I think this is a pretty important publication. It deserves wider coverage and discussion.<p>Strong ethical basis to AI/ML in health required, and there are many unknowns. So they take a cautious position: we should seek the positive applications but be wary of false claims, over claims, and risks.<p>The implicit language around "why" is cost: Clinicians are expensive to train. The goal is to reduce health cost (cost in all its forms. Time, Labour, delayed consequences, over-prescribing..) AND improve patient outcome, but the ordering of the motivations varies depending on who is doing the talking I guess.<p>They seem skeptical of most of the models. They seem to be saying the baseline is so variant, proving a model is "better" or "faster" is still increadibly subjective.<p>More work required.