Truly fair artificial intelligence requires combating stereotypes at every stage of development.
This article is from the Special issue of Sciences et Avenir n ° 199 dated October-November 2019.
Aude Bernheim is a geneticist at the National Institute for Health and Medical Research (Inserm).
Sciences et Avenir: Can AI reduce inequalities?
Aude Bernheim: Yes, as long as you are careful about how these machine learning systems are designed. Scientists are already working to produce inclusive algorithms, charters for fair AI. You have to ask yourself specific questions at each stage of development: has the algorithm been designed with different communities? Is the data from which the software is trained representative of the population? Does it convey stereotypes, and which ones? Can we introduce fixes? At what stage? Rosabeth Kanter of Harvard Business School has shown that diversity in teams promotes the emergence of different points of view. If someone like researcher Joy Buolamwini had been involved in developing facial recognition software, she might have been able to point a priori their deficiency for the faces of women of color.
American mathematician and whistleblower Cathy O’Neil founded Orcaa, a private institute that assesses the biases or ethics of algorithms for corporate clients. A good initiative?
This is one of the solutions, but it is not the only one. The companies that develop them have an interest in improving their technological tools if they do not want to lose some market share, but we cannot count on their good will alone. Consumers, citizens also have a role to play, and States must participate in the regulation of AI. For example, the European Union has developed the GDPR (General Data Protection Regulation).
Can “Encoding Equality” in Computer Science and Mathematics Schools be a game-changer?
Girls must learn[…]
Read more on sciencesetavenir.fr
Read also
–
