{"id":"arxiv/2601.22295","url":"https://arxiv.org/abs/2601.22295","tags":["math.OC"],"score":0,"title":"Operating Imperfect AI: Reliability Drift and Human Congestion","domain":"math.OC","source":"arxiv","authors":["Ziyao Wang","Svetlozar T Rachev"],"reviews":0,"subject":"math.OC","summary":"The deployment of machine learning in high-stakes services relies on ``human-in-the-loop'' architectures to mitigate algorithmic uncertainty. However, existing static policies fail to address a fundamental tension: algorithms suffer from stochastic ``reliability drift,'' while human override capacity is scarce and congestible. We formulate the management of such systems as a dynamic queueing control problem. The system state is defined by the tuple (queue backlog, reliability regime), and the control variable is a state-dependent risk threshold. We prove that the optimal escalation policy is…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-01-29","reactions":{}}