François Kammerer

Many think AIs would be morally significant if and only if they are phenomenally conscious in certain ways. This sentientist conception has been challenged, and alternative views emerged, on which agency, or the possession of desires, are sufficient for AI moral significance, even without consciousness. I argue that these alternative views face serious problems. They should probably be ruled out. I diagnose the mistake we made when formulating these views – a process I call “analytical drift”. I make methodological suggestions to avoid analytical drift in further theorizing about AI moral significance. Finally, I show how the space of options expands when we try to avoid analytical drift, by sketching a new alternative to sentientism about moral significance, inspired by hybrid theories of well-being and Weil’s concept of attention.