As policymakers increasingly turn to artificial intelligence to help inform decisions, questions remain about whether those tools can be trusted. A new Penn State study examined how legislators use AI to support evidence-based policymaking and identified what they need to trust it. The study, led by Jonathan Wright, assistant research professor at Penn State’s Prevention Research Center, found that policymakers see AI as a valuable tool for accessing research evidence more quickly, but they also raised concerns about transparency, accuracy and trust.

To better understand how policymakers are using AI, the researchers used a mixed-methods approach that combined interviews with 45 state legislators from 30 states and asked participants to test an AI assistant they developed, which allows users to search the Results First Clearinghouse, an evidence-based database of prevention and intervention programs. Unlike general AI tools, the assistant was restricted to information within the database and was not able to search the broader internet – to ensure that the search findings produced accurate, evidence-based information.

“Participants said AI could help them locate, summarize and understand research more quickly, making it easier to access evidence under tight legislative deadlines,” Wright said. “At the same time, they questioned whether AI-generated information could be trusted, citing concerns about transparency, reliability, potential bias and the possibility of inaccurate or misleading information.”

The findings, published in Prevention Science, suggest that policymakers want AI tools that are grounded in credible evidence, that clearly identify their sources and that are designed specifically to support research-informed decision-making.

“We’re seeing that policymakers are under a lot of pressure in general to make decisions about programs that affect children, families, and communities,” said Wright. “As they’re under this intense time pressure to make those decisions, AI is a very convenient tool to be able to find information that will allow them to make those decisions faster.”

As AI continues to evolve, Wright believes it will become an increasingly common part of policymaking. The challenge will be ensuring that these tools are designed to support accurate, evidence-based decisions.

“AI is not going anywhere,” Wright said. “Since it’s a tool that makes access to information easier, we expect that it’s going to find its way into policymaking.”

Looking ahead, Wright hopes this research will support the development of more specialized AI tools focused on topics related to public health. To this end, he’s led the development of the “AI for Informed and Responsible Evidence Use” (AIRE) protocol, a framework for building AI tools grounded in credible, evidence-based sources.

“It’s a development approach for building AI tools that are based on credible evidence, that’s shaped by user feedback, that’s transparent about their sources, and that’s improved through repeated testing,” Wright said. “The important thing is to create AI tools that use research as their evidence base and to restrict their responses to that evidence base.”

Wright said the promise of AI is not replacing the expertise of policymakers but helping them make better-informed choices.

“A lot of people are concerned about AI replacing human judgment,” Wright said. “What we hope is that if AI tools are developed well, the promise of AI and policymaking is not that it will replace human judgment, but that it can help public officials find and understand the best evidence faster, so their decisions are better informed.”

Co-authors on the study include D. Max Crowley, director of the Edna Bennett Pierce Prevention Research Center; and Alex Winters, Damon Jones, Jessica Pugel, Patrick O’Neill, Bethany Shaw, Sarah Hamel, Elizabeth Long, Michael Donovan and Taylor Scott, all with the Edna Bennett Pierce Prevention Research Center at Penn State. Jessica Pugel is with the Evidence-to-Impact Collaborative.

Funding for the research was provided by the National Institute on Drug Abuse, the National Institute of Child Health and Human Development, the Center for Socially Responsible Artificial Intelligence, the William T. Grant Foundation, The Pew Charitable Trusts and the National Science Foundation.