Mark Zuckerberg broke the centralization of AI power
Mark Zuckerberg has hit out at Anthropic and OpenAI, two of the world’s most valuable AI startups, saying the way they developed AI software could lead to a centralization of power that would limit people’s access to transformative technology.
In an interview with The New York Times on Tuesday, Meta’s 42-year-old CEO did not refer to Anthropic and OpenAI by name. But he said that if leading research labs wanted to develop AI technology in a tightly controlled way — as Anthropic and OpenAI have done — it would be akin to “abandoning our values” in American technology development and stifling innovation.
“So much of the discourse from the many other labs that they’re developing is overwhelmingly filled with doom,” Mr. Zuckerberg said. “There needs to be a voice or a few voices that bring realism to this debate.”
The comments from Mr. Zuckerberg, who rarely gives interviews to mainstream news organizations, escalate a debate over how artificial intelligence software should be built that has divided Silicon Valley.
Technologists such as Dario Amodei, CEO of Anthropic, and Sam Altman, CEO of OpenAI, have said that some of today’s leading AI models are too dangerous to be shared and developed without strict controls. But Microsoft, Nvidia, Google and Meta disagreed, arguing that many AI models must be developed largely openly so that people can develop the technology and build new businesses.
Disagreements escalated this month as breakthroughs in Chinese AI models appeared to erode the lead enjoyed by US companies such as Anthropic and OpenAI. The companies have lobbied regulators in Washington, raising concerns about China’s open-source AI models.
On Friday, Jensen Huang, CEO of Nvidia, published a letter supporting open source software development that others in Silicon Valley rallied to. This week, Mr. Huang announced a coalition to work on open source security and cybersecurity tools for AI, joined by other tech giants such as SpaceX and IBM.
Tech executives including Mr. Huang and Mr. Altman are traveling to Washington this week to further make their case to the Trump administration, which is mulling how to deal with the problem.
Amid the battle, Mr. Zuckerberg has repeatedly argued that more people should have access to AI technology, rather than fewer, a quiet criticism of Mr. Amodei and Mr. Altman.
In particular, Mr. Amodei highlighted the dangers of “alignment risk,” a problem with programming artificial intelligence in a way that is safe for humans.
“I think a lot of people have this idea that if you build some kind of singular AI, you can use some idealized form of alignment to make sure it’s benevolent to humanity,” Mr. Zuckerberg said in an interview. “Personally, I’m skeptical of this path.”
Instead, he called for “personalized superintelligence” for everyone, which would mean configuring AI robots to suit people’s needs and wants. A central “benevolent” AI superpower, he argued, would be unsustainable because humans have different opinions and value systems.
“I think it’s literally impossible to have a single benevolent superintelligence that’s connected to everyone at once,” Mr. Zuckerberg said.
Representatives for Anthropic and OpenAI did not immediately respond to requests for comment. In previous statements, OpenAI has said it supports open-source software and is working with the White House on safe ways to support AI development.
In the blog post on Monday Mr Amodei said Anthropic was focused on “keeping powerful chips out of authoritarian hands, stopping industrial distillation and requiring security testing of all sufficiently capable models, both open and closed”.
On Tuesday, top AI researchers from Anthropic, OpenAI, Meta and Google also signed an open letter titled “Pacing the Frontier” arguing for an international effort to develop technical and control tools to slow AI development for safety reasons. OpenAI sent a letter on social networks.
(The Times has sued OpenAI and Microsoft, alleging copyright infringement of news articles. Both companies have denied the claims.)
Meta’s own philosophy around AI development has fluctuated. For years, the company has championed open-source AI, giving away its AI models for free. Last spring, after falling behind Anthropic and OpenAI in the AI race, Mr. Zuckerberg took action.
He invested $14.3 billion in artificial intelligence startup ScaleAI and hired its founder Alexander Wang to revamp Meta’s AI division, which was renamed Meta Superintelligence Labs. Meta spent billions to assemble a team of stellar AI researchers around Mr. Wang, and they began to develop an AI model that was not open, but closed.
The move gave Meta the potential to sell access to new, higher-revenue models. This month, it began selling access to its latest AI model, Muse Spark, for the first time. Meta also continues to develop open-source models.
Mr. Zuckerberg has long believed that artificial intelligence will transform society in positive ways not yet understood. This month Meta released a national advertising campaign he said that investing in AI tools and releasing them in general meant that the company was “betting on people”.
The Wall Street Journal published opinion piece from Mr Zuckerberg this week, in which he criticized the “centralisation” of AI power and said he believed in “individual empowerment, invention and balance of power”. He did not name Anthropic and OpenAI in the article.
In an interview on Tuesday, Mr. Zuckerberg acknowledged that the AI debate has been contentious and noted that conversations with various parties in Silicon Valley and Washington have been productive.
“I don’t know if the goal is for everyone to agree on everything,” he said. “Almost by definition, if you believe in what I’m doing, you don’t have to convince everyone of everything.”
Looking back at the evolution of technology, he said the democratization of technological tools tended to win out in the end.
“I believe the general arc of the industry has been towards more openness and putting the power of technology in the hands of more people, not less,” he said.