Simplifying Capitol Hill

Congressional legislation isn’t meant for light reading — the contents can be dense and lengthy. But a team of Carnegie Mellon University graduate students created an artificial intelligence tool to help cut those bills down to size.

The Heinz College of Information Systems and Public Policy students teamed with Washington-based nonprofit Bipartisan Policy Center (BPC) to improve on the Congress.gov website by using ChatGPT’s technology to streamline its search and analysis possibilities.

“Let’s say there’s a defense authorization bill, and it’s very long,” says Jack Vandeleuv, a member of the team that created the app in fall 2023. “The analyst might use the tool to ask, ‘Is there anything in this bill that supports shifting arms to Ukraine?’ and the tool can pull that information, scan the results, provide an answer and link the answer to the source content in the bill.”

Jack and the rest of the team — Li-li Chen, Kathy Chiang, Zhichen Li and Sylvia (Yiping) Zhang — were led by faculty expert Andrew Garin, assistant professor of economics at Heinz College, and partner Tom Romanoff, then the director of the technology program at BPC, to create an AI tool that can search and analyze thousands of pages of congressional bills to help policy analysts conduct research.

The students built on the work of a previous Heinz College capstone project from spring 2023. The initial team of students conducted user research interviews, identified use cases and developed a desktop application that would summarize a bill.

The new tool can filter, analyze and summarize content using the past 10 years of bills currently in the Library of Congress’s Congress.gov database. It’s currently being piloted by a few users within BPC, Romanoff says, and news of its capability has started to create a buzz among his colleagues both inside and outside of BPC.

“People from several offices and committees are asking if they can get access,” Romanoff says, adding that he is rolling out the tool at a deliberate pace while addressing some issues. Like other generative AI tools, the app has problems with hallucination, when the AI system generates inaccurate or nonsensical responses. That issue can be addressed through an evaluation stage and it’s one of the reasons Romanoff has not yet released the tool broadly.

“We can all see the promise of generative AI technology, but we have to be realistic about its capacity,” Romanoff says.

Before the app becomes more widely available, he expects to provide education about how it can be useful and exactly what its shortcomings are — for example, that it should not be used as a definitive source for decision making, and to explain about hallucinations and the need for fact-checking.

Similarly, one of Jack’s takeaways from the experience is an appreciation for the potential of generative AI applications.

“You start to see a future where the average person on the street is better able to understand what Congress is doing because there’s a tool that can synthesize and explain complex topics that might be in a thousand-page bill,” he says. “We’re not there yet, but it’s exciting to see things trend in that direction.”

Demystifying AI

The Bipartisan Policy Center isn’t just working with Carnegie Mellon to make Congress more accessible — it’s also working to make artificial intelligence more accessible to members of Congress.

In March 2024, BPC, CMU and Google — the program’s lead financial sponsor — began working on the AI 101 Education Initiative, a literacy campaign designed to demystify the technology for policymakers.

Having informed government officials will be pivotal in the future, says Tom Romanoff, former director of the technology program at BPC.

“AI 101 is not aimed at any political outcomes, and it’s not partisan — it’s universal,” Romanoff says. “People are alarmed about AI and its potential impact on jobs and on our society. We’re bringing in experts to explain exactly what AI can and cannot do.”

The partnership combines BPC’s policy expertise with CMU’s educational and research capacity and Google’s ability to create resources and a platform for delivering the program. That collaboration is key to creating trust among stakeholders and credibility for the initiative.

The workshops will be shared with federal public servants initially, and eventually with legislators at the state level.

“AI 101 is a good example of how tech, government, and universities can work together,” Romanoff says. “We need to all be speaking the same language when we talk about AI, and those conversations need to happen today, because the technology is advancing so quickly.”