Integrating Ethics into Data Science: Insights from a Product Team

Abstract

AI ethics is a significant area of scholarly and public debate and, increasingly, of controversy. Throughout the last decade, there has been a growing trend of organisations engaged in AI development committing themselves publicly to upholding ethical principles. Meanwhile, technology giants have built up substantial internal research teams dedicated to the ‘ethics’ of AI, alongside a growing body of research produced by these teams as well as independent academic researchers.

In this context, there is a need for research on how ethical approaches are actually being applied within technology teams working in AI, and to what extent they actually change day to day practice of the engineers and data scientists building AI tools and products. In this article, we analyse the findings of an ethnographic study of Wellcome Data Labs, a data science team within the Wellcome Trust, with the aim of understanding how members of the team responded to its attempt to develop an ethical approach and embed ethical thinking inside its technical product team, and identifying to what extent the initiative resulted in new practices.

Drawing upon our findings, we argue that notions of ethics work in data science product development operate as a mutable and flexible concept in practice, and that mechanisms for interdisciplinary engagement become essential ways to make AI ‘ethics’ work meaningful and feasible, enabling team members with different expertise to navigate ways of reconciling their own understanding of the work with one another.

Keywords: Data science, ethics, expertise, interdisciplinarity, ethnography