A Comparative/Integrative Theory of Biological and Artificial Autonomy

Autonomous behavior appears both in living organisms and in artificial agents, but it does not arise from the same principles. This paper aims to develop a comparative account of autonomy between biological and artificial systems, asking what it really means for a system to act “on its own.” We first trace the biological roots of autonomy, from autopoietic self-maintenance and predictive regulation (homeostasis/allostasis) to the emergence of goal-setting, planning, and norm-sensitive behavior in social species. We then examine artificial systems, from reactive controllers to model-based reinforcement learning agents, and analyze how their apparent autonomy is scaffolded by designer-imposed objectives, external energy/maintenance, and engineered world models. We argue that the core fracture between biological and artificial autonomy lies in two properties currently unique to living systems: self-maintenance and intrinsic normativity (the fact that some states are good or bad for the system because they affect its continued existence). Finally, we propose a two-axis framework to locate any agent, biological or artificial, in terms of (i) substrate (“hard” embodied/self-sustaining vs. “soft” purely computational) and (ii) level of autonomy (from reactive need satisfaction to norm-guided, socially negotiated agency). This framework aims to clarify which aspects of biological autonomy current AI actually captures, which aspects it only simulates, and which it fundamentally lacks.