Elon Musk, Bill Gates, Mark Zuckerberg and other technology heavyweights debated the possibilities and risks of artificial intelligence Wednesday in a closed-door meeting with more than 60 U.S. senators who are contemplating legislation to regulate the technology.
Musk, the chief executive of Tesla and owner of X (formerly Twitter), warned about what he views as AI's potential to threaten humanity, according to a participant. Microsoft co-founder Gates said the technology could help address world hunger, said Sen. Chuck Schumer (D., N.Y.), who convened the session.
Other speakers included Facebook founder Zuckerberg and the CEOs of Google, Microsoft, Nvidia and IBM, along with union leaders, civil-rights advocates and others.
Schumer at one point asked the guests if they agreed that the government needed to play a role in regulating artificial intelligence. Everyone present raised their hands, Schumer said during a break in the daylong session.
"No one backed off in saying we need government involvement," Schumer said in an interview after the session ended. "They understood that there needed to be government responsibility, because let's say even these companies would be willing to install guardrails on themselves -- they'll have competitors who won't."
Despite that consensus -- and Schumer's vow to move toward passing legislation within months -- the meeting also laid bare some of the tension points ahead.
One debate centered on so-called "open-source" AI systems that are available for the public to download and modify. These systems allow companies and researchers to tap in to AI technology similar to the models that power ChatGPT without spending millions of dollars to train them.
Tristan Harris, head of the nonprofit Center for Humane Technology, argued that bad actors can abuse open-source AI systems, including the Llama 2 model recently released by Meta Platforms, the company led by Zuckerberg, according to people in the room. Harris said his nonprofit was able to get the Llama 2 model to provide instructions on how to create dangerous biological compounds, these people said.
Zuckerberg parried back, saying similar instructions can already be found on the internet, the people said. He added that open-source models could pose dangers, but that Meta was trying to build this technology as safely as possible, the people in the room said.
Zuckerberg told senators in his opening statement that open source "democratizes access to these tools, and that helps level the playing field and foster innovation for people and businesses," according to excerpts released by Meta.
Another point of tension related to workers who see AI as a potential threat to their jobs. Sen. Maria Cantwell (D., Wash.) recounted a moment where the head of the Writers Guild of America West, Meredith Stiehm, described the views of members who are on strike seeking a new contract with Hollywood studios in part to address those fears.
"That was, like, two sides right there," Cantwell said. "Lots of different viewpoints."
Some executives in the room argued that adding too many safety guardrails on AI systems risked American leadership in the technology. Another contested topic: the possibility that future AI systems could wipe out humanity.
Deb Raji, an AI researcher who attended Wednesday's meeting, said she urged the room to focus on current-day harms, including biased decisions in housing, hiring or criminal sentencing, that can come from hastily deployed AI systems.
She also advocated for building auditing methods for AI systems now before companies develop even more powerful AI systems.
The closed-door nature of the session drew criticism from some quarters. Sen. Josh Hawley (R., Mo.) questioned whether the meeting was designed "to prevent senators from asking tough questions the CEOs don't want to answer," and called on Schumer to bring AI legislation up for a vote.
"This is a lot of song and dance that covers the fact that actually nothing is advancing," Hawley told reporters Wednesday.
Schumer, who organized the meeting with a bipartisan group of three other senators, said the format was designed to allow a frank debate.
"There were people who have created AI systems and there were people who had problems with parts of the AI systems," Schumer said. "They talked directly to each other, they answered one another."
The tech leaders arrived just before 10 a.m. on Wednesday, stepping out of tinted SUVs and strolling past a phalanx of cameras -- with the exception of Musk, who arrived in a jet black Tesla sedan. They gathered around a table with name cards inside the marbled Kennedy Caucus Room.
Most of the tech leaders avoided the cameras stationed outside. Musk told reporters as he departed that he didn't believe Congress was ready to regulate AI, while adding that it also made sense for them to study the issue before writing legislation.
"I saw very little disagreement actually," he said of the meeting in a video posted by CBS News.
Schumer described Wednesday's meeting as the first in a series designed to develop legislation addressing rapid adoption of powerful artificial intelligence tools.
Lawmakers are far from consensus on what that legislation will say, but Schumer's involvement gives the effort more juice than Congress's previous tech-regulation efforts, many of which the majority leader never called up for a vote on the Senate floor.
One open question is what areas the legislation will cover. Issues previously raised by participants in Wednesday's meeting cover a range of potential concerns involving esoteric areas of law. Among them: Copyright violations, privacy invasions, racial discrimination, economic ties with China, and the use of AI by the government for military or other purposes.
"We'd like to get as comprehensive a proposal as we can, but if we can't reach agreement on certain areas, we'll still try to move forward on the ones where we can reach agreement," Schumer said. "It's such a cosmic issue with such consequences for society that you can't run away from it.
