Against the backdrop of growing concerns over artificial intelligence (AI) safety, and as initiatives like "Pacing the Frontier" look to major labs as a way to keep research safe, the issue of open-source models is becoming a central point of contention in the industry. Open-weight models, which are distributed freely without tight control over how they are used, are viewed by some labs as a threat that is difficult to monitor. However, as reported by TechCrunch, during the Ai4 conference held last week in Las Vegas, three of the world’s leading AI pioneers—Nobel laureate Geoffrey Hinton, World Labs co-founder and CEO Fei-Fei Li, and Coursera co-founder Andrew Ng—presented strong stances on the matter. Despite disagreements on specific tactics, the three offered compelling arguments in favor of keeping the field of AI open and accessible to the general public.
The Fear of Gatekeepers and Harm to Innovation
For the three speakers, the core concern was the possibility of a handful of major AI companies controlling the pace of technological progress. Andrew Ng highlighted the similarity to the current situation in mobile operating systems, where companies like Apple and Google act as gatekeepers, which can slow down innovation and give the platform-controlling companies decisive influence over what gets built on them. Ng expressed deep concern that a similar dynamic would develop in the AI field. "I don't want there to be gatekeepers," Ng said during the panel, adding that the existence of such gatekeepers limits how all of us can access AI.
The speakers explained that commercial companies have a clear incentive to protect their competitive advantages, in part by influencing the rules and regulations governing the industry. This influence could create a situation where only the largest, best-funded companies have the resources required to build the most advanced AI systems. Ng’s solution to this problem is to maintain a multiplicity of technology providers, allowing different models and companies to compete with one another, rather than permitting a limited number of players to dominate the entire market. "If I were to try to give one prescription, it would be to promote openness," Ng noted, "because AI is amazing technology and I want it to be in everyone's hands."
Geoffrey Hinton's Distinction Between Open Source and Open Weights
However, not all panel members agreed that distributing open-weight models is the right way to preserve competition and accessibility. Geoffrey Hinton drew a clear and important distinction between free, open-source software and open-weight models. According to Hinton, traditional open source is highly positive because it displays the lines of code to the public, allowing many people to inspect it and find bugs. In contrast, open weights mean training a massive, expensive model and then distributing the trained weights to the general public—a process that is fundamentally different in nature.
Hinton explained that he had previously opposed open weights because this approach makes it significantly easier for various actors to take large foundation models, which are very expensive to train, and retrain them at a much lower cost to carry out harmful activities, such as cyberattacks. Nevertheless, despite his reservations, Hinton acknowledged during the conference that open-weight models have already become a permanent and integral part of the AI landscape. "I think that battle's been lost," Hinton said. "We now have open-weight models, so the barrier to lots of people getting these big models—which was the cost of training foundation models—that barrier has disappeared. It's too late."
Addressing Risks and Benefits of Advancing Technology
Despite accepting the current reality, Hinton emphasized that this does not mean the risks associated with the technology should be ignored. Hinton’s position remained clear: AI will continue to advance, and he believes this is largely a good thing. He noted that the technology will boost productivity and significantly improve education and healthcare. At the same time, he defended the concerns regarding potential negative impacts: "Worrying about the possible bad effects of AI and the things that intelligent beings might do when they're smarter than us—I don't think that's unfair. I think it is unfair to label anybody who thinks like that as a fear-monger," Hinton added.
Andrew Ng Warns of Losing the Geopolitical Competition to China
Andrew Ng presented a slightly different perspective on the matter. In his view, the key question is not just whether open models carry risk, but who controls access to them and who will win the global market. Ng explained that whoever develops the cheaper model will enjoy a clear advantage. He warned that if China’s open-weight models gain widespread adoption across Asia, Africa, and/or developing nations, they could influence how billions of people are exposed to ideas regarding democracy, freedom, and human rights.
"One thing I hope we do is encourage American competitiveness and open-source AI," Ng said. He explained that AI constitutes a massive source of soft power, as demonstrated by the great success and achievements of Chinese models in Africa, for example. Ng’s major concern is that, due to heavy lobbying in the United States and fear-mongering, open-source AI development in America is struggling to compete with open-weight models originating from China. Ng expressed concern that if China finds a fundamentally more cost-effective way to build AI, more economical technologies will have a fundamental and decisive business adoption advantage in the global market.
Fei-Fei Li Proposes a Multi-Layered Model and Comparison to Nuclear Physics
Fei-Fei Li pushed back against the dichotomous framing of the issue. According to her, it is a mistake to present the issue as a binary choice between complete openness and complete closedness. "In complex software systems, as well as scientific systems, the matter is much more complex and nuanced," Li explained. To illustrate her point, Li presented the field of nuclear physics as an example: scientific papers in this field are published completely openly to the public, whereas uranium itself is tightly regulated, while work in research laboratories is situated somewhere in between these two extremes.
The main lesson from this example, according to Li, is that openness does not have to be an all-or-nothing decision. Different layers of the ecosystem can operate at entirely different levels of openness. Li also emphasized the importance of collaborations between public and private institutions, citing the Human Genome Project as an example. The knowledge generated by the project became a platform on which others could build, allowing pharmaceutical companies to profit, scientists to advance their research, and society as a whole to benefit from the results. "So I think we have to use AI as that kind of infrastructure," Li said, adding that certain levels of openness are required in scientific discovery, education, and global partnerships, alongside profitable business models for entrepreneurs, while also accepting closed-source systems. In her view, the sweeping debate assuming that only one model can be tolerated is a false debate, and we need to reach a more complex and detailed level of analysis.
Broad Agreement on the Necessity of Government Regulation
Despite differences in tactics, there was full agreement among the three speakers that some regulation will be necessary to keep AI on the right track. "What we want to do is develop AI in a direction that helps people, and regulation will help us do that," Hinton concluded. He strongly emphasized his position on who should determine the rules: "You can't leave it to people like Elon Musk and Mark Zuckerberg to decide how AI should be done."