AI
AI leaders defend open development amid safety pushback from major labs
Geoffrey Hinton, Fei-Fei Li, and Andrew Ng argued that keeping AI development open is essential to prevent monopolistic control, even as concerns about risks grow.
TechCrunch — AI
Three of the world's most prominent AI researchers — including Nobel laureate Geoffrey Hinton — recently made a forceful case for preserving openness in artificial intelligence development, even as major AI laboratories grow increasingly wary of freely distributed models.
At the Ai4 conference in Las Vegas, Hinton, World Labs CEO Fei-Fei Li, and Coursera co-founder Andrew Ng each articulated why allowing closed systems to dominate the field poses greater dangers than the risks associated with open-source models, though they differed sharply on the details.
The gatekeeper concern#
The researchers shared a fundamental worry: that concentrating AI capabilities in the hands of a few corporations could stifle innovation and entrench competitive advantages, much as Apple and Google's dominance in mobile operating systems has shaped what software can run on those platforms.
Ng emphasized the need for competitive alternatives. He warned that when large firms control access to transformative technology, they can shape both the rules governing the industry and what applications emerge. "If I were to try to give one prescription, it would be to promote openness," Ng said, "because AI is amazing technology and I want it to be in everyone's hands."
Without multiple competing providers, he cautioned, only the largest and best-resourced companies would have the capacity to build the most sophisticated AI systems.
The open-weights debate#
Hinton drew a crucial distinction between open-source software — where underlying code is publicly available for inspection and debugging — and open-weight models, which release the trained parameters of large AI systems for public use. The latter, he argued, makes it trivial for others to repurpose expensive foundation models for harmful purposes, including cyber attacks.
Yet Hinton was pragmatic: he acknowledged that the availability of open-weight models is already irreversible. "I think that battle's been lost," he noted, recognizing that the financial barrier to accessing frontier capabilities has effectively dissolved.
That did not mean dismissing the safety concerns. Hinton maintained that worrying about potential harms from increasingly capable AI systems is justified and should not be dismissed as fearmongering, even if he ultimately believed openness served a greater good.
