Trustworthy AI – reliability, fairness, and privacy in learning from data: are we ready?

Davide Anguita

Università degli Studi di Genova, Italia

Abstract

Artificial Intelligence (AI) is finally becoming widespread and is around us, automating simple (or not so simple) tasks, helping to improve our health, increase our safety or support critical decisions. The availability of huge amount of data, produced by an increasingly digital world, and easy-to-use Machine Learning (ML) tools, available on increasingly powerful computing devices, has made the application of this technology within everyone’s reach. However, this simplicity comes with some danger and there are already some signs that, if we will not be able to improve AI trustworthiness, we could sooner or later enter a new “AI winter”. In the last years, working toward this goal, a lot of research has concentrated on the development of Trustworthy AI, adding properties like, among others, reliability, fairness, and privacy to ML algorithms. The focus of this talk will be on presenting some food for thought on the gaps between theory and practice respect to these properties in ML research.