Toward a test of medical AI superintelligence

Researchers urgently need a rigorous, task-based framework to define and measure medical AI ‘superintelligence’, because existing benchmarks are misleading and insufficient.