Embodied Cognition and Strong AI

This chapter examines what embodied cognition (EC) contributes to debates about the possibility of strong artificial intelligence (AI). We begin by reconsidering the classical distinction between weak and strong AI in light of the embodied turn in cognitive science and recent developments in neural and generative AI. Although large language models can display highly sophisticated linguistic performance, we argue that such performance alone does not establish human-like intelligence, since strong AI requires the possession of meaningful concepts rather than the manipulation of concept labels. We then analyze a prominent embodied approach according to which conceptual meaning is grounded in the reactivation of sensorimotor systems. On this view, because artificial systems possess sensorimotor capacities different from those of humans, they cannot instantiate genuinely human concepts and therefore cannot realize strong AI. We challenge this conclusion by distinguishing concepts from conceptions. Concepts specify the conditions under which something belongs to a category, whereas conceptions include the body-dependent, experiential, and sensorimotor information typically associated with members of that category. We argue that the sensorimotor reactivations emphasized by embodied theories are better understood as constituting conceptions rather than conceptual meanings. Consequently, artificial systems may possess concepts sufficiently similar to human concepts even if their conceptions of the world differ substantially from ours. This distinction reopens the possibility of strong AI while requiring a refinement of what “strong AI” should mean. Finally, we argue that sensorimotor grounding does not fully solve the problem of meaning, since it presupposes rather than explains how sensorimotor states themselves acquire semantic content.