A self-representational theory of phenomenal consciousness can avoid problems of misrepresentation by making self-representations contingently infallible when they occur within conscious mental states. This approach acknowledges the logical possibility of self-misrepresentation while preventing actual errors. Three other self-representational models—by Kriegel, Van Gulick, and Gennaro—each reveal the need for such infallibility. The account is naturalistically acceptable only if self-representations have neo-Fregean, directly referring indexical content and conscious states have a specific ontological structure, drawing on work by Perry and Kaplan.
Talkative artificial systems can create a realistic impression of intelligent agency by generating fluent, context-sensitive text. Realism about AI interprets this linguistic skill as evidence for underlying cognitive capacities, treating these technologies as subjects of psychological states. This paper challenges the realist view from the perspective of embodied cognition, questioning how digital systems could undergo interoceptive, affective, perceptual, and action-oriented mental states tied to an animate body. The author proposes that an anti-realist view aligns with a fictionalist attitude: when conversing with a talkative AI, we often imagine or pretend that our interlocutor is an autonomous agent, similar to engaging with fictional characters. These fictions are distinctive for being interactive, co-constructed, self-involving, and capable of blending reality with imagination.