Sen. Gary Peters (D-Mich.), chair of the Senate Homeland Security and Governmental Affairs Committee, will introduce long-awaited legislation on Tuesday laying down new rules for how the government buys and uses cutting-edge artificial intelligence systems — a bill that could shape how the broader AI industry handles myriad risks associated with the technology.
The bipartisan PREPARED for AI Act, cosponsored by Sen. Thom Tillis (R-N.C.), would:
Make it easier for the government to buy new AI tools by streamlining AI procurement at federal agencies.
Launch pilot programs that promote agency flexibility and greater competition between potential AI contractors.
Lay down a series of rules that agencies and AI developers must follow if they plan to use AI — especially in areas where algorithms could violate civil rights or restrict access to crucial government services, such as health care, financial services, public housing, transportation and more.
Fully ban certain government uses of AI, such as those involving facial mapping for social control.
Peters said the intention in part is to use the government’s clout to ensure the safe spread of AI in America overall.
“There's a lot of talk in Congress right now about what sort of regulation we should have for the economy at large. That's a difficult task,” Peters told POLITICO in an interview. “But we can have a bigger impact by actually thinking about the federal government and our procurement process.”
The federal government is the largest single purchaser of goods and services in the United States, and Peters said that the baseline rules it sets for AI contractors will likely percolate out across the broader U.S. economy.
“The standards that we set for federal use are going to be a powerful message on how these companies and many others who are looking to do business with federal agencies should develop their technologies,” Peters said.
Greasing the skids: The bill aims to relax some restrictive provisions found in the government’s hidebound procurement processes. It would allow agencies to more quickly adopt AI systems before they’re out of date, and establish pilot programs that allow agencies to research and test specific AI tools before committing to their wide-scale adoption. It would also introduce procedures meant to increase competition between AI companies looking to partner with the government.
Peters compared the effort to the Pentagon’s plan to speed up the acquisition of advanced defense tech (a push that has achieved mixed results so far). He framed his bill as an attempt to expand the Pentagon’s more streamlined approach to all federal agencies.
Watching for risks: The PREPARED for AI Act also requires AI companies partnering with the federal government to craft systems that avoid risks to individuals’ civil rights or privacy. It would force agencies to audit and evaluate how AI companies are developing those tools, and would require companies to share data with the government about potential risks associated with the technology.
Peters acknowledged those provisions will increase the burden on agencies hoping to deploy AI systems, to some extent undercutting other parts of the bill meant to increase ease of AI adoption. But the senator noted that the current procurement process “is already very burdensome in many respects,” and that the government must ultimately procure and deploy AI systems responsibly.
Full stop: The bill also entirely bans agencies from using AI-enabled facial recognition systems to track an individual’s emotions and take action against them, and from using those systems to deduce or infer race, beliefs and opinions, trade union status, sexual orientation or other personal traits.
It also blocks agencies from using AI tools to classify, rate or score the “trustworthiness or social standing of an individual” — an apparent nod to the Chinese government’s system of social credit scores, which uses technology to carefully track the behavior of its citizens and reward or punish them.
Peters said those full-stop restrictions stemmed from expert testimony at committee hearings held over the past year on the government’s use of AI. “Most of the experts testified that these are very real concerns, and that we should address them directly at the outset,” Peters said.
Codifying cooperation: The legislation would codify the establishment of chief AI officers at each federal agency, as well as the creation of a cross-agency AI Council and an AI Governance Board. The Office of Management and Budget tasked federal agencies with standing up these AI-specific institutions earlier this year.
The path forward: Peters said he plans to mark up the PREPARED for AI Act sometime in July. “The quicker we get this out of committee, the quicker we can start moving on appropriate floor action,” he said.
A Peters aide told POLITICO that the senator is prepared to push for the bill to be considered as a standalone or to be included in one of the larger legislative vehicles that often move at the the end of the year, such as the year-end defense bill.