Symposium 2023
Featured Article
Algorithmic Rulemaking vs. Algorithmic Guidance
Peter Henderson & Mark KrassAlgorithms are coming to government. One legal question raised by this change is the extent to which the Administrative Procedure Act (“APA”) will regulate the use of algorithms as decision-support tools for agency adjudicators. Under the APA, “rules” are officially binding statements of policy subject to notice and comment as well as rigorous pre-implementation judicial review, whereas “guidance,” officially defined as non-binding advice, is effectively unreviewable. The implementation of algorithmic tools often occupies a gray zone between the two. To...
Generative AI Will Break the Internet: Beyond Section 230
By Graham H. RyanView PDFGraham H. Ryan, J.D., is a litigation and appellate partner at Jones Walker LLP, with extensive experience litigating complex commercial issues in all phases of litigation, appeals, and regulatory proceedings, including those involving technology, artificial intelligence, Section 230, and related matters. He holds an international designation as an Artificial Intelligence Governance Professional from the International Association of Privacy Professionals, and has been published on legal and constitutional issues arising from emerging Internet technologies.The law that “created the Internet” has...
Algorithmic Rulemaking vs. Algorithmic Guidance
Algorithms are coming to government. One legal question raised by this change is the extent to which the Administrative Procedure Act (“APA”) will regulate the use of algorithms as decision-support tools for agency adjudicators. Under the APA, “rules” are officially binding statements of policy subject to notice and comment as well as rigorous pre-implementation judicial review, whereas “guidance,” officially defined as non-binding advice, is effectively unreviewable. The implementation of algorithmic tools often occupies a gray zone between the two. To help clear the thicket, we provide a deep dive into the computer science and economics literature to provide a set of workable heuristics that help distinguish algorithmic rulemaking from algorithmic guidance. These heuristics align with best practices in the computer science literature and provide insights into agency incentives for adopting safer algorithms. We suggest that the specter of rulemaking may have value in nudging agencies toward best practices aligned with existing algorithmic safety recommendations. Specifically, avoidance of APA rulemaking may encourage agencies to prevent automation bias and other potential harms from algorithmic deployments. In this way, distinguishing algorithmic rules and guidance under the existing framework of the APA may dovetail with best practices in computer science.