Sam Altman Says AI Should Scare the Public, But Trust Companies to Do It Right
SAN FRANCISCO — Public concern about artificial intelligence is warranted, but people should trust AI companies to act responsibly as the technology gets more powerful, OpenAI chief executive Sam Altman said on Tuesday.
“The speed of the AI development has made it easier to imagine dire consequences if the technology isn’t kept in check,” Altman said at Salesforce’s Dreamforce conference in San Francisco.
“The world should have confidence that we’re going to do the right thing because it’s the right thing, and because we feel the weight of our responsibility,” Altman said.
And it is no longer difficult to imagine concerns about AI, he also said.
“It doesn’t take as much imagination as it once did for us to imagine how the situation could go wrong,” said Altman. “I think the world has a right to be afraid of AI,” he said.
His remarks came as a debate is building in the technology industry over how quickly advanced AI should be developed and who should take responsibility for managing its risks.
The debate erupted after former Anthropic researcher Jacob Coxon publicly warned that AI could potentially lead to the extinction of the human race by the end of the decade. Coxon said he departed over concerns about how the industry is handling standard artificial intelligence systems that are becoming increasingly capable and potentially self-improving.
The warnings were echoed by Anthropic chief executive Dario Amodei, who called for the pace of frontier AI development in the industry to be slowed down. In an essay in September, Amodei suggested steps like more input from independent evaluators and international coordination on AI safety. Altman later said he agreed the industry needed to “pace the frontier.”
But Altman also pushed back on the idea that commercial competition or international rivalry meant AI companies couldn’t act responsibly at Dreamforce.
He said AI firms should have enough time to develop the technology safely and coordinate on safety, but he insisted the industry must also earn public trust.
Conflicting opinions on regulation
The discussion at Dreamforce also revealed divisions among top technology executives over the role of government.
According to Jensen Huang, CEO of Nvidia, AI safety is more of an engineering problem than something that requires new laws.
“We don’t need new laws and regulations,” Huang said. Companies can innovate faster while keeping products safe and should hold off a release if they don’t have confidence that a system is ready, he argued.
Likewise, Meta CEO Mark Zuckerberg said that individual AI companies have the responsibility and incentive to make their systems safer and better aligned.
“Companies have both commercial and legal incentives to stop harmful behavior,” Zuckerberg said, adding that alignment will become more important as artificial intelligence systems become more capable.
His position differs from proposals to coordinate industrial action or strengthen external oversight. Meta also said it delayed the release of its Muse AI system while building safety and security measures.
A debate Silicon Valley is leaving behind
The dispute comes as AI companies are under increasing scrutiny over both the benefits and risks of ever more autonomous systems.
Amodei has warned that AI development could outstrip companies’ ability to understand and control the technology. His plan would include independent reviews and more coordination among AI developers.
Altman has said he supports some of that thinking, while Huang and Zuckerberg have stressed that companies can take safety measures on their own, without new government regulations.
The competing positions point to a central question in the AI debate: Can safety largely be managed by voluntary action on the part of technology companies, or will governments and other outside institutions eventually need a greater role?
For now, the big AI companies are split on how that balance should be struck, even as their leaders increasingly say the risks of more powerful systems deserve more attention.
