← 研踪 PaperTrailarXiv:2609.32083 · v2
When Should a Human Take Back Control? Optimal Delegation under Turbulent AI Risk
首次提交 2026-09-26 07:43
版本更新 2026-11-16 13:32
雷达观测 2026-11-17 17:12
摘要
When should a human take control from an autonomous AI agent, how much checking is worthwhile, and when should autonomy resume? These decisions become harder when incidents cluster and recent errors signal elevated risk. Exploratory analyses of AgentLogs coding-agent traces and ODCV-Bench task runs show higher held-out log likelihood for Hawkes incident-history models than matched no-history baselines, with positive gains whose conditional intervals exclude zero. AgentLogs evaluates continuous-time incident timing and type; ODCV evaluates failure occurrence at discrete decisions. Motivated by this predictive evidence, we formulate a continuous-time delegation problem that jointly chooses productive action, verification effort, and switching rates between human-directed, AI-assisted operation and autonomous AI operation. The model assumes that verification weakens incident propagation and balances project benefits against operating, supervision, risk, and handoff costs. Our Hawkes-PPO method summarizes observed incidents with exponential memory filters and learns these decisions from rewards. In synthetic experiments, it achieves a higher evaluated objective than either fixed regime and PPO without history filters, and approaches an approximate full-information benchmark. Further experiments show how incident excitation and supervision costs shape delegation.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-16 13:32 · 最新观测版本
When Should a Human Take Back Control? Optimal Delegation under Turbulent AI Risk
When should a human take control from an autonomous AI agent, how much checking is worthwhile, and when should autonomy resume? These decisions become harder when incidents cluster and recent errors signal elevated risk. Exploratory analyses of AgentLogs coding-agent traces and ODCV-Bench task runs show higher held-out log likelihood for Hawkes incident-history models than matched no-history baselines, with positive gains whose conditional intervals exclude zero. AgentLogs evaluates continuous-time incident timing and type; ODCV evaluates failure occurrence at discrete decisions. Motivated by this predictive evidence, we formulate a continuous-time delegation problem that jointly chooses productive action, verification effort, and switching rates between human-directed, AI-assisted operation and autonomous AI operation. The model assumes that verification weakens incident propagation and balances project benefits against operating, supervision, risk, and handoff costs. Our Hawkes-PPO method summarizes observed incidents with exponential memory filters and learns these decisions from rewards. In synthetic experiments, it achieves a higher evaluated objective than either fixed regime and PPO without history filters, and approaches an approximate full-information benchmark. Further experiments show how incident excitation and supervision costs shape delegation.
查看此版本解读 →查看此版本原文 ↗