CSAIL researchers developed an AI system that can predict whether an AI or human is a better decision-maker. The system relies on two separate machine-learning models to make decisions, like diagnosing a patient, in tandem with humans. One model makes the decision or diagnosis, and the other predicts whether the AI should defer to a human.
- The paper comes from Hussein Mozannar and David Sontag, who are researchers at CSAIL, MIT’s Computer Science and AI Laboratory
- Their decision-making system could apply not just to clinical applications like assessing chest X-rays, but the moderation of social media posts.
- The second model, known as the "rejector," can improve its predictions over time.
- The research is important as more medical environments use AI and automation to support human clinicians.