The design of an explainable AI
Abstract. In the wake of the fourth industrial revolution, the range for potential uses of AI has increased, and with it its potential for harm. Nowadays, new applications are being consistently developed in the fields of healthcare, climate change regulation, security, workplace management and control, and governance mechanisms. At the same time, AI entails a number of potential risks, from being wilfully used for criminal purposes to unexpected and potentially harmful consequences for the health and freedom of its users, such as gender or race based discrimination, transparency and privacy concerns, and opaque decision-making. In following this trend, EU institutions have given rise to a prolific regulatory framework of principle-based ethics codes and guidelines. The scope of this paper is to analyse from a European legal perspective the current fairness principles regulating the design of AI systems, the relevant legal concerns in case law, and how a symbolic approach to AI could help insure fairness, interpretability and transparency. In particular, the focus shall be on the latest projects concerning computable representation of EU directives and how to build upon those to augment the efficiency of such systems and their adherence to the aforementioned ethical principles.