Скачать или смотреть Think Like a Doctor:Theory of Mind for Explainability and Trust in HumanAI Collaboration-Wico Mulder

Despite significant advancements, many AI systems in the medical domain function as passive tools rather than trusted collaborators in hybrid human-AI settings. Although these systems demonstrate high accuracy in prediction tasks, they often lack the adaptability, explainability, and communication efficiency needed to establish trust and collaborate effectively with healthcare professionals. This hinders their potential to become trusted diagnostic companions in hybrid teams.
We are developping a framework to support decision-making in a medical context. The framework facilitates a two-way communication channel, enabling an AI agent and a doctor to exchange observations and justifications when classifying ambiguous cells in histological tissue samples. This exchange mirrors real-life consultations between medical professionals, where diagnostic indicators for ambiguous regions often hinge on neighboring areas.
Central to our research is the integration of computational Theory of Mind to guide the agent through the exchange of information. We found that AI agents capable of inferring the doctor's reasoning during interactions can strengthen explainability and trust in human-AI collaboration.