Joshua D. Blank, University of California, Irvine, School of Law, and Leigh Osofsky, University of North Carolina, Chapel Hill, are publishing Automated Agencies Minnesota Law Review, Vol. 104, 2023 in volume 104 of the Minnesota Law Review. Here is the abstract.
When individuals have questions about federal benefits, services, and legal rules, they increasingly seek help from government chatbots, virtual assistants, and other automated tools. Most scholars who have studied artificial intelligence and federal government agencies have not focused on the government’s use of technology to offer guidance to the public. The absence of scholarly attention to automation as a means of communicating government guidance is an important gap in the literature. Through the use of automated legal guidance, the federal government is responding to millions of public inquiries each year about the law, a number that may multiply many times over in years to come. This new form of guidance is thereby shaping public views of and behavior with respect to the law, without serious examination. This Article describes the results of a qualitative study of automated legal guidance across the federal government. This study was conducted under the auspices of the Administrative Conference of the United States (ACUS), an independent federal agency of the U.S. government charged with recommending improvements to administrative process and procedure. Our goal was to understand federal agency use of automated legal guidance, and offer recommendations to ACUS based on our findings. During our study, we canvassed the automated legal guidance activities of all federal agencies. We found extensive use of automation to offer guidance to the public by federal agencies, with varying levels of sophistication and legal content. We identified two principal models of automated legal guidance, and we conducted in-depth legal research regarding the most sophisticated examples of such models. We also interviewed agency officials with direct, supervisory, or support responsibility over well-developed automated legal guidance tools. We find that automated legal guidance offers agencies an inexpensive way to help the public navigate complex legal regimes. However, we also find that automated legal guidance may mislead members of the public about how the law will apply in their individual circumstances. In particular, automated legal guidance exacerbates the tendency of federal agencies to present complex law as though it is simple without actually engaging in simplification of the underlying law. While this approach offers advantages in terms of administrative efficiency and ease of use by the public, it also causes the government to present the law as simpler than it is, leading to less precise advice and potentially inaccurate legal positions. In some cases, agencies heighten this problem by, among other things, making guidance seem more personalized than it is, ignoring how users may rely on the guidance, and failing to adequately disclose that the guidance cannot be relied upon as a legal matter. At worst, automated legal guidance enables the government to dissuade members of the public from accessing benefits to which they are entitled, a cost that may be borne disproportionately by members of the public least capable of obtaining other forms of legal advice. In reaching these conclusions, we do not suggest that automated legal guidance is uniquely problematic relative to alternative forms of communicating the law. The question of how to respond to complex legal problems, in light of a public that has limited ability or inclination to understand complex legal systems, is a difficult one. There are different, potential solutions to this problem, which each present their own series of cost-benefit tradeoffs. However, failure to appreciate, or even examine, the tradeoffs inherent in automated legal guidance, relative to the alternatives, undermines our ability to make informed decisions about when to use which solution, or how to minimize the costs of this form of guidance. In this Article, after exploring these challenges, we chart a path forward. We offer policy recommendations, organized into five categories: transparency; reliance; disclaimers; process; and accessibility, inclusion, and equity. We believe that our descriptive as well as theoretical work regarding automated legal guidance, and the detailed policy recommendations that flow from it, will be critical for evaluating existing, as well as future, government uses of automated legal guidance.
Download the article from SSRN at the link.
