ACUS hosted the first panel of a multi-part symposium on Artificial Intelligence in Federal Agencies on Thursday, June 25, in partnership with the Institute for Technology Law and Policy at Georgetown University Law Center. Over 180 attendees—many from agencies and the Hill—tuned in.
The first panel, moderated by Hillary Brill of the Institute for Technology Law and Policy, featured the authors of the ACUS-commissioned report, Government by Algorithm: Artificial Intelligence in Federal Administrative Agencies: California Supreme Court Justice Mariano-Florentino Cuéllar, Stanford Law Professors David Engstrom and Dan Ho, and NYU Law Professor Cathy Sharkey.
Justice Cuéllar opened the panel by discussing the history of AI in federal agencies and described how governmental use of AI pushes up against some core constitutional and administrative law doctrines, including due process and reason-giving in agency decision making. Justice Cuéllar emphasized the need to grapple with certain fundamental questions, including determining what tasks can be delegated to AI programs and how we can ensure the legality and morality of government’s AI uses.
Discussing the role of AI in agency enforcement, Professor Engstrom explained how AI-based mechanisms can help agencies direct their resources more efficiently. “All of these enforcement tools are united by this common focus on shrinking the haystack of a pool of violators,” he noted. Engstrom also mentioned the growing importance of accountability in judicial review, and the way that further use of AI-based tools may force a reexamination of the “hiving off” of judicial review of enforcement decisions, a reference to the Supreme Court’s decision in Heckler v. Chaney which generally prohibits judicial review of agencies’ enforcement decisions.
Professor Ho explained how agencies are using technology to ensure fairness and consistency in adjudication. He also addressed concerns related to the governance of AI, including the possibility that sophisticated parties may attempt to “game” agency algorithms through adversarial machine learning.
Professor Sharkey discussed the role of AI at the Food and Drug Administration (FDA). As the world’s leading drug regulator, the FDA hosts a vast amount of data—a critical component of AI usage. Sharkey emphasized that “the potential of being able to harness immense data sets using these tools is mind boggling.” She described the FDA’s efforts in developing internal AI expertise as a kind of “internal incubator” or “regulatory sandbox.”
The discussion among the panelists was wide ranging and touched on the dangers of bias in AI applications, opportunities for agencies to design AI-based tools in house, and designing AI programs that offer clear explanations for the conclusions reached.
ACUS looks forward to the continued exploration of the nexus between AI and administrative procedure, including a forthcoming report from University of Pennsylvania Law School Professor Cary Coglianese and the continued compilation of resources on our website. The next panel, Artificial Intelligence and Administrative Law Doctrines: Challenges and Opportunities, is scheduled for Thursday, July 9, from 2:00 p.m.–3:15 p.m. ET (online registration available here). Moderator David Vladeck will lead panel members in a discussion of the interplay of federal agencies’ use of AI and core administrative law and constitutional law doctrines. Future panels will discuss AI from the perspective of agency officials, and potential bias in government’s use of AI.