MedMSA: A Framework for Transparent Clinical AI Under Uncertainty

The MedMSA (Medical Model Synthesis Architectures) framework has been introduced to enhance AI reasoning in medical environments by integrating language models with formal probabilistic models. This system utilizes language models to access pertinent prior knowledge while developing a probabilistic model that facilitates calibrated and verifiable conclusions amid uncertainty. A proof-of-concept showcases its application in differential diagnosis, tackling the issues of opacity and uncalibrated reasoning present in existing AI systems. Further information about the framework can be found in arXiv:2605.09716.