Administering by Algorithm: Artificial Intelligence in the Regulatory State

This policy lab explores the growing role that artificial intelligence (AI) and related technologies are playing in the federal administrative state. Already, a wide range of federal agencies are utilizing software that uses machine learning and related techniques associated with AI to make and support decisions. Examples include efforts by the Social Security Administration to improve decisional quality in the adjudication of benefits claims, by the EPA to model the toxicity of chemical compounds, and by the IRS to predict tax non-compliance and identify audit targets. Other uses of AI are in the pipeline or already in place but remain outside public view. Such use will almost certainly increase as AI becomes more sophisticated and cheaper and as the private sector’s increasing reliance on AI to make decisions presses agencies to keep pace in order to regulate effectively. As agency use of AI proliferates, administrators, lawyers, and judges will have to ask how well agency deployment of machine learning systems conforms to well-established principles of constitutional and administrative law.

Students enrolled in this policy lab will have a unique opportunity to help set the terms of that debate via a first-of-its-kind report to be submitted to the Administrative Conference of the United States, an independent federal agency charged with recommending improvements to administrative process and procedure. Students will spearhead completion of a report designed to explore the use of AI in the administrative state at multiple levels. The first part of the project will be a mapping exercise, with descriptive and predictive components. The chief descriptive task will be to canvass the hundreds of agencies that make up the federal administrative state and document agency use of AI across a wide range of substantive policy areas. A related, more predictive part of the project will draw on Stanford University’s distinctive concentration of technical knowledge in AI and related fields to assess where AI may be most likely to be deployed by agencies in the near- and medium-term. The final part of the project will turn to normative issues, by contributing to a framework for thinking about the many legal, policy-analytic, and philosophical questions raised by agency use of AI to perform regulatory tasks. To the extent possible, we will consider how agency use of AI may affect the administrative state in general terms, and will explore some of the implications of core administrative law doctrines — such as the nondelegation doctrine, arbitrary and capricious review, due process, and rules governing reliance on subordinates for decisions — for agency use of AI.

Students enrolled in the lab will work in teams, with each allocated a cluster of agencies. As the project unfolds, teams will drill down on practices, both actual and predicted, in specific agencies that exemplify the legal and normative tensions that will arise as agencies increasingly deploy AI technologies. Students will be encouraged to deploy a range of methodologies, including careful secondary and legal research, case studies, survey work, and stakeholder interviews, among others. Some of this work may also require travel to Washington, D.C. or agency regional offices in order to fully understand agency practices or, upon the project’s completion, to present findings.

The policy lab is open to all students at Stanford University, and will ideally attract both law students and non-law students from technical fields who can contribute a sophisticated understanding of the current trajectory of AI technology. For law students, past coursework or a strong background or interest in administrative law is highly recommended. Students from all parts of the University who wish to enroll in the policy lab may also consider taking Justice Cuéllar’s fall quarter course at the Law School, “Regulating Artificial Intelligence.”

Law students wishing to undertake R credit will perform additional research or take on additional tasks analyzing the issues and results of the collective research. R credit is possible only by consent of the instructor. After the term begins, and with the consent of the instructor, students accepted into the course may transfer from section (01) into section (02), which meets the R requirement.

See course details
Consent of Instructor Form

Faculty

Daniel E. Ho 4

Daniel E. Ho

  • William Benjamin Scott and Luna M. Scott Professor of Law
  • Professor of Political Science
  • Professor of Computer Science (by courtesy)
  • Senior Fellow, Stanford Institute for Human-Centered Artificial Intelligence (HAI)
  • Senior Fellow, Stanford Institute for Economic and Policy Research
  • Director of the Regulation, Evaluation, and Governance Lab (RegLab)

Clients & Deliverables

  • Client: Administrative Conference of the United States
  • Deliverables: Convening at NYU Law School; client briefing; and full report, “Government by Algorithm: Artificial Intelligence in Federal Administrative Agencies” (Jan. 2020).
Administering by Algorithm: Artificial Intelligence in the Regulatory State 2
ACUS AI Report