AI leaders argue for open access with layered safeguards
At the Ai4 conference in Las Vegas, Geoffrey Hinton, Fei-Fei Li and Andrew Ng made the case for preserving meaningful openness in artificial intelligence while accepting that regulation is needed. The three researchers differed on open-weight models, but each warned against treating the issue as a simple choice between unrestricted release and fully closed systems.
Ng, Coursera’s co-founder, said he did not want AI access controlled by gatekeepers. He compared the concern to mobile operating systems, where platform owners can influence the applications built on top of their technology. In his view, maintaining multiple providers and competing models is important because companies have incentives to defend competitive advantages and influence the rules governing the sector.
Open source and open weights are not the same
Hinton drew a firm distinction between open-source software and open-weight AI. Open-source code can be inspected and modified, enabling users to identify errors. Open weights, by contrast, provide the trained parameters of a large model. Hinton said that release can make it cheaper for others to adapt expensive foundation models for harmful uses, including cyberattacks.
He nevertheless said open-weight models are now an established part of the AI landscape. The training cost that once formed a barrier to access has already ceased to be decisive for many users. Hinton also argued that concern about harmful effects from systems more intelligent than people should not be dismissed as fear-mongering, even as he expects AI to improve productivity, education and healthcare.
Competition, governance and different access layers
Ng framed openness partly as a competition issue. He warned that the most cost-efficient models may gain a business adoption advantage, and said AI can become a source of soft power when models shape how large populations encounter ideas about democracy, freedom and human rights. The policy debate over open-weight AI restrictions and industry pressure reflects the pressure to avoid broad restrictions while keeping the ecosystem competitive and accountable.
Li, the co-founder and CEO of World Labs, rejected an absolute divide between open and closed approaches. She compared AI to nuclear physics, where published scientific knowledge is open, uranium is regulated and laboratory work occupies an intermediate space. Her example was intended to show that different layers of complex scientific and software systems can operate under different levels of access.
Li also pointed to public-private collaboration around the Human Genome Project, where shared knowledge became a base for scientific work, commercial activity and broader social benefits. She said AI needs openness in scientific discovery, education and global partnerships, alongside viable business models and accepted closed-source systems.
What enterprises should take from the debate
The speakers agreed that regulation has a role in directing AI toward outcomes that help people. For businesses, the implication is to evaluate AI deployments beyond a binary open-or-closed label: assess model access, the risk of downstream modification, security controls, vendor concentration and the governance requirements attached to each use case.