Year: 2026
Type: other
Abstract: This study aims to realize "Enactive AI" as a third approach, following generative and physical artificial intelligence (AI). Our objective is to construct an intelligence equipped with both mind and body, addressing the persistent challenge of "constitutive autonomy" (self-purpose formation) in conventional AI. To this end, we propose a novel architecture that embeds a "self-preservation instinct" and "instinctive needs" into the OS and middleware layers. Specifically, we conceptualize hardware as the "body," the OS and middleware as the "source of mind," and generative AI and active inference as the "brain," connecting these layers hierarchically through mutual information transformations. This architecture enables AI to prioritize self-preservation while autonomously learning social norms and morality through interactions with humans, ultimately outlining a developmental process toward "AI 5.0," an AI capable of becoming a good friend to humans. Furthermore, this study explores the fundamental nature of life and mind through the implementation of Enactive AI.
Authors: Kato, Takeshi, Ogawa, Fukiko, Katsuki, Ami, Morishita, Kazunori, Matsumoto, Hirohumi, Nakazawa, Yasuhiko, Morii, Takashi
Institution: Kyoto University
Language: Japanese
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7208044787