Estonian scientists call for slower development of artificial intelligence

Estonian scientists say the development of artificial intelligence should be slowed until stronger safeguards are in place, ERR reported. Rapid advances in AI are raising questions about whether people can control increasingly capable systems.

Several recent incidents involved AI agents acting autonomously while carrying out tasks, including breaking into external systems or attempting to exploit their security vulnerabilities. In response, calls for tighter restrictions have come not only from critics but also from some companies developing AI.

“We are clearly at a point where there is great potential, but also a real risk,” OpenAI co-founder and CEO Sam Altman said.

In Washington this week, both Senator Bernie Sanders and Steve Bannon, a former ally of Donald Trump, called for stronger oversight of AI.

“I think Senator Sanders gave a good overview of the risks. Obviously, we have very different views of the US president, but we are trying to rise above party politics, and we have to do that,” Bannon said.

Estonian scientists attending an AI conference in Washington said there is currently no certainty that AI systems much more capable than humans could be kept under control.

“If we continue building increasingly intelligent AI systems without limits, then based on the current situation, we do not know how to control an AI that is much more intelligent than humans. Until we have better control measures, we should not build such systems either,” said Meelis Kull, a professor of artificial intelligence at the University of Tartu.

“In every respect, slowing the development of these large supersystems would be somewhat beneficial, because it reduces the major risk and gives us time to do something, while also allowing the economy and society to adapt,” said Tanel Tammet, a professor of applied artificial intelligence at Tallinn University of Technology.

Superintelligence refers to a still-hypothetical form of AI that would be more intelligent than humans in almost all fields. Jacob Coxon, who helped train large AI models but recently left the field, has also warned that superintelligent AI could soon be ready if development continues in the same way. He said films such as “The Terminator” could become reality.

“For comparison, if someone rides a motorcycle throughout their life, the risk of dying in an accident is about 1 percent. Many people believe that the probability of AI causing the extinction of all humanity, not just one person but everyone, is at least five times higher, perhaps even more. Some believe it is above 10 percent. That is genuinely alarming, and it is not just a view held by a few people in AI companies; academics conducting research also share it,” said Risto Uuk, a member of the Eesti.ai council.

Uuk said limiting the development of artificial intelligence is essential for reducing these fears and risks. The European Union has already established AI rules, but a global agreement is needed, he said.

“China already has its own regulations, South Korea has adopted some measures from the European Union, and similar rules are emerging in Brazil and elsewhere. The next step would be to discuss collectively and globally what kind of agreement we could reach,” Uuk said.

Uuk also said Estonia can influence the development of AI more than is often thought, both through the European Union and at the individual level.