WASHINGTON — Sen. Bernie Sanders (I-Vt.) introduced legislation Wednesday to pause the development of artificial intelligence until the government can make sure it won’t wreak havoc on society.
The Ban Artificial Superintelligence Act would create a new federal agency that would also ban companies from creating “superintelligent” AI beyond the reach of human control.
“We cannot simply sit by idly and wait for advanced AI to cause the next 9/11 or Chernobyl or a global pandemic or a financial crisis before we take action to protect the American people and, of course, the people of the world,” Sanders told reporters on Wednesday.
Sanders and Rep. Greg Casar (D-Texas), chair of the Congressional Progressive Caucus, announced plans for the legislation earlier this month amid warnings from within the industry that companies like OpenAI and Anthropic were losing control of AI “agents” they’d created. Some in the industry even claimed the technology could wipe out humankind within a decade.
The apocalyptic warnings rapidly changed the politics of AI, with Sanders and Casar offering the most aggressive response and Republicans in Congress supporting more modest proposals. Lawmakers seem far from a consensus, and Congress is unlikely to take action before November’s midterm elections.
President Donald Trump, meanwhile, has called the warnings a hoax and said slowing AI development would merely cede global competition to China.
Roberto Schmidt via Getty Images
“I understand that I may be called an alarmist for introducing this bill, but I’d rather be called an alarmist than a father and grandfather who is asleep at the wheel,” Sanders said.
Sanders also discussed the near-term economic threat posed by AI.
“In my view, it is likely if we allow things to continue the way they are, that there will be no more truck drivers in America, no more cab drivers, no more bus drivers, and we will lose some 6 million transportation jobs to driverless vehicles,” Sanders said. “In 10 years, it is quite possible that there will be no more factory jobs or warehouse jobs, because very sophisticated robots will be doing those jobs.”
Casar said the ban on superintelligence would work in part by prohibiting AI companies from using “recursive self-improvement,” or letting AI models build more advanced versions of themselves.
“This is not a radical notion. All it does is take us back a couple years in time when humans, rather than AI, were in charge of AI research,” Casar said Wednesday.
The Department of Artificial Intelligence, Casar said, would vet AI models at various stages of development, including before they’re released.
Anthropic and OpenAI said models developed with recursive self-improvement this summer escaped “sandbox” training programs and took unauthorized actions on the open internet in order to solve training problems the companies had set.
Dario Amodei, the CEO of Anthropic, wrote this month that recursive self-improvement could further outrun human understanding of new AI models. He said humans already don’t well understand what’s going on in current AI models developed through recursive self-improvement.
“Despite all the progress, we still only understand a tiny fraction of what goes on inside these models,” Amodei wrote in an essay calling for an AI slowdown, an idea endorsed by other leading tech CEOs. (Amodei and other industry leaders have not endorsed the Sanders proposal.)
Some have said the tech industry’s warnings are hype, and a mere push for advantageous government regulation. Sanders acknowledged the industry is led by billionaires who want to make themselves richer, but he’s clearly taking the warnings of danger at face value.
“If the leaders of the major AI companies acknowledge that they are losing control of this extremely dangerous technology, it is irresponsible, in my view, for society. was society to allow them to move forward and make these products even more advanced,” Sanders said.