OpenAI CEO Sam Altman says in Abu Dhabi that AI is now “mostly smarter” than the best scientists and that humans must retain control of the future

OpenAI CEO Sam Altman says in Abu Dhabi that AI is now “mostly smarter” than the best scientists and that humans must retain control of the future

On November 2, Altman spoke at the opening of the ADIPEC oil and gas exhibition and conference. That same day, state-owned oil company ADNOC announced a partnership with OpenAI and technology group G42 under which ADNOC will become the first energy company to deploy OpenAI Frontier, a platform for AI agents that carry out assignments autonomously.

Speaking with the editor-in-chief of The National, Altman addressed people's anxiety about AI: “At the beginning of this year, and this is very approximate because capabilities are uneven and this does not hold across the board, AI was mostly not as smart as our smartest scientists. Now it is mostly smarter than they are.” In the same answer, he said people would remember the importance of human judgment and of having a person accountable for it.

In his opening address, he said AI had begun solving mathematical problems in recent months that humans had been unable to solve. As an example, Altman cited the occasion “when we proved the first Millennium Prize Problem.” The Millennium Prize Problems are seven difficult problems selected in 2000 by the Clay Mathematics Institute, which offered a million dollars for each solution. According to Altman, 10 thousand agents, each a copy of a model due for release soon, worked for about four days. “There was no hierarchy. There were no bosses. Everything was completely decentralized”: the agents chose their own directions, shared results, and learned from one another.

Altman did not identify the problem, but the figures match the account of a proof concerning the Navier-Stokes equations that OpenAI published on September 8: 10 thousand agents and about 88 hours of work. The equations describe the flow of water and air, and the problem asks whether flow velocity can become infinite in a finite time, a phenomenon mathematicians call “blow-up.” OpenAI reported that the agents proved such a blow-up for a flow subject to a specifically chosen external force. Altman gave the cost on the Cheeky Pint podcast: around 10 million dollars.

Clay's official formulation allows a flow subject to an external force, and on September 11 the institute described the problem as “apparently solved.” Most research groups had studied flow without an external force, which University of Chicago mathematician Luis Silvestre considers the central problem: “The Clay problem is solved, while the main problem concerning the Navier-Stokes equations remains unsolved,” he told Scientific American.

Asked about OpenAI's responsibility, Altman said:

“We believe it is absolutely critical that humans remain in control of the future. There are many other things we can debate, but we believe there should be no room for debate on this. There should be no handover of control to AI. There should also be no gradual shift toward ‘we will let AI decide what is best for us.’”

AI, he continued, should be “a tool built for people and serving people, not the other way around”; this is a value “we intend to really fight for.” He added: “I wish this did not need to be said.”

Altman said the agents worked without bosses; humans set their task. According to OpenAI and Scientific American, researchers focused the entire effort on the Navier-Stokes equations, launched 10 thousand agents, and gave them a starting proof for a fluid without friction that about 100 other agents had found. ADNOC's press release states that the first application will be preparing field development plans, with consequential decisions remaining under human expert oversight.