The Importance of Prediction in Designing Artificial Intelligence Systems

Understanding machine behaviors powered by artificial intelligence (AI) techniques has become urgent today both to shape these systems according to beneficial ends and to avoid negative impacts on individuals and societies. These systems have been progressively designed in a way that hides their internal logic from its users, namely as black boxes. Some emphasis has been devoted in the very last years on methods for explaining black box models (Guidotti et al., 2018), in particular to account for several decisions which are based on these models and impact on our lives. The availability of transparent technologies (European Commission, 2018b; IEEE, 2019) is essential in leveraging our trust in the possibility for users to know the reasons of automated decisions. Very recently, a whole new interdisciplinary field of research named machine behavior has been proposed and is devoted to “the scientific study of behavior exhibited by intelligent machines” (Rahwan et al., 2019, 477). Machine behavior as a field expands and incorporates computer science and includes approaches from other sciences, such as methods of social sciences for studying the behaviors of humans and animals.