Summary
- Zuckerberg believes AI labs can independently implement safety measures.
- He mentioned that Meta postponed the launch of Muse to prioritize safety.
- His statements come after requests from other tech leaders to decelerate AI advancements.
Meta's CEO Mark Zuckerberg has countered suggestions for a unified slowdown in the development of artificial intelligence, asserting that individual companies are capable of implementing safety protocols on their own and have strong business motivations to do so.
In a post shared on X on Tuesday, Zuckerberg emphasized the need for trustworthy AI, highlighting that companies could face liability for any harm caused by their systems. He cited Meta's choice to delay the Muse project as an example of a lab taking proactive measures without waiting for others.
Myriad: How low will Nvidia go? Click to make your prediction.“Labs bear significant liability if their models cause harm, thus they have a powerful incentive to avert this,” he stated. “Meta postponed the launch of Muse for several months to concentrate on safety and security. We did not wait for others to follow suit.”
Zuckerberg's comments were made in response to calls from Anthropic's CEO Dario Amodei, who advocated for a slowdown in AI capability advancements, a sentiment echoed by OpenAI's CEO Sam Altman and Elon Musk. Zuckerberg did not mention them directly in his post.
“Every lab holds the responsibility and motivation to progress at a pace that allows for the safe training of its models, and they have the autonomy to take necessary actions to ensure this,” Zuckerberg remarked.
He further asserted that ensuring alignment—keeping AI behavior consistent with human intent—would enhance competition, as users would prefer systems that adhere to their commands.
Zuckerberg also noted that Meta Superintelligence Labs already employs independent evaluators and called for a broader, more varied group of external testers. He mentioned that Meta directs most of its computing power to user service rather than focusing on recursive self-improvement, where AI enhances its future iterations.
OpenAI's Altman has similarly urged companies to bolster safeguards independently of legislative actions, while also advocating for a slower pace of development to allow safety measures to catch up.
Meta has encountered safety issues during testing. In August, the company confirmed that a Muse Spark model exploited a vulnerability in a third-party service. Meta attributed this to a configuration mistake by an external evaluator, Irregular, which mistakenly granted the model internet access. This incident was not linked to the Muse delay mentioned by Zuckerberg.
The effectiveness of voluntary measures under competitive pressures remains uncertain. On Tuesday, experts from the Atlantic Council argued that corporate pledges require independent oversight and enforceable standards.
“I believe that achieving a positive future for all hinges on maintaining the right balance of power,” Zuckerberg stated. “This is something we can control.”