Talk by Shaul Duke: The ethical and epistemic challenges of clinical AI for conditions with soft diagnosis

Developers of clinical AI are currently venturing outside the boundaries of conditions with ‘standard diagnosis’, and towards conditions with ‘soft diagnosis’. Yet this shift has wide implications for ethical and epistemic issues. In this talk, I will analyze these implications using the case study of a tool developed for the purpose of early detection of lung cancer cachexia. It is based on 29 interviews with medical staff from three European medical centers. Results show that AI tools to detect conditions with soft diagnosis are at high risk of automation bias, that they may render human oversight ineffective, and that they cause an epistemic shift away from treating and diagnosing clinicians and towards the AI system.

Shaul Duke is a guest researcher, Department of Food and Resource Economics, University of Copenhagen.

Please bring your own lunch!