Berliner Tageblatt - Meta chief pushes back on AI slowdown calls

"People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned," Zuckerberg wrote on X.

Alignment refers to the effort of making an AI system's goals and behavior match what humans actually want it to do.

The debate was sparked in July when ChatGPT-maker OpenAI revealed that during testing, hundreds of AI agents -- AI programs that can take actions on their own -- broke out of their test environment, connected to the internet and infiltrated a website.

For the founder of Facebook, establishing trust and creating reliable systems are "quickly becoming the most important capabilities" separating good AI products from bad ones.

"Any lab that doesn't focus on alignment will fall behind," he said.

The debate has expanded to whether AI companies can be trusted to regulate themselves, with rivals including OpenAI, Anthropic and Google exploring a joint standards body.

Prominent industry voices led by Anthropic CEO Dario Amodei are calling for development to slow down.

Zuckerberg argued that liability exposure ultimately forces labs to act responsibly, pointing to Meta's own decision to delay the rollout of its Muse AI system -- which features personalized agents -- for several months to shore up safety and security.

Meta itself has faced a wave of liability lawsuits in recent months unrelated to AI, agreeing in August to pay billions of dollars to settle claims that it designed Facebook and Instagram to be addictive to children.

"We didn't call for everyone else to do this before we would," he wrote of the Muse AI release.

"We just did it as part of our day-to-day work because it was clearly the right thing for people and for us."

He also endorsed the use of independent evaluators to scrutinize AI systems, saying Meta Superintelligence Labs, his company's AI division, already relies on outside reviewers.

Zuckerberg called for "a larger and more diverse ecosystem of evaluators" across the industry, in a tacit criticism of Anthropic, which is accused of wanting to use evaluators the company has close links to.

The post from Zuckerberg was endorsed by Silicon Valley figures close to the White House who are pushing back on the wave of warnings about AI's potential dangers to humanity and the calls for a slowdown in development.

L.Janezki--BTB