EQUALITY IN CODE: ANALIZING GENDER AND RACIAL BIAS IN AI ALGORITHMS

The Artificial Intelligence (AI) technology has been through a process of development of relevant speed, becoming day by day more structured. Such evolution had great impact in our society that is increasingly developing it, in order to bring always more benefits. Artificial Intelligence has become an everyday assistant, facilitating users’ lives in quantitative ways. However, Artificial Intelligence systems are not perfect, rather they present different downsides. Focus hasn’t been posed on the harmful impact they are causing to marginalised communities. No matter what kind of AI systems, whether facial recognition, work platform or risk assessment tools. As long as input data contains expression of gender and racial biases – among the others –, the subsequent performance would not be better than human performance. In recent times, extensive research has started to formulate and develop solutions to improve algorithmic technical accuracy and transparency. However, technical improvement is not the solution to bias mitigation. Therefore, the law plays extensive role in ensuring appropriate measures to limit and prevent bias as far as possible. The key legislation in the regulation of AI systems’ biases and risks of their adverse impact, is represented by the copyright fair use doctrine, under U.S. law, and the newly introduced Artificial Intelligence Act under EU law, which however, is considered to have sown the seeds for subsequent legislation beyond the EU borders.