The problem of crowdwork remains the crowd

Around 2017, demand for microtasking crowdwork changed quickly and significantly, both in quantity and quality. Florian Alexander Schmidt tried to figure out, among other things, whether this was "a short-lived phenomenon or [something offering] long-term economic prospects for crowdworkers". This post is my own summary of the resulting report, titled "Crowdsourced Production of AI Training Data" and published in February 2019. What caused the sudden change in the demand for microtasking crowdwork was the equally sudden need of lots of high quality training data for autonomous vehicles. Those data are fed to the so-called self-learning algorithms that "drive" self-driving cars.