Hinton Says We Might Have Been Better Off Without AI, Calls for Independent Review Before Release

Geoffrey Hinton has proposed that powerful AI models be reviewed by an independent body before they are released to the public. He made the comments in a November 3 interview with the Guardian. Hinton said AI could still deliver a future that benefits humanity, but added that, over the long run, it might have been better never to have developed it at all.

Since leaving Google in 2023, Hinton has spoken publicly about the risks of AI. In this interview, he again raised the question of what AI’s advance will ultimately mean for humanity, and of who should judge those risks and on what evidence.

What would be reviewed before a powerful AI model is released

What Hinton wants examined is the statistical evidence on the benefits and risks of powerful AI models. Instead of releasing a model first and watching what happens, he argues, companies should be required to submit statistical evidence to an independent body, before release, showing that the benefits outweigh the harms. In other words, put one checkpoint in place before ordinary users ever get their hands on the system.

Hinton compared the process to the way drugs are vetted in the United States before going on sale. Just as a new medicine is weighed for both its effects and its side effects before it can be marketed, powerful AI should have its benefits and risks assessed together when deciding whether to release it.

Another notable feature of the interview is that Hinton spoke about both the upside and the danger. He said a future in which AI benefits humanity is possible, while voicing concern about the long-term outcome. His remark that it might have been better not to develop AI at all is likewise a judgment about what lies ahead.

Why the power grid and the financial system

Hinton predicted that a major AI-related incident could hit the electricity grid or the financial system in 2026 or 2027. He pointed to areas such as power and money, where a major incident could immediately disrupt daily life.

He also leveled criticism at some executives at major AI companies. Hinton said there is a problem with the very incentives driving them, and argued that an independent body should review the evidence before powerful AI is released.

The interview raises a single question: what review is needed before people use AI, and who should conduct it. Hinton proposes placing that checkpoint before a model is released. Whether the proposal feeds into an actual policy debate — and, if it does, which models would be covered and what evidence would be examined — is what to watch next.