Three leading AI researchers oppose market closure, disagree on almost everything

The Ai4 conference in Las Vegas became the stage for an internal debate among three of the field’s most prominent researchers, centered on the future openness of artificial-intelligence models. Geoffrey Hinton, Nobel laureate; Fei-Fei Li, chief executive officer and co-founder of World Labs; and Andrew Ng, co-founder of Coursera, agreed on a single principle: it is undesirable for a handful of companies to dictate the pace of technological progress. Their argument draws a parallel to the mobile-operating-system market, where Apple and Google created a reality in which a dominant player can shape what is built on the platform and delay innovation. Ng voiced opposition to creating such gatekeepers in AI, warning that a centrally controlled model would restrict everyone’s ability to access the technology.

The economic logic behind the concern is straightforward: firms have an incentive to protect their competitive advantage, and one way to do that is to influence the rules that govern the industry. The possible outcome is a market in which only the largest, deepest-pocketed companies—those with the resources to build the most advanced systems—survive. Ng’s proposed remedy is to preserve a number of competing suppliers rather than allowing a few players to dominate, and to promote openness so that the technology reaches as many hands as possible.

Hinton, by contrast, refused to equate open-source with open-weights, two concepts that are often conflated in industry discourse. Open-source allows inspection of code and the identification of bugs, whereas open-weights means training a massive model and then releasing only its parameters. He opposed the latter because it lowers the entry barrier: anyone who receives a foundation model that required massive capital to train can fine-tune it for malicious purposes such as cyber-attacks. Nevertheless, Hinton acknowledged that, in practice, the field has already moved past this barrier; the cost of training foundation models has disappeared, and open-weight models are now a routine reality.

Recognizing the broader reality did not cause Hinton to abandon his concerns. He stressed that AI will continue to advance, which he views as largely positive for productivity, education and health, but added that the fear of intelligent systems acting against humanity cannot be easily dismissed, and that labeling every deployer as a fear-monger is inappropriate. Ng shifted the discussion toward geopolitics, arguing that the central question is not only whether open models are risky, but who controls access and the market. He warned that open-weight models from China spreading across Asia, Africa and the developing world could shape how billions encounter ideas about democracy and human rights, and called for strengthening American competition and open-source AI, recognizing the technology as a significant source of soft power.