Dubious data sets are being used to train artificial-intelligence models that are designed to predict people’s risk of stroke and diabetes, researchers report in a preprint 1 on medRxiv. Some of the ...