Machine learning and AI research for Patient Benefit: 20 Critical\n Questions on Transparency, Replicability, Ethics and Effectiveness

Abstract

Machine learning (ML), artificial intelligence (AI) and other modern\nstatistical methods are providing new opportunities to operationalize\npreviously untapped and rapidly growing sources of data for patient benefit.\nWhilst there is a lot of promising research currently being undertaken, the\nliterature as a whole lacks: transparency; clear reporting to facilitate\nreplicability; exploration for potential ethical concerns; and, clear\ndemonstrations of effectiveness. There are many reasons for why these issues\nexist, but one of the most important that we provide a preliminary solution for\nhere is the current lack of ML/AI- specific best practice guidance. Although\nthere is no consensus on what best practice looks in this field, we believe\nthat interdisciplinary groups pursuing research and impact projects in the\nML/AI for health domain would benefit from answering a series of questions\nbased on the important issues that exist when undertaking work of this nature.\nHere we present 20 questions that span the entire project life cycle, from\ninception, data analysis, and model evaluation, to implementation, as a means\nto facilitate project planning and post-hoc (structured) independent\nevaluation. By beginning to answer these questions in different settings, we\ncan start to understand what constitutes a good answer, and we expect that the\nresulting discussion will be central to developing an international consensus\nframework for transparent, replicable, ethical and effective research in\nartificial intelligence (AI-TREE) for health.\n

Ähnliche Arbeiten