Sam Altman of OpenAI suggests an international organization oversee the ‘most powerful’ AI for’ reasonable safety’

Sam Altman of OpenAI says that the "most powerful" AI should be watched over by an international body to make sure "reasonable safety."

7,498

“I think there will be a time shortly like we’re not talking many years from now, where boundary artificial intelligence systems are able of causing major global damage,” Altman said on the show on Friday.

Those systems will “harm more than one country,” he says, and he wants them to be controlled by “an international agency looking at the most powerful systems and making sure reasonable safety testing.”

To Altman, figuring out the right amount of control will be like balancing two scales.

“I’d be very worried about too much government interference here.” I think we do way too much or a little too much of this wrong. We might mess this up if we don’t do enough, he said. 

Laws are already being made to control the quickly changing technology.

In March, the EU passed the Artificial Intelligence Act, which will classify AI risks and make it illegal to use AI in certain situations. President Joe Biden also signed an order last year asking the biggest AI models in the world to be more open about their work. Bloomberg says that this year, California has been the leader in controlling AI, with politicians there looking at more than 30 bills. 

However, Altman said that an international body would be more adaptable than national laws, which is important because AI changes so quickly.

“The reason I’ve pressed for an agency-based strategy because I kind of like the big-picture things and not like a write-it-in-law is in a year it will all be written wrong,” said he. He thinks that politicians, even if they are “true world experts,” probably won’t be able to write laws that will control events in 12 to 24 months. 

To put it simply, Altman believes that AI should be controlled in the same way that an airplane is.

“When like a major loss of human life is a serious potential, like airplanes, or a variety of other instances where I think we’re happy with some sort of testing framework,” he told me. “When I get on an aircraft, I don’t think about it.” “I think it will be secure in the end.”

Comments are closed.