As lawmakers eye regulation, AI could reshape Congress itself
Lawmakers and staff grapple with how AI could change their own workplace at Congressional Hackathon
In the same week that tech industry leaders and senators came to a consensus that government should regulate artificial intelligence, a group of lawmakers, staff and advocates leaned into the potential powers of AI to change the legislative branch and how it works.
Rather than something to be feared, AI is a tool that can help streamline constituent services, demystify complex legislation and improve civic discourse, according to a series of ideas presented at the fifth Congressional Hackathon, a meeting of policy wonks and data nerds that unfolded over several hours Thursday in the Capitol Visitor Center auditorium.
“I understand the fear,” said Speaker Kevin McCarthy, R-Calif., who co-hosted the event with Minority Leader Hakeem Jeffries, D-N.Y., referring to anxieties around AI. “But I also want to understand the opportunity. And I don’t want to see things only happening in the private sector.”
The Congressional Hackathon was first convened in 2011 to bat around ideas for improving the inner workings of the Capitol, and subsequent events were held in 2015, 2017 and 2022. This year’s hackathon was not solely about AI and its potential for good and evil, but the issue permeated much of the discussion and has increasingly demanded attention on the Hill. Critics have griped that Congress has been too slow to grasp the rapidly evolving implications and risks of AI, leaving lawmakers scrambling to catch up.
On Tuesday, a Senate subcommittee held a hearing on AI oversight, at which tech executives — including the president of Microsoft, which has invested billions in the parent company of ChatGPT — urged lawmakers to rein in AI, especially systems that control critical infrastructure.
The next day, Elon Musk and Mark Zuckerberg, as well as other top AI experts, were on the Hill for a closed-door meeting with senators. Majority Leader Charles E. Schumer, D-N.Y., who summarized the forum for reporters, said the parties agreed that Congress should play a role in regulating the technology but disagreed on how.
And on Thursday, the House Administration Committee hinted at how those larger debates could hit close to home for lawmakers in their own workplace. While generative AI holds the tantalizing promise of “dramatically increased efficiency” for rank-and-file staffers, it also poses questions of ethics and transparency, a brief report warned. The committee has asked congressional support agencies to provide monthly updates on how they plan to use AI.
But the mood at the hackathon was considerably more buoyant. While some participants in the public event expressed their anxiety over the potential for misinformation spread by AI, the majority saw opportunities to harness the technology.
Dylan Irlbeck, a congressional innovation fellow with the Senate Finance Committee, spoke about the potential for using large language models, which can scrape and quickly process large amounts of text from the web, to quickly write bill summaries that appear on Congress.gov, which are currently authored by the Congressional Research Service.
“The problem is that CRS cannot and does not provide these in a timely way,” Irlbeck said. “We have this new technology — large language models — that have a lot of amazing properties that make them really good at summaries, including language understanding … contextual awareness, scalability … customization by task and also processing speed. They can generate these summaries in mere seconds.”
Irlbeck said AI could be used as a “copilot, not autopilot.” CRS analysts would need to approve the AI-generated summaries, but would not have to undertake the more rigorous process of writing them.
Andy Curran, CEO and founder of the WeVote Foundation, previewed a social network set to launch in early 2024 that would create hyper-focused forums on which users could comment on specific pieces of legislation. That legislation would be summarized and simplified using AI tools to minimize confusion and encourage participation among users, according to Curran.
And Lars Erik Schonander, a policy technologist at the Foundation for American Innovation, described a tool he built that uses AI to compare House Rules packages from one Congress to the next. A similar tool might one day be used to make appropriations bills — which he called “text-file hell” — more transparent by quickly extracting numbers from what are often dense and complicated documents.
As the afternoon wore on, attendees swapped ideas from their breakout groups, covering not just the new emergence of AI, but old problems that have dogged Congress for years, like how best to communicate with constituents.
Rep. Derek Kilmer, D-Wash., made an appearance. The member of the House Administration panel, who has spearheaded efforts to modernize Congress, fielded questions about time-honored topics.
Asked about congressional scheduling and whether it would be better for collegiality if members and their families lived in D.C., Kilmer responded, “You shouldn’t have to have your family sleep here to not be an asshole.”