Talkative AI and the fiction of artificial minds

Abstract

Talkative artificial systems can sustain a realistic impression of intelligent agency, by generating quick, fluent, and context-sensitive strings of text in response to human input. Realism about AI treats this linguistic proficiency as evidence for an underlying suite of cognitive capacities, and so regards these technologies as subjects of psychological states. Firstly, I raise a challenge for the realist view from the perspective of embodied cognition: how might digital systems undergo interoceptive, affective, perceptual, and action-oriented mental states that seem to bear an important connection to the animate body? Secondly, I propose that the anti-realist view is consistent with taking a fictionalist attitude towards artificial speakers. When we converse with a talkative AI, we often imagine, pretend, or make-believe that our interlocutor is an autonomous agent with thoughts and experiences of its own, just as we do for more traditional fictional characters. These fictions are distinctive in being interactive, co-constructed, and self-involving; and in being capable of blending the world of the real with the world of the imaginary.