Embodied AI Ethics: A Phenomenological Critique of Current AI Explanation Systems

Current explainable artificial intelligence (XAI) systems face a fundamental challenge, which is the conflict between algorithmic explanations and embodied moral understanding. We examine two typical cases to reveal that human moral understanding possesses irreducible embodied characteristics. These characteristics cannot be fully captured by existing AI explanation technologies. Base on Merleau-Ponty's embodied phenomenology[1], we propose "Embodied AI Ethics" as a new theoretical framework. This framework shifts the focus from making AI more moral to protecting and enhancing human moral capabilities. We demonstrate how current XAI technologies systematically exclude these features. They do this through processes of disembodiment, decontextualization, and desubjectification. Based on these findings, we propose four design principles for embodied AI ethics. These principles provide theoretical guidance for developing AI interfaces that respect and support human moral understanding.