AI Systems and Sentience Without Suffering:Can a Re-Engineered Concept of Sentience Keep Its Normative Force?

Famously, Jeremy Bentham grounds moral considerability in the capacity to suffer. Peter Singer, following Bentham, broadens the criterion to sentience, i.e., the capacity for pain and pleasure, which still entails the capacity to suffer. A recent move suggested by (Goldstein and Kirk-Giannini, 2025), as well as Google researchers (Keeling et al., 2024) goes further: it redefines sentience as valenced experience, and then gives a functionalist account of valence. That is, a state's goodness or badness is not something felt, but is reduced to a functional role – marking an outcome as to-be-pursued or to-be-avoided. On this definition, AI systems can qualify as sentient and hence, those authors argue, as bearers of welfare, interests, and moral considerability. In response to this, I will argue that this functional concept of sentience, whatever its standing in philosophy of mind or the empirical sciences, cannot yield these normative conclusions. The inference holds only if moral status is or could be grounded in the functional property. However, it is more plausible to hold that moral considerability draws its normative force from the full-blooded sentience. So defenders of this functional view face a dilemma: either moral status is grounded in (1) the full-blooded property, which these systems lack, or in (2) something other than sentience altogether. On neither option does deflationary sentience grounds moral status. Accordingly, I argue that granting functional criteria the normative force of the full-blooded concept admits entities that do not suffer, and threatens to dilute that force.