Prescribing Partnership: Operationalizing Trust and Control in Human-AI Delegation

Résumé fourni par la source

As artificial intelligence integrates into the workforce, new frameworks for human-AI collaboration are essential. We examine a crucial inquiry: How can specialists effectively assign high-stakes responsibilities to developing AI agents? We present the Dynamic Delegation Boundary Framework; a contingency model predicated on AI epistemic uncertainty and task criticality. This implements contemporary delegation theories via a two-phase formal model that integrates decision theory with Qlearning. In clinical coding assessments with a customized LLM, our HAIC system demonstrated substantial enhancements: an F1-score of 0.82 compared to 0.74 (p<0.001), a 45% reduction in effort, and a 37% decrease in cognitive burden (d=1.45). The optimal delegation threshold converged at θ*=0.75, whereas Q-learning refinement approached θ≈0.72. This research presents: (1) a contingency theory for human-AI delegation, (2) a formal governance model, and (3) a methodology for delegation tweaking as an alternative to algorithmic control.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.