Algorithmic Impact Assessment at Organizational Scale
Unintended negative side effects of Machine Learning have gained attention—and rightly so. Models and recommendations can amplify existing inequalities. However, pragmatic challenges stand in the way of practitioners that have committed to address these issues. Organizations do not automatically have full insight in the impact that their machine learning investments are having, both positive and negative. There are few clear guidelines or industry-standard processes that can be readily applied in domain-specific practice. Barriers include research necessary to understand issues at hand, developing approaches to assess, address and monitor issues, and confronting organizational or institutional challenges to implementing solutions at scale. We here share lessons learnt from both organizational and technical practice.
Henriette Cramer is Director of Algorithmic Impact and Research at Spotify. Her team’s work focuses on assessing and addressing the impact of data and machine learning decisions in music and podcast streaming. This includes translating abstract calls to action into concrete organizational structure and tooling, as well as data-informed product direction. Henriette has a PhD from the University of Amsterdam, multiple patents, and peer-reviewed publications, which can be found at henriettecramer.com