The term ‘value alignment’ frequently enters contemporary discussions around the ethics and safety of artificial intelligence. In this talk I argue that based on how this problem usually gets articulated, there are actually many value-alignment problems. To ’solve’ value alignment, one would have to solve at least eight distinct problems that get wrapped up under this one title. I further argue that this confusion, or lack of nuance, is holding back progress. While many of these problems cannot realistically be solved be solved with technical solutions, some can be and, hence, technical AI safety researchers should focus their attention on the appropriate set of sub-problems.
Karina Vold is a philosopher of cognitive science and artificial intelligence & an assistant professor at the University of Toronto's Institute for the History and Philosophy of Science and Technology.