 The article discusses the challenges and limitations of artificial intelligence, AI, in healthcare, including intrinsic difficulties in machine learning, logistical issues in implementation, and barriers to adoption, and socio-cultural changes. Robust peer-reviewed clinical evaluation is necessary for evidence generation, but conducting these may not always be feasible. Regulation that balances innovation with potential harm and post-market surveillance are required to ensure patient safety. Mechanisms for direct comparisons of AI systems must be developed, including independent test sets. Developers of AI algorithms must address potential dangers such as dataset shift, confounder fitting, unintended bias, generalization challenges, and negative consequences on health outcomes. If these goals can be achieved, the benefits for patients are likely to be transformational. This article was authored by Christopher J. Kelly, Alan Karthikesalingam, Mustafa Suleiman, and others. We are article.tv, links in the description below.