Using Model Cards for ethical reflection on machine learning models: an interview-based study

Abstract

How do tools designed for documenting machine learning models contribute to developers’ ethical reflection? We set out to answer this question regarding Model Cards, a tool proposed for such purpose. We conducted a thematic analysis of eight semi-structured interviews based on speculative design sessions. Each participant assumed the role of developer of an artificial intelligence model in one of two scenarios: loan applications or university admissions. We found evidence that designers may have been selective about which ethical issues – from among those they had reflected on – they recorded in the Model Cards. While participants were hesitant to grant full autonomy to the artifact to be developed, we identified they still tended to rely on a third party (outside the design team) to mediate the relationship between the system and other stakeholders. Our findings contribute to our understanding of documentation tools, their epistemic value, and how they can be leveraged to engage in a more ethically informed design process of artificial intelligence systems.