded incredibly cool when it first launched. At the time, though, I didn’t really understand what you’d actually use it for. As machine learning evolved, the need for large-scale data labeling became obvious, and suddenly Mechanical Turk made a lot more sense.
I even gave it a try. After spending a few hours completing tasks and earning only a few dollars, I realized it wasn’t worth my time. But that’s not true everywhere. In many parts of the world, a few U.S. dollars can go a long way, which made Mechanical Turk an attractive source of income for a lot of people.
The irony is that AI—or, as we like to joke on the show, “how ML makes money”—has increasingly taken over many of those same low-level classification and identification tasks. The very human workforce that helped train modern AI systems has, in many cases, been displaced by the models they helped create.