Our seminar is concerned with the ethical implications of artificial intelligence in the context of medicine. A brief empirical case regarding a smart insulin pump is presented, to illustrate the complexity of patient-AI interactions. Using the concept of articulation work, these interactions are divided into four opacity levels for explainable AI, and the implications of these are discussed. We argue that a higher degree of transparency – better designed XAI than is implemented today – is needed to preserve patient autonomy.
