Artificial intelligence and discriminatory bias: Is a new concept of algorithmic discrimination necessary?

Abstract

AI systems incur discriminatory biases and, in particular, gender biases, which present differences with respect to the biases that generally exist in people and in society. Biases in AI tend to increase general biases and, in any case, present specific manifestations. The question to be addressed in this study is whether the current legal concepts on discrimination are sufficient to provide protection against such biases, or not, and in that case whether the creation of a new concept of algorithmic discrimination would be necessary. The answer to the question requires analyzing the causes of discriminatory biases in AI systems, their manifestations and the difficulties of legal qualification. Afterwards, what the European Artificial Intelligence Act contributes at the issue. Finally, some considerations that are not intended to be closed, but rather proposals for debate.