With AI becoming more common in research and policymaking, the Congressional Research Service (CRS) is asking Congress for $1.6 million to build its own closed AI model, citing accuracy and confidentiality.
CRS is a nonpartisan research agency within the Library of Congress that helps Congress by providing objective policy and legal analysis. Now, CRS is requesting additional funding to improve its research tools and resources, including advanced AI models, to continue providing lawmakers with timely and reliable information.
Although Representative Julie Johnson (D-TX) said there is no substitute for CRS staff, the agency’s productivity can be enhanced. However, AI should not replace human judgment or become the sole source of congressional information, she said.
“It seems troubling to me that the government of the United States Congress would rely on ChatGPT for all of our data,” Rep. Johnson said.
When CRS tested AI tools for drafting bill summaries, fewer than 3% of the generated responses met its standards for accuracy, clarity, objectivity and relevance. Rep. Johnson said the reliability of publicly available AI models, such as ChatGPT, is also “alarming.”
“I'm stunned by that number,” Rep. Johnson said. “I bet many of my colleagues up here are stunned by that number, and I think it really highlights that we as policymakers can't have a false sense of security of the accuracy of AI.”
CRS Director Doctor Karen Donfried cited this as a major reason why the agency needs a closed AI system.
“If the library were to develop that AI enterprise platform, it would serve the entire legislative branch, and the library is the repository of that data,” Donfried said. “The implications are far beyond the library and CRS.”
Donfried said using open systems also adds a risk of breaching confidentiality, but a closed system requires higher levels of Large Language Models that need to be trained to protect sensitive information and maintain strict confidentiality standards.
“That's a pledge we have to you, we will always treat any congressional information in a confidential manner,” Donfried said.
Rep. Johnson stressed that if Congress were to provide this funding, it would also need to be vigilant about data sourcing in these closed models, and part of that would include adding guardrails on how the models are trained and what information they are allowed to access.
“They work at their best when they're narrowly tailored to a specific function, or within a specific data set,” Rep. Johnson said. “So it is incumbent upon us to fund the library's request in tandem with this to make sure that we can have our own internal AI model that's designed for the maximum productivity for CRS.”
