スタンフォード HAI、行政の「政策スラッジ」削減に AI ツールを提供

Bureaucracy moves slowly, and decades of outdated provisions clogging the bylaws of cities and states only make it worse. Sometimes dubbed “policy sludge,” such obsolete processes and stale reporting requirements can trap civil servants in red tape and cause program dysfunction.

Scholars at the Stanford Institute for Human-Centered AI (HAI) and Stanford RegLab are working with cities and states across the country to find and cut out this sludge. They developed an AI tool to examine 500 million words of state statutes across all 50 U.S. states to reveal common patterns in the sludge, which they detail in a new paper, “The Abundance of Reports and Incapacity of States,” forthcoming in the Yale Journal on Regulation. RegLab also partnered with the states of New York, California, and Maryland to help these governments start paring these provisions from their books to speed the pace of government.

“This is a direct example of how AI can benefit millions of people,” says Daniel E. Ho, faculty director of the RegLab,associate director at HAI, and a Stanford Law School professor. “These tools can clean up outdated requirements, reduce burdens on civil servants, and enable government to serve regular people who don’t want to face delays in filing for a new business, building a house, or securing a license. We hope that more states and cities will adopt this approach to clear clutter.”

Based on this work, New York Gov. Kathy Hochul issued an Executive Order this month, directing state agencies to carry out a “regulatory reset” to remove outdated requirements, burdensome fees, and obsolete and unnecessary reports and commissions. “People have lost confidence in their state to be able to deliver the way they want it to,” Hochul said in a recent New York Times story about the order. “We have to shake it all up – be creative, and rethink government on a daily basis.”

Mapping the Clutter

In the new paper, coauthored by Ho and RegLab researchers Emily Robitschek, Ananya Karthik, Gabe Malek, and Derek Ouyang, the team compiled a corpus of 500 million words of state statutes. They developed an AI system to methodically scan through all provisions of the code to identify reporting requirements, commissions, and fees.

They tested and validated the tool in collaboration with the San Francisco City Attorney last year; the findings of that project led to legislation that streamlined over a third of San Francisco’s reporting requirements.

The scholars are working with other state partners to understand the prevalence, growth, costs, and benefits of such requirements. RegLab worked with Maryland to identify and analyze the estimated benefits and costs of reporting requirements across four state agencies. In California, the team collected information about how often reports are downloaded and validated the findings against a tracker of reports by the California legislature. The cross-state analysis also allowed the team to understand how such requirements vary by institutional factors, like government size and political representation.

“The AI tool built by the RegLab team was instrumental in New York’s Regulatory Reset, enabling us to convert unwieldy legalese into digestible datasets that will empower agency staff to systematically review mandated reports and identify opportunities for reform,” said Zoe Jacobs, director of regulatory reform and delivery in the office of Gov. Kathy Hochul.

Key Findings: More Bloat, Less Value

The cross-state analysis and government collaborations have revealed several striking findings.

First, reporting requirements have ballooned over time. In California, reporting requirements grew by 400% from 2000 to 2025. In Maryland, it could take up to 14 weeks to *read* reports, when the legislative session is only 13 weeks.

Second, the explosion of reporting requirements means many reports are never filed. In California, 30% of ongoing reports may have never been completed. Civil servants themselves question the utility of these reports: Maryland agencies identified 20% as candidates for elimination or consolidation.

Third, the impact and costs of reporting requirements can vary dramatically. A single report was calculated to consume 3,500 staff hours and over $870,000 to produce. Others cost a few hours of staff time and are viewed thousands of times. “States need a rational process to clean up clutter while identifying and preserving reports that add value,” said Malek, who served as a law student fellow in the Maryland governor’s office working on the project. “That process begins with better cost and impact data.”

Fourth, the team was able to test common conjectures about government bloat. Ezra Klein and Derek Thompson’s book Abundance), for instance, claims blue states are more bureaucratic. Ho and his coauthors find that reporting requirements are indeed more prevalent in Democratic states, but this partisan difference is small relative to the scale of reporting requirements. In other words, the problem exists across party lines.

The project also surfaced a myriad of obsolete provisions, underscoring the need for code cleanup. New York still has a provision on the books that requires the Board of Regents to report on actions taken against “subversive” teachers, a Red Scare measure found unconstitutional by the Supreme Court in 1967.

“The original idea of codification – the process of compiling statutory law – was to make law legible and accessible,” Ho said. “What this work shows is that if left unaddressed, policy sludge can have the exact opposite consequence, making the law illegible and programs inoperable.”

Giving Tools to Governments

Based on extensive engagement with jurisdictions around policy sludge, the RegLab team is also providing a model state statute to enable jurisdictions to get a handle on reporting requirements going forward. The model statute, for instance, provides for automatic sunsetting, a digital repository, and lightweight tracking of costs and benefits of reporting obligations.

The team also released results of the full scan across the 50 states and a website to enable anyone to begin to explore reporting requirements. In addition to New York’s Executive Order, state action is underway closer to home, as the state of California is using this data to convert paper reports into digital dashboards where possible.

Ho noted, “We hope this public resource triggers thoughtful, human-centered analysis to determine what processes serve people, and which ones trap civil servants and citizens in red tape.”

Authors:

Daniel E. Ho* is the William Benjamin Scott and Luna M. Scott Professor of Law, professor of political science and of computer science (by courtesy), senior fellow and associate director at Stanford HAI, senior fellow at the Stanford Institute for Economic Policy Research (SIEPR), fellow at the Center for Advanced Study in the Behavior Sciences, and director at the Stanford Regulation, Evaluation, and Governance Lab (RegLab). *

Preeti Hehmeyer* is the executive director at Stanford RegLab. *

Gabe Malek* is a JD candidate at Stanford Law School, a Knight-Hennessy Scholar, and law student affiliate of Stanford RegLab. *

Derek Ouyang* is a research director at Stanford RegLab. *

Ananya Karthik* is a research fellow at Stanford RegLab.*

*This work was supported in part by Stanford HAI and Stanford Impact Labs. Stanford RegLab receives general support from the William and Flora Hewlett Foundation. *