The Embodiment Challenge for Artificial Intelligence

Summary

Despite recent advances, artificial intelligence systems fundamentally differ from human cognition because they lack biological embodiment, situatedness, and autonomy. Humans learn through embodied, intersubjective experiences tied to survival and adaptation, whereas AIs cannot undergo structural changes due to their disembodied nature. The embodiment challenge argues that without these self-sustaining, survival-driven processes, artificial systems cannot replicate natural cognition. Creating a genuinely cognitive artificial intelligence would require first achieving artificial life.

Study at a glance

Abstract

Abstract Recent advances in the field of artificial intelligence (AI) renewed the enthusiasm for what AIs can do. While some argue that AIs can be viable models for understanding human cognition, others go further and claim that future AIs might display cognitive capacities. However, although some current models excel at tasks such as language processing and pattern recognition, their operations fundamentally differ from human cognition. Unlike humans, who learn through embodied, intersubjective experiences tied to survival and adaptation, AIs typically cannot undergo structural changes due to their disembodied nature. To understand the significance of this difference, I turn to the radically embodied framework and raise the embodiment challenge for AI, namely: due to the lack of biological embodiment, situatedness, and autonomy, artificial systems cannot replicate the self-sustaining, survival-driven processes essential to naturally cognizing systems. I conclude that creating a genuinely cognitive artificial intelligence would require achieving artificial life first.