The automotive industry requires a great deal of photographic and video material to train cars using machine learning so that they can correctly categorize objects. As Philip Kessler explains, however, leaving this process to humans is laborious and time-consuming. The computer scientist founded the startup understand.ai with his colleague Marc Mengler, and has unveiled a solution that should make the labeling process ten times faster and more accurate. It is carried out on a mostly automated basis, writes the Karlsruhe Institute of Technology (KIT) where Kessler studied, but final quality control is always carried out by a human to rule out errors.
The Hans Böckler Foundation at the German Trade Union Confederation (DGB) recently scrutinized the working conditions of the “crowdworkers” who up to now have generally been responsible for evaluating and flagging this type of pictorial data worldwide. According to the report, new providers have adapted to customers’ higher quality demands. Crowdworkers feel that they are treated with more respect and paid more reliably than by traditional providers, although the level of pay still often leaves something to be desired.