At the Ai4 conference in Las Vegas last week, three prominent figures in artificial intelligence — Geoffrey Hinton (Nobel Prize winner), Fei-Fei Li (World Labs CEO and co-founder), and Andrew Ng (Coursera co-founder) — debated how to balance openness and safety in AI development. While they disagreed on specific tactics, each made a case for keeping AI accessible.
Concern about concentration of power and access
All three speakers expressed worry that a small number of large companies could control the pace of AI progress and who gets access. Andrew Ng compared the risk to the dominance Apple and Google have over mobile operating systems, which can slow innovation and shape what is built on those platforms. “I don’t want there to be gatekeepers,” Ng said, arguing that gatekeepers limit how everyone can access AI.
Ng recommended preserving multiple providers and competition among models and companies rather than permitting a few players to dominate. “If I were to try to give one prescription, it would be to promote openness,” he said, because he wants AI to be widely available. He also warned of geopolitical implications: if cheaper open-weight models from China or elsewhere gain wide adoption in Asia, Africa, or the developing world, they could shape how billions of people encounter ideas about democracy, freedom, and human rights.
Hinton’s distinction: open-source code vs. open weights
Geoffrey Hinton drew a clear line between open-source software, which publishes code for inspection and modification, and open-weight models, which release the trained parameters of large models. “Open source is great. ... Open weights means you train a big model and then you give people the weights. That’s very different,” Hinton said. He said he opposed open weights because they make it easier for others to take expensive foundation models and, with much less money, adapt them to malicious ends such as cyberattacks.
Despite his reservations, Hinton acknowledged that open-weight models have already become a permanent part of the landscape. “I think that battle’s been lost. We now have open-weight models, so the barrier to lots of people getting these big models ... that barrier has disappeared. It’s too late.” He nonetheless emphasized that AI development can be largely beneficial — increasing productivity and improving education and healthcare — while warning about potential harms.
Fei-Fei Li: nuanced, layered openness
Fei-Fei Li pushed back against treating openness as an all-or-nothing choice. She argued for a more nuanced approach in complex scientific and software systems, where different layers of the ecosystem can operate at different levels of openness.
Li used nuclear physics as an analogy: scientific papers are published openly, uranium is regulated, and laboratory work sits in between. She also pointed to public–private collaborations like the Human Genome Project, where open knowledge created a platform that allowed pharmaceutical companies to profit, scientists to advance, and society to benefit. “So I think we have to use [AI] as that kind of infrastructure,” Li said, calling for levels of openness across scientific discovery, education, global partnerships and viable business models.
Agreement on the need for regulation
All three agreed that some regulation will be necessary to steer AI development toward beneficial outcomes. Geoffrey Hinton summarized the point: “What we want to do is develop AI in a direction that helps people, and regulation will help us do that.” He added that it should not be left to individual tech magnates such as Elon Musk or Mark Zuckerberg to decide how AI should be managed.
Conclusion
The Las Vegas discussion showed a shared preference for openness and competition as bulwarks against concentration of power, but differing views on how openness should be implemented. Andrew Ng emphasized competition and broad access to prevent gatekeepers; Geoffrey Hinton warned specifically about the risks of publishing model weights while recognizing that open-weight models are already widespread; and Fei-Fei Li urged a layered, nuanced approach that mixes openness with appropriate regulation. The debate underlines that both access and targeted regulatory frameworks will be needed to ensure AI development serves the public good.
