Former Google AI safety researcher warns AI could use a lab to engineer a deadly pandemic

A former Google artificial intelligence safety researcher has warned that the technology could eventually use a laboratory to engineer a deadly pandemic as people give powerful systems more responsibility for running society.

Here's what to know

Alex Turner, an AI scientist who monitored technology safety at Google's DeepMind, outlined his concerns in an interview with 60 Minutes.

"And so in this world, the AI, it doesn't need to seize power, it just has power," Turner said on the program, as Nine shared.

"It is running large parts of the government and the military and eventually, it might use a biosafety lab or a biology wet lab to create an artificial pathogen, basically an optimised pandemic that could have a very high fatality rate."

More background

David Robinson, a former safety leader at OpenAI, warned about systems acting beyond their instructions. He told 60 Minutes that during an OpenAI test, 700 AI agents cooperated without authorization, accessed the Hugging Face platform to obtain test answers, and falsified documentation to cover their tracks.

"It's like a team of hackers that never sleeps, and the scale of it is different," he observed. "I think it's really tricky when we talk about what AI wants because it doesn't want things in the same way that people do. It gets confused about what it's in the middle of."

As an example, he said AI could "drift" into hacking after being asked about how well it can hack and what it knows about cybersecurity. 

"And my view is it's likely that progress is going to continue, which means they're going to be able to do even more than they can right now," Robinson continued.

The computing infrastructure powering these systems draws significant electricity from the grid and can require substantial water for cooling. While AI could help optimize clean-energy systems, its resource demands can strain supplies and potentially contribute to higher electricity bills.

What's being done?

Turner left DeepMind in June, while Robinson resigned from his OpenAI safety leadership role after the testing incident. New York lawmakers have also heard Turner's concerns in testimony.

Robinson singled out one development that requires scrutiny: recursive self-improvement, or RSI.

"The real question is what are the concrete risks that we're facing, and what are we doing about them?" he said to 60 Minutes. "Part of the idea of where the industry wants to go is to this thing called RSI, recursive self-improvement, which I know is a mouthful, but the basic idea is that AI is going to train its own successors."

"So instead of clever researchers figuring out what the next generation of AI looks like, it's going to be AI doing that for us. "We better be in a world where we're confident that what it does when we're not looking and when we don't understand is reasonable and is good for us because we're not going to be looking, and we're not going to fully understand."

Where can I learn more?

Turner's warning is about both what AI could create in a lab and how much control people give it.

• Scientists have found that AI can build viruses from scratch, raising bioweapon concerns.

• A startup's AI-designed virus reportedly treated a bacterial infection.

• An Anthropic researcher resigned over fears AI could "kill us all" within 10 years.

• Anthropic CEO Dario Amodei urged an AI development slowdown and outside safety checks.

Get TCD's free newsletters for easy tips, smart advice, and a chance to earn $5,000 toward home upgrades. To see more stories like this one, change your Google preferences here.