Open to the Princeton University community.

In late 2017, the U.S. House of Representatives unanimously passed the Foundations for Evidence-Based Policymaking Act, a bill that supports the collection of data needed to evaluate and improve government programs, including regulations. This bipartisan legislation follows a long history of federal executive orders, statutes, and guidance, which require agencies to use benefit-cost analysis before adopting economically significant rules and which require agencies to review existing rules to ensure they are having their intended effect. Although there is broad agreement that agencies should analyze the effectiveness of their regulations, there has been comparatively little guidance to agencies on how, exactly, they should do that. This presentation will focus on two recent recommendations of the Administrative Conference of the United States (ACUS), a government agency dedicated to finding ways to improve administrative processes in the federal government. The recommendations aim to help agencies plan for meaningful analysis and retrospective review of their regulations.

Following on the ACUS recommendations aimed at improving regulatory analysis, the session will also feature a separate presentation and discussion of recent work by Cary Coglianese on the potential for federal government agencies to use machine learning to inform their decision-making — and even possibly to automate significant portions of their regulatory and adjudicatory tasks. The kind of data that could be used to carry out the ACUS recommendations featured in the first part of this session could also be integrated into machine-learning frameworks to support ex ante government decisions. But would a shift to a system of “regulation by robot” and “adjudication by algorithm” be consistent with prevailing legal norms? In a nation committed to government “of the people” and “by the people,” any consequential public-sector adoption of automated tools for decision support and decisionmaking will inevitably raise questions about the legal permissibility of reliance on artificial intelligence. This session will suggest that prevailing legal doctrines governing federal agencies should afford these agencies considerable room to rely on algorithms, concluding that the key questions about algorithmic governance will be the kinds of policy questions about good government that should be asked for any new policy, programmatic effort, or management innovation.

Agenda for Enhancing Government’s Analytic and Learning Capabilities

I. Retrospective Review and Learning from Regulatory Experience (Executive Orders, statutes, OMB Guidance, and ACUS Recommendations) (30 minutes)

II. Questions and Answers (30 minutes)

III. Regulating by Robot (30 minutes)

IV. Questions and Answers (30 minutes)