Estonian scientists say AI development should be slowed down

Estonian researchers at an AI conference in Washington called for slowing the development of superintelligent AI, warning there is currently no certainty such systems could be kept under control. University of Tartu Professor of Artificial Intelligence Meelis Kull said "until we have better methods of control, we shouldn't build such systems," while Tallinn University of Technology Professor of Applied Artificial Intelligence Tanel Tammet said slowing large super-systems would reduce major risk and give the economy and society time to adapt. Eesti.ai advisory council member Risto Uuk cited public risk estimates that AI could cause human extinction at 5 percent or higher — at least five times a motorcycle rider's roughly 1 percent lifetime risk — and said a global agreement is needed beyond existing EU, Chinese, Korean and Brazilian rules.

Estonian scientists say AI development should be slowed down Recent incidents involving AI systems acting autonomously have intensified calls for stronger oversight, with researchers warning that safeguards must keep pace as the technology becomes increasingly capable. Recently, there have been several cases in which AI agents have acted on their own while carrying out tasks, breaking into other systems or attempting to exploit security vulnerabilities. Developments following these incidents have moved quickly. Calls for restrictions on AI are no longer coming only from critics, but also from some of the companies developing the technology. "We are clearly at a point today where there is enormous potential, but also real risk," OpenAI co-founder and CEO Sam Altman said. Few issues are capable of bringing political opponents together around a common goal in the United States. Yet this week in Washington, both Sen. Bernie Sanders and former Donald Trump ally Steve Bannon called for stronger oversight of AI. "I think Sen. Sanders gave a good overview of the risks. Obviously, we have quite different views of the U.S. president, but we're trying to rise above party politics and we have to do that," Bannon said. Estonian researchers attending an AI conference in Washington said there is currently no certainty that artificial intelligence far more capable than humans could be kept under control. "If we keep building increasingly intelligent AI systems without limits, then, based on where things stand today, we don't know how to control an AI that is vastly more intelligent than humans. Until we have better methods of control, we shouldn't build such systems," University of Tartu Professor of Artificial Intelligence Meelis Kull said https://news.err.ee/1610142064/ai-professor-today-s-chatbots-will-not-end-civilization . "Slowing the development of these large super-systems would be somewhat beneficial in every respect because it would reduce the major risk and give us time to do something about it, while also giving the economy and society time to adapt," Tallinn University of Technology Professor of Applied Artificial Intelligence Tanel Tammet said. Superintelligence is the term used for a so far hypothetical form of AI that would be more intelligent than humans in almost every field. Jacob Coxon, who helped train large AI models but recently quit, has also warned of the danger, saying that if development continues on its current path, artificial superintelligence will soon be a reality and science fiction films such as "The Terminator" could become real. And people are genuinely afraid. "For comparison, if someone rides a motorcycle throughout their life, their risk of dying from it is about 1 percent. In other words, many people think the probability that artificial intelligence could cause all of humanity to go extinct — not one person, but everyone — is at least five times higher, perhaps even more. Some think the probability is above 10 percent. That's actually very concerning and these aren't just a few isolated individuals at AI companies saying this; they include academics conducting research," Eesti.ai advisory council member Risto Uuk said. Setting limits on the development of artificial intelligence is therefore essential to reducing these fears and risks. The European Union has already introduced rules for AI, but a global agreement is needed. "China already has its own regulations, Korea, for example, has adopted some things from the European Union, Brazil — regulations are starting to emerge everywhere. The next step would be for us to come together globally and discuss what kind of agreement we could reach," Uuk said. Estonia can influence the development of AI more than is often assumed, both through the European Union and at the individual level. "Europe actually has an enormous amount of power and influence over these technology companies because the European market is very large. The European Union can theoretically say that if you don't follow proper safety practices, if you don't ensure security and safety, you don't get access to our market. The same applies to Estonia because Estonia is an EU member state," Uuk said. "I would like Estonians to lose this sense of helplessness. There is a great deal we can have a say in and at the individual level, we can vote and elect politicians who are concerned about artificial intelligence. We can learn about artificial intelligence ourselves, use it and see how it can be beneficial and how it can be harmful. There are all sorts of things we can do, so there is no reason for us to adopt a position of hopelessness," Uuk added. This is all the more relevant because, paradoxically, AI itself can help mitigate the risks associated with AI. "If we were to say that we shouldn't use AI, we'd be putting ourselves in an especially difficult position. We should actually try to make the greatest possible use of AI in our activities and economies and get involved with it because that will at least keep us near the leading edge of this wave," Tammet said. -- Editor: Marcus Turovski. Johanna Alvin Source: "AK. Nädal"