Trusting Black-Box Algorithms? Ethical Challenges for Machine Learning in Healthcare

Trusting Black-Box Algorithms? Ethical Challenges for Machine Learning in Healthcare

The rise of medical ML demands in-depth ethical analysis of its risks and opportunities to provide guidance for regulatory bodies and ethics committees alike. Guided by the overarching question whether and under which conditions patients and health-care professionals can expedite trust in clinically applied black box programs, the project aims to develop an informed account of the ethical and legal challenges posited by them. The project will be carried out by integrating both theoretical examination and empirical data collection and has three specific objectives.

The first objective is to set the stage for my investigation by mapping the field. On the one hand, I need to identify which ethical issues are considered most pressing by different expert groups, which I aim to achieve through qualitative interviews. On the other hand, I need to evaluate trust as a proposed solution to these challenges and evaluate which conception of trust may be adequate to deal with the uncertainties posed by medical ML.

The second objective tackles the question under which conditions trust in medical ML may be justified. Conceptually, this part shall particularly address the issues of fairness and transparency, commonly considered to constitute the two most vital ethical concerns particular to clinically applied ML (Vayena et al., 2018). By conducting semistructured interviews, I hope to extract attitudes, fears and opinions from acknowledged experts on both topics. I will furthermore address them theoretically, in particular by drawing on writings from philosophy of science.

The third objective is to combine the results from the preceding parts to promote a sensitive, ethical and balanced framework for medical ML. The theoretically and empirically informed results shall be integrated in the sense of critical applied ethics to evaluate social practices concerning medical ML, improve bioethical theory addressing the field and provide guidance for ethics committees and regulatory bodies.