The Rule-Relocation Problem in Bio-Inspired AI

Abstract

Bio-inspired AI is increasingly presented as overcoming externally imposed objectives by internalizing regulation through drives, homeostasis, or predictive control. This Perspective argues that this narrative is often mistaken: in many contemporary designs, normativity is not eliminated but relocated, with constraints compiled into internal variables, learning rules, or deployment infrastructures while authorship remains external. To make this relocation explicit, this Perspective introduces a lifecycle map that locates where normative ``whistles'' enter AI systems from training through deployment, and proposes a constructive design pattern---the separation of an explicit Safety Envelope from an internal Adaptive Space---that preserves robust control while making normative authorship auditable and governable.

Keywords