AI can be a game changer for regulatory reform
A recent panel discussion sheds light on the potential and provides a playbook for states
Above image created with ChatGPT.
Nearly every year I attend the Allied Social Science Associations (ASSA) meeting organized by the American Economic Association. It is the world’s largest annual gathering of economists. This year’s meeting was held in Philadelphia.
One of the most interesting panels on this year’s program was sponsored by the Society for Benefit-Cost Analysis. The panel was titled: “AI in Regulatory Analysis: Emerging Tools and Institutional Innovations” and featured:
Reeve Bull
Director of Virginia’s Office of Regulatory Management (ORM)
Casey Mulligan
Chief Counsel for Advocacy at the U.S. Small Business Administration and Professor of Economics at the University of Chicago
Will Rinehart
Resident Senior Fellow at AEI
and
Jonathan Wolfson
Wolf and Saber Advising and Vulcan Technologies
Patrick McLaughlin, a senior fellow at the Hoover Institution and a pioneer for using AI tools to measure and quantify regulation, moderated the discussion.
Reeve Bull’s ORM used AI tools to help achieve Virginia’s significant reduction in regulations:
The efforts of the newly created Office of Regulatory Management (ORM) contributed to the elimination or simplification of over 88,000 requirements and 12 million words of guidance, saving the Commonwealth over $1.2 billion. The Virginia ORM mandated economic analysis for all agency actions and focused on reducing burdens, helping the Youngkin administration achieve their goal of 25% regulation reduction.
Source: https://rtp.fedsoc.org/paper/the-virginia-model-how-the-commonwealth-built-a-best-in-class-regulatory-system/
Reeve made clear during the panel discussion that most of Virginia’s reform was not assisted by AI, but the state began to make use of the technology in 2025. Through his application of AI tools in Virginia’s significant reduction in regulations, Reeve identified five areas where AI proved to be most useful in reviewing regulations:
Finding duplication: regulations that overlap or restrict the same activity
Simplify regulation guidance documents: guidance documents are meant to help citizens better understand how regulations affect them, but can create many issues of their own— particularly if they remain obtuse and hard to understand.
Compare statute and regulation: Is regulation in compliance with statute? This seemingly trivial issue can have enormous consequences. Consider the case of universal recognition. Our licensing boards actually writing rules to implement the law? Or do licensing boards retain rules that are no longer legal. It can be time consuming to review this for humans, but Reeve saw that AI can play a big role here.
Perform preliminary cost benefit analysis: many states are lacking in performing a careful analysis of whether the costs associated with a new regulation exceeds the benefits.
Compare state to state requirements: this is clearly relevant for occupational licensing where you can see wide differences in licensing requirements for the same job across states. Good data are available, but there can be a bit of a learning curve with where to look.
Reeve, and all of the panelists, stressed that AI can not replace the work of humans. Humans should be reviewing and supervising the work of AI.
Jonathan Wolfson noted that most states do not have the resources to staff an agency to conduct rigorous review and analysis of regulations. Previous research by Patrick shows that regulatory analysis is often of poor quality. State governments would be well served devoting some taxpayer resources to sound regulatory review. States can obtain much greater productivity and impact if they complement these agencies with AI tools. Reeve’s five examples provide a good start and as AI improves it might be capable of performing additional tasks.
Will Rinehart spoke about a fascinating project he has been working on to estimate compliance costs for new AI regulations using platforms like ChatGPT, Claude, and Grok.
Source: https://www.williamrinehart.com/ai-research/#llms-for-cost-benefit-analysis
These estimates provide very useful guidance to states considering their own regulation of AI. States could also potentially replicate this methodology for better understanding the effects of new regulations.
Casey Mulligan made clear that AI is not replacing jobs at his agency. Instead, the Small Business Administration is actively hiring. He emphasized that AI is a complement rather than a substitute for human ingenuity. In his words, humans are needed to go “the last mile” and it currently too expensive or just not feasible to perform good regulatory analysis with machines alone.
Overall, a fascinating discussion. State policymakers would be well served to take lessons from Virginia’s experience. There are also promising new methods and exciting new partners like Vulcan Technologies that can assist states conduct meaningful and impactful reviews of regulations. Coupled with AI tools, state governments can make small investments in regulatory review that can make real and measurable differences for human flourishing.