Fast or Slow? Human-Inspired Self-Evolving Framework for Resilient AI Systems

Abstract

This paper proposes a disruptive shift toward human-like self-evolving loops as a foundation for resilient AI systems. At the core of our proposal is the PURER loop (Perceive, Update, Reason, Execute, Reflection), a cognitive- inspired framework that enables intelligent AI agents to treat experiences (including failures) as fundamental learning triggers. PURER is centered on five tightly coupled stages: (1) Perceive by gathering raw and engineered states, (2) Update by revising memory and internal world models, (3) Reason by planning and prioritizing actions based on reward models, (4) Execute by enacting behaviors via tool calls, and (5) Reflect by learning from outcomes to adjust future behavior. Unlike current static models that rely on periodic updates via retraining, ad-hoc recovery pipelines, or vulnerable fine-tuning practices, PURER enables continuous exploration, functionality transfer, and adaptation without manual intervention.

Publication
In Proceedings of the 56th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2026 Disrupt Track)