This talk will not be live-streamed or videotaped.
No RSVP required for current Princeton faculty, staff, and students. Open to members of the public by invitation only. Please contact Jean Butcher at if you are interested in attending a particular lunch.
Scholars of privacy law and data protection are increasingly interested in the growing use of statistical predictions and automated decisions by powerful institutions One branch of recent work, situated in the administrative law literature, centers on government’s own use of automated decisions, and considers how data-driven administrative activity can best be made consistent with constitutional mandates, statutory requirements and normative aspirations.
At the same time, parallel and often isolated lines of work explore and assess particular government applications of predictive analytics within specific administrative domains, including in policing, child protection, education, and immigration. Some of this work builds on earlier scholarship, reaching back to the mid-twentieth century or before, about the inherent difficulties and limits of quantifying social life.
A recurrent theme across these contexts is that public sector leaders often lack even the most basic understanding of how predictive analytics works, and of how such tools can best be used and governed. Basic challenges that are inherent to any use of these new methods are being rediscovered, and redescribed, in one domain of application after another. Practical insight from the field of data science — much of it refined in commercial settings — is not reliably visible to these communities.
In this talk, David provides a unifying, clear, accessible and factually grounded framework for public sector leaders to use when evaluating potential applications of predictive analytics to public interest goals. He identifies recurrent challenges and means to address those challenges.
The challenges of particular interest include excessive policymaker and official optimism about the correctness of automated predictions; difficulty in defining the outcomes that must be maximized, such as a police department’s overall performance or a child’s welfare; a strikingly common failure to monitor outcomes, assess predictive performance, and re-calibrate models over time; and an inability to measure elasticity or responsiveness — that is, an inability or refusal to specifically predict how potential interventions, such as social supports or punitive sanctions, will change the probabilities that were initially predicted.
In the latter part of the talk, David will explore possibilities for updating institutional design in the public sector to align with the operational requirements of sound statistical prediction.
David Robinson is a managing director and cofounder of Upturn. David works to empower the public, advocates, and policymakers to influence the high tech systems that shape our daily lives, from courtroom algorithms to “predictive policing” systems to gig economy platforms. He has a long-term interest in concrete ways that data-driven systems can support, or frustrate, civic goals of justice, equity and opportunity. Before Upturn, David served as the inaugural associate director of Princeton University’s Center for Information Technology Policy. He has also worked as a journalist. David serves as an adjunct professor of law at Georgetown University Law Center, where he teaches a seminar he proposed and designed on Governing Automated Decisions. He holds a J.D. from Yale Law School, and bachelor’s degrees in philosophy from Princeton and Oxford, where he was a Rhodes Scholar.