Bill Gates says artificial intelligence has not yet slipped beyond human control, but he believes society needs to take its risks much more seriously as AI systems become increasingly capable. Speaking to NBC News, the Microsoft co-founder said the immediate danger is not AI acting on its own. Instead, he pointed to people with bad intentions using the technology to carry out attacks, fraud and other harmful activities.

Gates said AI is different from earlier technologies because its capabilities are improving rapidly. He argued that governments, companies, researchers and policymakers need to think about how the technology should be developed and what safeguards should be put in place.

The risks are already changing

Gates said AI can now give small groups capabilities that were previously available mainly to large organisations or countries. He specifically pointed to cyberattacks, fraud and the possibility of using AI to help develop biological weapons.

“Now small groups with AI can do what only the biggest countries could do,” Gates said.

He said someone looking to attack an electricity network, interfere with banking systems or defraud people could potentially use AI to make those efforts more powerful. Gates also discussed the possibility of AI eventually becoming autonomous enough to act beyond human control. But he said that is not the immediate threat.

“There is not a current risk of these AIs getting completely out of control,” he said.

According to Gates, humans can still shut down the computers running today's AI systems. He said the more urgent concern is preventing people from using those systems for harmful purposes.

“The thing that's urgent has to do with bad people using AI, not the AI going off on its own,” Gates said.

Gates talks about growing food using AI

Gates argued that safeguards and monitoring should be built into advanced AI systems rather than added after a serious incident takes place. He said monitoring would allow developers and authorities to see what sophisticated AI systems are doing and keep records that could help identify harmful activity.

“It’s completely irresponsible not to require every AI to have the safeguards and monitoring capabilities,” Gates said.

Gates also said a simple kill switch would not be enough. In his view, preventing serious incidents requires being able to understand and monitor what an AI system is doing. He wants governments to play a role as well. Gates said legislation is needed so safeguards and monitoring are required across the industry instead of depending only on companies to regulate themselves.

He also argued that countries should work together on the issue. Gates said the US and China, despite competing in AI, should have a common interest in preventing major attacks and other dangerous uses of the technology. At the same time, Gates stressed that AI has significant potential benefits. During the interview, he discussed efforts by the Gates Foundation and other organisations to improve AI support for languages that are poorly represented today.

He said AI could eventually be combined with humanoid robots to help build houses, grow food and provide other services. But for now, he said the focus should be on getting through the period of rapid change without causing too much damage.

“My ability to predict what the world looks like even in 10 years is much less now than during my entire life,” Gates said.

– Ends