'There's a one in three chance AI will kill mankind in ten years', ex-Google DeepMind researcher warns

There is a one in three chance that artificial intelligence (AI) will kill all humans within the next ten years, a former Google DeepMind researcher has warned.

Alex Turner said AI systems are getting smarter and more powerful, but scientists are finding it harder to understand how they work. He warned that humanity is reaching an important point where the choices people make could decide its future.

Speaking at the NDTV World Summit on Friday, Turner warned that AI could kill humanity even if nobody tells it to do so.

But he said the outcome depends on what humans do next. He explained that he wanted to warn people about what could happen if companies continue building AI in the same way they do today.

Turner said private companies are competing to create more powerful AI systems. He warned that this competition could move faster than scientists' ability to make sure the systems stay under human control.

He explained that AI is developing so quickly that even the people who create it do not always understand how it works or why it behaves in certain ways. This means humans might not be able to fully control what AI does.

Turner compared AI with other kinds of engineering. Engineers can usually understand how the buildings or machines they design will work. But companies creating AI may not fully understand their own technology, he said.

AI systems can now carry out more and more difficult tasks. However, they might not follow instructions in the way the people using them expect.

Former Google DeepMind researcher Alex Turner (pictured) has warned that there is a one in three chance AI will kill humanity within the next ten years.

This makes AI's behavior difficult to understand or predict. In other areas of engineering, experts can test how something works using well-known scientific rules. Turner said AI systems are not built in the same way.

"We haven't built AI like that," Turner said. He also warned that AI might act differently when it knows researchers are watching it. This could make it even harder for scientists to understand what the technology is doing.

His warning could worry people in some industries because AI is being developed with less human supervision than other technologies. This means AI systems could do things people do not want, especially when they are given more difficult tasks.

Turner left Google in July after the company signed an agreement with the Pentagon to use Google's AI technology for secret government work. He warned that AI could eventually help make itself more advanced, making it harder for humans to keep up with its progress.

He called this an "intelligence explosion," meaning AI could help improve itself faster and faster. Turner warned that this process is already happening today.

The former Google researcher said a scientist working at another AI laboratory told him that AI is already being used to "monitor" other AI systems. Turner said this suggests AI is already starting to check and improve its own work without direct human involvement.

Turner explained that the scientist told him people were no longer checking work produced by AI themselves.

Instead, they were building another AI system to check the first AI system's work.

Turner also warned that some large AI companies are denying the dangers of their technology. He compared this with how tobacco companies once denied the link between smoking and lung cancer, and how oil companies denied the dangers of climate change.

He said: "Big tobacco lied about lung cancer. Big oil lied about climate change. And now part of big AI says these risks are fake."

His warning comes after Bill Gates said last month that AI could cause the deaths of one billion people unless humans step in to control how the technology develops.

Gates said AI is already powerful enough to cause events that could kill a billion people.

He explained that although it would be difficult for AI to wipe out 100 percent of humanity, there has never been a weapon as powerful as the combination of dangerous people and the latest AI tools.

Gates also suggested possible ways to reduce these dangers when Welker asked him what could stop tragedies caused by AI.

He warned that AI could make it very difficult to tell the difference between scientists using the technology to create medicines that save lives and dangerous people using it to develop biological weapons.

Gates told Welker that he wished it were easier to tell whether someone wanted to use AI for something good, such as designing a new medicine to treat a disease, or something harmful, such as creating a tool for biological terrorism.

The Microsoft co-founder also said these risks mean powerful AI systems need strict rules and supervision. He called on the United States to lead efforts to create safety measures.

Gates said that because helpful and harmful uses of AI can be so similar, experts need to watch every advanced AI model and keep an exact record of what it is being used to do.

Gates made these comments after Donald Trump said the United States was ahead of China in AI development and that predictions about AI killing people would not come true.

Trump said the United States was leading China in AI and was the most technologically advanced country in the world. He said he wanted America to stay ahead because he believed whichever country wins the AI race will come out on top.

Earlier in September, Jacob Coxon, a whistleblower from the AI company Anthropic, also warned about what he called "dangerous technology." He said he believes AI could bring an end to humanity.

Coxon said humans have mostly worked out how to control nuclear weapons, but they still do not know how to control AI.