In 2017, a computer scientist named Joy Buolamwini realized that software supposed to recognize her face could not do this. The reason ? The software algorithm did not detect his black skin. This discrimination is the result of an “algorithmic bias”. “Early facial recognition algorithms did not recognize dark skin for the simple reason that the people who thought and developed and implemented it were mostly white“, explains the author and computer scientist Aurélie Jean.
An algorithm learns from a database as a set of images. If the data submitted to it excludes a category from the population, the algorithm will also exclude it. The consequences of an algorithmic bias can be serious. Researchers have found that autonomous cars dodge white pedestrians better than black pedestrians.
88% designed by men, the algorithms also reproduce the sexism of our society. In late 2019, the algorithm used by Apple's credit card would have disadvantaged women in obtaining credit. Aurélie Jean recalls, however, that algorithms are not the “culprits”: “We humans are macho, racist, sexist.“To stem this regrettable phenomenon, Aurélie Jean praises better diversity in the development environment.”For example, in a team of 10 people, if you have two or three women, you realize that sometimes it's enough to create a new dynamic“, she continues.
Today, only 15% of computer engineers are women.