Extend human–AI hypothesis testing beyond the baseline setting

Extend the human–AI hypothesis-testing framework beyond binary labels, two simple hypotheses, a single AI source, and a finite AI-report alphabet by treating composite hypotheses, multiple AI systems with heterogeneous costs and accuracies, and richer or continuous report spaces.

Background

The paper develops SCALE for cost-aware hypothesis testing with binary ground-truth labels, two simple hypotheses, one AI information source, and finite-valued AI reports. The authors explicitly identify relaxing these structural assumptions as an unresolved direction that would broaden the framework's applicability to more realistic human–AI inference settings, including heterogeneous AI systems and non-discrete outputs.