Abstract

In ethical debates on moral status, a common assumption is that in cases of uncertainty it is safer to be over-inclusive rather than under-inclusive when defining the boundaries of our moral community. Some argue that, since it cannot be definitively ruled out that AI might one day possess morally relevant capacities—such as consciousness or the ability to feel pain—a more ethically cautious approach may be to treat AI as a moral patient. But is over-inclusiveness truly the more cautious approach to AI? I argue that the risks associated with the over-attribution of moral status to AI are more likely in the short term, as they are already materializing. This is illustrated by cases of so-called AI psychosis, as well as by the potential to divert resources and care away from beings that clearly need them toward parasocial relationships with AI. By contrast, the risks associated with under-attributing moral status to AI—existence of which I do not deny—can, to a certain extent, be mitigated by other ethical reasons for refraining from interacting with AI in problematic ways. Such reasons (e.g., concern for one’s own moral character, respect for norms governing certain social practices, or aversion to symbolic violence) do not, however, require attributing moral status to AI itself.