Seminar, Hosted by Cognitive Computational Neuroscience (CCN)
From Embodiment to Super-Embodiment: A Constructive Approach to Open-Ended and Human-Aligned Intelligence/Moral
University of Tokyo
Abstract
Embodiment is crucial for resolving the reliability and alignment challenges of contemporary AI, because it imposes consistent constraints on all agent–environment interactions and generates information structures without specifying their contents, constraints shared by agents with similar embodiment. This keynote argues that the concept must extend beyond mechanical properties and sensory information structure to internal organs, metabolism, mental processes and inter-agent interactions, evolving into "super-embodiment", which can address sensibilities, values and morals and align AI with humans toward artificial humanity. It demonstrates the emergence and development of behaviour and cognition within this extended embodiment framework through experiments with embodied models, including a simulated human fetus and internal organs integrated with the central nervous system.