Understanding Medical Black Boxes: A Philosophical Analysis of AI Explainability (360G-Wellcome-213660_Z_18_Z)
I plan during the next two years to develop a major, multi-year project into AI explainability in medical contexts. This project will connect existing literatures in philosophy of science, philosophy of medicine and medical ethics, where problems of understanding and explanation have been extensively studied, to the emerging literature on explainability in machine learning and the ethics of AI. The aim will be (i) to enhance our understanding of the problems AI systems raise for explainability in medical contexts and (ii) to collaborate with machine learning researchers to develop technical research apt to address these problems. The existing literatures on explainability and understanding in medicine are vast and have not previously been systematically connected to the ethics of AI. To lay the groundworks for a later grant proposal, this application proposes to conduct three pilot-studies, focusing on potential challenges from AI to: (1) mechanistic understanding, (2) clinical judgement and diagnostic reasoning and (3) informed consent. A part-time research assistant will assist in scoping the relevant literatures. Travel to groups at other universities and a workshop in Cambridge will furthermore help establish contacts with a network of researchers interested in the ethics of AI and AI explainability in medical contexts.
£86,561 26 Jul 2018