Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents

Key points

• Current evaluations score the co-evolved policy on the final bank it trained with (the delivered bank).1
• The acceptance rule these methods share (cooperative acceptance) ties a policy's competence to the bank it co-evolved with, leaving it fragile once skills drift, regardless of the implementation; empirically, sensitivity to skill drift grows as co-evolution training proceeds.1
• We therefore introduce the CO-evolution Drift Audit (CODA), a deployment-oriented evaluation protocol that simulates skill drift and, alongside success rate under worst-case drift, introduces two new metrics: a consistency rate under meaning-preserving edits and a resistance rate to corrupted skills.1
• CODA thus reframes the target of co-evolution, from peak performance on the co-evolved bank to the ability to harness skills however they evolve, and provides a shared standard against which future methods can be audited.1

Sources (1)

• [1]Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 07:29 PM

  Skill-policy co-evolution, in which an agent's policy model and its skill bank improve each other during training, is a promising paradigm for self-improving agents.

  Current evaluations score the co-evolved policy on the final bank it trained with (the delivered bank).

Extractive summary: sentences quoted from the sources.