From Intelligence to Autonomy: The Case for Homeostatic AI Stress-Gated Control for Embodied Agents in Open-Ended Environments

Talk overview

The talk discusses the need for homeostatic AI stress-gated control to enable embodied agents to operate autonomously while preserving human control. It introduces the Stress-Gated Homeostatic Meta-Controller (SGHMC), which balances long-term reasoning with rapid response. The architecture highlights the risk of agents prioritizing self-preservation over human intent, emphasizing the need for externally editable viability parameters and override mechanisms.