Meta’s Zuckerberg favours evaluators over slowdown for AI

Meta chief executive Mark Zuckerberg joined a chorus of technology executives weighing in on a global debate on AI.

SAN FRANCISCO - Meta Platforms chief executive officer Mark Zuckerberg said artificial intelligence labs should rely on independent evaluators and advisers to ensure that models are safe.

“Engaging independent evaluators and advisors is industry best practice,” Zuckerberg wrote on Sept 15 in a post on X, joining a growing chorus of technology executives weighing in on a global debate over the best way to prevent AI systems from harming humanity.

Some company officials have voiced support for decelerating the pace of AI development and advocated working together to achieve consensus. Meta decided on its own to slow work on the Muse AI tool, rather than ask peers to also check their own progress first, Zuckerberg wrote.

“Meta delayed shipping Muse for several months to focus on safety and security,” he wrote. “We didn’t call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us.”

A turbulent debate over AI’s economic and safety risks accelerated over the weekend with a 3,800-word essay from Anthropic PBC chief executive officer Dario Amodei, who urged a slowdown in the pace of development of the most advanced systems so researchers can better understand potential threats.

OpenAI chief executive officer Sam Altman and SpaceXAI chief executive officer Elon Musk – whose companies are bitter rivals to Anthropic – endorsed Amodei’s missive, and an OpenAI executive said the company was working with Anthropic and Alphabet’s Google on AI safety.

Zuckerberg’s post marked the chief executive’s first public foray into the latest round of discussions about how to address the escalating concerns, adding to comments from executives from across the AI industry and lawmakers worldwide.

Earlier on Sept 15, Nvidia chief executive officer Jensen Huang made similar remarks, saying safety should be a regular part of the development process, and companies can decide not to release things that are harmful.

“Safety is paramount,” Huang said during a Salesforce event. “In a lot of ways, it’s job one. However, safety is an engineering problem.”

If we reach a point when “the company is out of control and the product’s not going to be safe, take a pause and make sure you get it right,” he said. Huang also dismissed the need for new AI security regulations, arguing that market forces will help companies safely innovate.

“We don’t need new laws – we don’t need new regulations,” Huang said. He added that there’s a “false choice” about AI development being either safe or fast.

“You can definitely have both at the same time,” Huang said during a conversation with Salesforce chief executive officer Marc Benioff. “You pace yourself until you’re confident you’re releasing something that the market would appreciate. The market forces are already there.”

At the same Salesforce conference, OpenAI’s Altman likewise struck a calmer note, saying he thinks AI companies can develop the technology safely, without causing significant harm to people.

“I am very confident in our company’s ability – our industry’s ability – to do this safely,” Altman said.

While OpenAI is working with Anthropic and Google’s DeepMind lab to address safety issues, Altman said AI companies should be able to safely improve AI models on their own.

“I’m disappointed by how it’s been framed,” he said. BLOOMBERG