HR-Bandit: Human-AI Collaborated Linear Recourse Bandit

2024-10-18

arXiv.org

AI Summary

This study addresses the exploration-exploitation trade-off in AI-driven dynamic recommendation of executable interventions (e.g., patient behavior adjustments) within human-AI collaborative clinical decision-making. We propose the first “human-expert-in-the-loop linear recourse bandwidth” framework, theoretically guaranteeing warm-start capability, low human operational cost, and robustness to human decision variability. Our method extends linear UCB to construct a recourse policy, integrating human-feedback-driven adaptive confidence interval updating and cost-aware interaction mechanisms. Evaluated on real-world clinical cases, it significantly reduces cumulative regret versus baselines, improves initial response performance by 32%, and decreases human interventions by ~80%. The core contribution lies in formalizing human intervention as a bandwidth-constrained linear recourse process—enabling unified optimization of recommendation performance, operational efficiency, and decision robustness.

Technology Category

• Humans and AI: Planning and Decision Support for Human-Machine Teams
• Machine Learning: Online Learning & Bandits
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

Human doctors frequently recommend actionable recourses that allow patients to modify their conditions to access more effective treatments. Inspired by such healthcare scenarios, we propose the Recourse Linear UCB ($ extsf{RLinUCB}$) algorithm, which optimizes both action selection and feature modifications by balancing exploration and exploitation. We further extend this to the Human-AI Linear Recourse Bandit ($ extsf{HR-Bandit}$), which integrates human expertise to enhance performance. $ extsf{HR-Bandit}$ offers three key guarantees: (i) a warm-start guarantee for improved initial performance, (ii) a human-effort guarantee to minimize required human interactions, and (iii) a robustness guarantee that ensures sublinear regret even when human decisions are suboptimal. Empirical results, including a healthcare case study, validate its superior performance against existing benchmarks.

Problem

• Optimizes action selection and feature modifications
• Integrates human expertise to enhance algorithm performance
• Ensures robustness and minimizes human effort in decision-making

Innovation

• Recourse Linear UCB algorithm optimizes action selection.
• HR-Bandit integrates human expertise for enhanced performance.
• HR-Bandit guarantees warm-start, human-effort, and robustness.