In response to this research project, 3 image training datasets were terminated, 2 citations were censored, 1 author apologized, and face recognition training datasets became a front page story” – This is how the artists Adam Harvey und Jules LaPlace summarize the early results of their collaborative art and research project “MegaPixels”. Harvey and LaPlace are examining different facial recognition databases, publishing them on their website and exploring their ethically dubious nature. To this end, they investigated the included pictures, the motivation of their inclusion, as well as the financial backing of the data collection efforts.
Selfies, profile pictures, pictures of celebrities, YouTube tutorials, video snapshots as well as authentic, “in the wild”¹ recordings are used in building and expanding the picture libraries and datasets. The where, and how of the usage of these datasets, who can access and the why, are often unexplored and barely controlled. Among other uses, neural networks train on these datasets to improve facial recognition capabilities. A greater the number of pictures, a larger variety of lighting conditions, perspectives and cutouts increases the reliability with which algorithms can determine and evaluate the biometric features of a face.