Medical AI has made significant strides in diagnosis and prognosis, but its understanding of treatment is lacking. Instead of relying on human opinions and synthesized data from texts like biomedical publications, researchers argue that actual treatment outcomes should be incorporated into training and evaluation. This neglect limits the potential of medical AI and causes deficiencies in current models and benchmarks. --- Why it matters: This matters because accurate treatment outcomes are essential for developing effective medical AI systems. By ignoring this aspect, researchers risk creating models that don't translate to real-world improvements in patient care. Source: https://arxiv.org/abs/2608.14598