Check our article at Tech Policy Press.
At the Harvard Kennedy School, my research team has been working on a practical problem in AI governance: how can policymakers compare hundreds of AI bills, executive orders, and regulatory proposals without reducing them to slogans?
AI policy is rarely about one thing. A single bill can touch national security, safety, civil rights, market competition, and industrial policy at the same time. Yet public debate often flattens these policies into one label: “pro-innovation,” “pro-safety,” “anti-China,” “anti-Big Tech,” or “anti-regulation.”
Our project used AI to make that complexity more visible. We built a framework that scores AI policy documents across multiple dimensions, using expert-informed rubrics, computational text analysis, and validation against a larger corpus of governance documents. Our goal is to show what the bill emphasizes, what it overlooks, and where different policy goals overlap, not to tell policymakers whether a bill is good or bad.
This Tech Policy Press essay was an unexpected byproduct of that work. When we benchmarked commercial AI models against our framework, we found that different systems read the same policy differently. Some models captured secondary policy dimensions. Others narrowed the analysis around the dominant headline category.
That finding matters because governments are already using AI for more than dramatic military scenarios. They are using, or will use, these tools to summarize, compare, and interpret complex documents. In that world, AI procurement is not just a technology decision. It is also a decision about what government systems are more likely to notice and what they may miss.
Hope you like the article. You can see it by clicking here.
My coauthors:
Isabel Adler is a dual Harvard Kennedy School master’s in public policy (MPP) and Tuck School of Business at Dartmouth (MBA) candidate. Formerly with the Knight First Amendment Institute.
Jeffrey Zhou is an A.B. candidate in Computer Science and Economics at Harvard College.
Claudio Mayrink Verdun is a Research Scientist at the Massachusetts Institute of Technology and holds a Ph.D. in mathematics and electrical engineering from the Technical University of Munich.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.