How AI can help reshape Congress
This week the U.S. House of Representatives Committee on House Administration convened an unusually well-attended hearing with a panel of government technology leaders from an alphabet soup of congressional agencies, including the Library of Congress (LOC), Government Publishing Office (GPO), Government Accountability Office (GAO), and the House Chief Administrative Office (CAO). The purpose of this gathering of the acronyms: to discuss how AI could make Congress govern more effectively.
Congress has been cautious in its adoption of AI compared to the governments of India, Italy, and Brazil, all of whom are turning to AI to transcribe and translate parliamentary proceedings and for sophisticated legislative record searching. Still, that’s not to say the U.S. has been totally complacent when it comes to AI adoption in the legislature: Last April, the House doled out 40 ChatGPT licenses (the only approved large language model for use on official devices) to a group of bipartisan staff and set up an AI Working Group to discuss how AI can help staff in their day-to-day work.
While many of the examples of AI use cited by the hearing’s panel were related to other innovations, such as using AI to digitize and search records at the Library of Congress or Copyright Office or translate text to speech for the blind, small experiments in “legi-tech” are underway.
Out of 150 offices surveyed by the House Administrative Office (which manages back-end operations for the House of Representatives), the most popular AI uses are “producing that first draft and then giving it to a human to go from there,” says John Clocker, the Deputy Chief Administrative Officer of the House. First drafts of testimony, witness questions, and speeches are popular ways of taking advantage of the technology’s ability to generate language. “If you have writer’s block, it’s going to get you over that writer’s block and it’s going to give you a good framework to actually customize in the Member’s voice,” he says.
The Government Publishing Office is using AI to supplement its proofreading team, freeing them up from routine tasks like recognizing the difference between the capital letter “S” in the State of Wisconsin and the lowercase New York state of mind to enable the agency to publish government documents more efficiently and with greater accuracy.
The Library of Congress is working with the Congressional Research Service, leveraging AI to assist analysts in creating high-quality summaries of pending bills for Congress.gov. The tool, which they anticipate will launch in March, will use natural language processing to assign bills to legislative analysts at CRS based on the summary and identify similar pending bills. The agencies are also experimenting with using AI to extract geographic information from within legislative text. Geographic information could aid in analyzing policy impacts, ensure equitable resource allocation, and enhance legislative oversight and public accountability by highlighting the regional focus of laws.
At the hearing, the Library’s CIO committed to explore whether the new toolkit will also make it possible to identify Member contributions to legislation. Right now, the public can see who sponsored the bill but not who wrote which lines.
Tools for better bill analysis could promote greater transparency and accountability and, at the same time, reduce duplication between bills and the ability to pull together provisions across different drafts to forge better legislation. However, at present these tools are only slated to be available to staff at the Congressional Research Service, not to Members of Congress, their staff, or the public.
To be sure, Congress faces greater risks than most organizations from organized cybersecurity attacks, but there is so much uncharted opportunity to use AI to make Congress more efficient and more effective and to transform how we make laws for the better.
The nonprofit POPVOX Foundation recently released a report on AI and legislatures that enumerated myriad ways Congress could be using AI, such as: 1) creating chatbots to enable the public to ask about a Member’s voting records or get help from their congressional office 24/7; 2) transcribing and summarizing all legislative activity and translating it into multiple languages; 3) responding to constituent queries to their Members with personalized responses; or 4) reviewing legislation to improve its clarity but, above all, its quality.
Imagine if Members and Committees leveraged AI to speed up the institution’s ability to do deep research and analysis in order to understand the problems it needed to solve and identify solutions to incorporate into legislation? Or, what if Members and Committees overcame the lack of data scientists employed by Congress to crunch large quantities of data faster to gain a deeper understanding of major issues from climate change to poverty and healthcare? Better yet, suppose they engaged the American people in conversation about their experiences to inform the crafting of legislation. Of course, there are risks from bots spreading disinformation, but with the ability of generative AI to distinguish fact from falsehood and summarize large quantities of text, meaningful two-way conversations between government and the governed are becoming a practical reality.
By embracing responsible use of AI and safeguarding security and privacy, Congress has the opportunity both to modernize and speed up its operations but also to deepen its connection with the American people, ensuring that the legislative process becomes more transparent, efficient, and, above all, responsive to the people Congress serves.
(25)