Speaker: Marija Slavkovic (Univ. of Bergen, Norway), invited Professor at LIP6

Titre: Social choice and AI value alignment

Abstract: The AI value alignment problem can be seen as the problem of aligning the desirable behaviour and ethical impact of technology with what the algorithm is actually doing. We need to be able to introduce the desired values as a parameter that is used by an algorithm or by which the algorithm behaviour is evaluated. The question of how to do this  very quickly leads to the problem of: whose values. In the absence of an ideal benevolent dictator, decisions on what is the value a machine should be built around is a collective decision. The talk discusses social choice for making ethical decisions and identifying societal values. We will discuss some approaches and go through the challenges of using social choice.