Hinton, Ng, Li defend open AI despite safety concerns

The debate at Ai4 reflects a fundamental tension in AI governance: how to balance innovation and safety without ceding control to a small group of corporations. Ng's concern about gatekeepers echoes real anxieties about mobile computing, where Apple and Google's dominance over operating systems has shaped which apps reach users and how the ecosystem evolves. By that logic, allowing a handful of AI labs to monopolize access would concentrate not just computing power but also ideological influence—a concern made more acute when Ng frames it as a geopolitical competition between U.S. and Chinese models. Yet Hinton's candid admission that "the battle's been lost" on open weights reflects a different reality: the cost barrier to training foundation models has already eroded, and suppressing open-weight models is no longer feasible. Li's middle path—accepting that different layers require different controls—sidesteps the harder question of enforcement: how do you keep uranium regulated while scientific papers flow freely when the underlying physics is published? The three speakers' agreement on one point, however, is telling: all three see some form of regulation as necessary, though they pointedly did not specify what that should look like or who should enforce it.