Learning When to Automate: Queue Control in Human-AI Service Systems
The UCB-DPP policy addresses a human-AI service system where automation decisions and human scheduling are coupled. It learns unknown parameters like chatbot success rates and human service speeds by combining Upper Confidence Bounds with Drift-Plus-Penalty control. This results in sublinear regret, $O(K ilde{p}_T ext{ log } T)$, while guaranteeing mean-rate stability of the human service queues.