A new set of recommendations aims to help improve the way datasets are used to build Artificial intelligence (AI) health technologies and reduce the risk of potential AI bias.