Attention, Consciousness, and Linguistic Cooperation with AI

Abstract

Contemporary debates on Artificial General Intelligence (AGI) center on what philosophers classify as descriptive issues. These issues concern the architecture and style of information processing required for multiple kinds of optimal problem-solving. This paper focuses on two topics that are central to developing AGI regarding normative, rather than descriptive, requirements for AGIs epistemic agency and responsibility. The first is that a collective kind of epistemic agency may be the best way to model AGI. This collective approach is possible only if solipsistic considerations concerning phenomenal consciousness are ignored, thereby focusing on the cognitive foundation that attention and access consciousness provide for collective rationality and intelligence. The second is that joint attention and motivation are essential for AGI in the context of linguistic artificial intelligence. Focusing on GPT-3, this paper argues that without a satisfactory solution to this second normative issue regarding joint attention and motivation, there cannot be genuine AGI, particularly in conversational settings.