Abstract

This essay examines which conceptual frameworks for understanding intelligence are best suited to artificial systems, with particular reference to the question of Artificial General Intelligence (AGI). I argue that AGI should not be framed in analogy to human cognition or biological embodiment, since such anthropocentric approaches risk mischaracterizing the nature of artificial entities and conflate the distinct philosophical question of whether machines can be cognitive with the question of whether they can be intelligent. Following a comparison of pragmatic-behavioral, cognitive, and embodied approaches, I defend a pragmatic characterization in the tradition of Mollo (2024), on which intelligence is understood in terms of flexibility, generalizability, goal-directedness, and adaptive learning. Against accounts that treat physical embodiment as a precondition for genuine understanding (Chemero, 2023; Mitchell & Krakauer, 2023), I draw on Wittgenstein's notion of forms of life and on Collins (1996) to argue that contextual understanding is primarily a product of social embeddedness rather than of shared biological structure. On this view, the absence of sensory-motor experience does not in itself preclude artificial systems from acquiring informal understanding, provided sufficient immersion in shared sociocultural contexts. While present LLMs likely fall short of the degree of linguistic socialization such acquisition would require, the framework developed here permits recognition of the prospects for intelligence in AI in a more meaningful and non-anthropocentric manner.