.jpg)
Sam Altman said a U.S. agency could set standards for proper AI-system safety testing before a system is released. Jim Lo Scalzo. Rising con...
 |
| Sam Altman said a U.S. agency could set standards for proper AI-system safety testing before a system is released. Jim Lo Scalzo. |
The agency could be charged with granting licenses for AI platforms, setting operating standards and enforcing compliance with the rules, according to proponents including Sam Altman, chief executive of ChatGPT creator OpenAI.
It is one of many ideas being kicked around in Washington as lawmakers contend with a new technology with humanlike abilities to complete an array of tasks, including holding conversations and creating videos—but which also could be used to commit sophisticated crimes and spread false information.
“The word ‘guardrails’ has been used repeatedly,” said Rep. Anna Eshoo (D., Calif.), who helped arrange for Mr. Altman to meet with House Speaker Kevin McCarthy (R., Calif.), House Minority Leader Hakeem Jeffries (D., N.Y.) and dozens of others members following his appearance this week on Capitol Hill. Other tech leaders and some Republicans are wary of creating a new bureaucracy.
“Anybody who proposes a government review commission that would have to approve things is talking about a regulatory body with an awful lot of rules that we don’t know how to write right now,” Eric Schmidt, the former CEO of Google who chaired a congressional AI commission, said in a recent interview. He said lawmakers should focus on clear threats, such as an AI-driven deluge of false information online.
After Altman’s testimony Tuesday, the Computer & Communications Industry Association, a trade group that includes Alphabet Inc.’s Google and Amazon.com, warned against “creating additional layers of bureaucracy.”
The idea of a technology regulator dates back years, stemming from lawmakers’ feeling that major internet platforms have become central to Americans’ daily lives without the kind of oversight that applies to other critical industries such as finance, telecommunications or medicine.
“We are way behind the curve, but that’s often where we reside,” said Sen. Michael Bennet (D., Colo.), who plans to introduce a bill Thursday creating a five-member Federal Digital Platform Commission that would develop internet platform codes of conduct with industry input. “We need to have an administrative agency that can iterate as the tech iterates.”
Altman said AI systems “above a certain scale of capabilities”—for instance, the ability to manipulate a person’s behavior—should be licensed. A federal agency could set standards for proper safety testing before a system is released and be empowered to take the license away from companies who don’t comply, he said.
Altman also urged the government to advocate other countries take similar steps, citing the example of the International Atomic Energy Agency, a global body that promotes nuclear safety. Sen. Lindsey Graham (R., S.C.), the top Republican on the Senate Judiciary Committee, agreed with Mr. Altman’s comparison.
“You just can’t go build a nuclear power plant…you have a Nuclear Regulatory Commission,” he said. “I just don’t understand how you could say that you don’t need an agency to deal with the most transformative technology, maybe ever.” Other Republicans weren’t persuaded. “Having seen how agencies work in this government, they usually get captured by the interests that they’re supposed to regulate,” said Sen. Josh Hawley (R., Mo.).
Several Democratic senators praised the idea at the hearing, and the Biden administration has begun examining whether checks need to be placed on AI tools. Senate Majority Leader Chuck Schumer (D., N.Y.) is leading discussions on a bipartisan AI bill that he says will maintain American leadership in innovation while ensuring AI is used responsibly and transparently. The scant details he has released don’t mention the creation of a new agency.
Eshoo cautioned that congressional efforts on the technology are nascent. “If you want to get a comprehensive bill, it’s early,” she said. Although no dedicated AI agency yet exists, the Biden administration has said it would apply existing laws to AI in areas including lending, employment decisions, fraud and competition.
The U.S. Copyright Office has launched a review of concerns that ChatGPT and other AI systems are unfairly trained on copyright material, such as songs or books. Alondra Nelson, a former Biden White House official now at the progressive Center for American Progress think tank, said the administration also could look to use the federal government’s procurement authority to set standards for AI systems that it uses.