Lawmakers propose ‘blue-ribbon’ AI commission

A bipartisan group of lawmakers are backing creating a federal commission to develop a strategy for regulating artificial intelligence, the latest addition to a growing crop of legislative proposals tackling new AI tools like ChatGPT.

Reps. Ted Lieu (D-Calif.) and Ken Buck (R-Colo.) on Tuesday are introducing a bill that would require Congress and the White House to appoint 20 people across government, industry, civil society and the computer science field to an AI commission.

The group would issue three reports over two years to policymakers with recommendations for “mitigating the risks and possible harms” of AI while “protecting” U.S. tech innovation.

Lieu said in an interview that the “blue-ribbon commission” would give policymakers breathing room to understand what impact generative AI tools like ChatGPT will have on society and to receive input on what response may be needed before acting.

“It can be disruptive to society, from the arts to medicine to architecture to so many different fields, and it could also potentially harm us and that's why I think we need to take a somewhat different approach,” he told me on Monday.

Under the bill, appointments to the commission would be split evenly between Democrats and Republicans “to ensure bipartisanship.” The president would appoint 8 commissioners, while party leaders from both the House and Senate would collectively tap the other 12.

Members would need to have a background in government, industry, civil society or in computer science, but none of those groups would be allowed to make up a majority of the group.

Rep. Anna G. Eshoo (D-Calif.) is co-sponsoring the bill and Sen. Brian Schatz (D-Hawaii) is expected to introduce a Senate counterpart, Lieu spokeswoman Jenna Bushnell said.

Lieu said that while creating a commission would not preclude Congress from stepping in before receiving any recommendations, it would be “prudent” for lawmakers to hold off on “overarching legislation” until the group could weigh in.

“I just think we need some experts to inform us and just have a little bit of time pass before we put something massive into law,” said Lieu, one of a handful of lawmakers on Capitol Hill with a computer science background. Lieu first teased the idea in a January op-ed.

In the meantime, Lieu said, Congress could legislate in “discrete areas,” such as in national security. Lieu, Buck and other lawmakers recently introduced legislation to prevent AI systems from controlling the launch of nuclear weapons without “meaningful human control.”

Lawmakers previously introduced a flurry of bills to set privacy guardrails for AI tools, to require companies to vet their algorithms for biases and to create new regulatory agencies or departments to oversee digital services, but most have not passed.

But legislative efforts are gaining new life this year as scrutiny mounts in Washington about the rising popularity of AI chatbots like OpenAI’s ChatGPT or Google’s Bard.

The surge in AI interest has triggered a wave of hearings, briefings and meetings on Capitol Hill, where lawmakers have tapped a wave of academics and industry leaders to help get them up to speed on the tools, as my colleague Cat Zakrzewski and I reported over the weekend.

Lieu, one of dozens of lawmakers who attended a private dinner with OpenAI CEO Sam Altman last month ahead of his congressional testimony, said it was clear that night that “a number of members of Congress had never seen generative AI before.”

“Before we rush into any sort of massive congressional bill, I think it's important that we give members of Congress the opportunity to learn about this very complicated subject,” said Lieu.

But Lieu said it’s unreasonable to expect most of Congress to develop a deep technical understanding of AI to be able to legislate, as some industry leaders have suggested.

“Members of the Congress do deal with hundreds of different issues of which we're not experts on … You don't have to know exactly how AI algorithms work, but I think it's important to understand the output of these algorithms and what they potentially could do,” he said.