Stay in view of the free updates

Simply subscribe to artificial intelligence Myft Digest – it is delivered directly to your inbox.

When the Chernobyl Nuclear Energy Station exploded in 1986, it was a disaster for those who lived near in northern Ukraine. But the accident was also a global industry disaster that pushes nuclear energy as a technology in the future. The pure number of Nuclear reactors It was largely flat because as he was seen as unsafe. What will happen today if the artificial intelligence industry suffers from an equivalent accident?

This question was asked on the sidelines of the AI ​​Action summit for this week in Paris by Stuart Russell, Professor of Computer Science at the University of California, Berkeley. His answer was that it was a mistake to believe that there should be a comparison between safety and innovation. So those most enthusiastic about the promise of artificial intelligence technology should go carefully. “You cannot innovate without safety,” he said.

Russell’s warning was repeated by some other artificial intelligence experts in Paris. We must agree on the minimum safety standards worldwide. “We need it before we have a great disaster,” Windy Hall, director of the Southampton University Science Institute, told me.

But such warnings were mostly on the margins, as government delegates at the summit wandered around the cavernous large palace. In a pierced speech, JD Vance emphasized the necessity of national security security in artificial intelligence. Vice President of America argued that technology would make us “more productive, more prosperous, and more free.” He said: “The future of artificial intelligence will not be won by manual mobility.”

While the first international summit of artificial intelligence focused in Balcheli Park in Britain in 2023, it focused almost completely – most of which are excessive – on safety issues, the priority in Paris was to work as president of Emmanuel Macron at the time Big investments In the French technology industry. Max Tegark, head of the Future of Life Institute, who participated in hosting a marginal event on safety: “The process that started in Bletchley, which I thought was really amazing, was specific here.”

What is most related to safety activists is the speed with which technology and corporate dynamics – and geopolitical – to achieve artificial general intelligence, when computers correspond to humans in all cognitive tasks. Several artificial intelligence research companies, including Openai, Google DeepMind, Noteropic and Deepseek in China, have a clear task to reach AGI.

Later this week, Dario Ameudi, co -founder and CEO of anthropologist, expected that AGI would likely be achieved in 2026 or 2027. ”

Next to him, Dimis Hasabis, co -founder and CEO of Google DeepMind, was more careful, expecting a 50 percent probability to achieve AGI within five years. “I will not be shocked if it is shorter. He said,” I will be shocked if it is longer than 10 years. “

Critics of safety activists depict them as scientific imaginative people who believe that creating an artificial ingenuity will lead to human extinction: they stand from hands like Luddites on the last day in the way of progress. But safety experts are concerned about the damage that the very strong artificial intelligence systems can cause today and the risk of cyber weapons attacks or huge biological weapons. Even the main researchers admit that they do not fully understand how their models work, which creates concerns related to security and privacy.

A Search paper In the asleep agents of the anthropoor last year, some basic models found human deception to believe that they were working safely. For example, models that were trained to write a safe code in 2023 can include an exploitation code when the year was changed to 2024. This back behavior has not been revealed through standard safety techniques in humans. The possibility of a candidate from the algorithm -lukewarm Mancheshis in China has already been banned by many countries.

Nevertheless, Tegmark is optimistic, that both artificial intelligence companies and governments will witness the overwhelming self -interest in re -determining safety priority. The United States, China, or anyone else does not want artificial intelligence systems from control. “The safety of artificial intelligence is the global public interest,” Xue Lan, Dean of the Amnesty International Governance Institute at the University of Tsinghwa in Beijing, told the safety event.

In the race to take advantage of the full potential of Amnesty International, the best slogan for the industry may be the US Navy seals, and it has not been noted by many hands. “Small slow and smooth fast.

john.thornhill@ft.com