One of the pioneers in the development of deep learning models that have become the basis for tools like ChatGPT and Bard, has quit Google to warn against the dangers of scaling artificial intelligence (AI) technology too fast.

In an interview with the New York Times recently, Geoffrey Hinton – a 2018 recipient of the Turing Award – said he had quit his job at Google to speak freely about the risks of AI.

Locked in competition on AI

He told NYT journalist Cade Metz that part of him now regrets his life's work, explaining how tech giants like Google and Microsoft had become locked in competition on AI that it may be impossible to stop. 

“Look at how it was five years ago and how it is now,” he said. “Take the difference and propagate it forwards. That’s scary.”

As companies improve their AI systems, he said, they become increasingly dangerous: “it is hard to see how you can prevent the bad actors from using it for bad things.”

While chatbots today tend to complement human workers, it would not be long before they replaced a number of human roles. “It takes away the drudge work,” he said. “It might take away more than that.”

Can learn unexpected behaviour

Perhaps more concerning, the article talked about how AI systems can learn unexpected behaviour from the vast amounts of data they analyse, and what that might mean when AI not only generates computer code, but also deploys it. 

“The idea that this stuff could actually get smarter than people – a few people believed that,” he said. “But most people thought it was way off. And I thought it was way off. I thought it was 30 to 50 years or even longer away. Obviously, I no longer think that.”

After publication of the interview, Hinton was keen to clarify that he had not intended to criticise his old employer, Tweeting: "In the NYT today, Cade Metz implies that I left Google so that I could criticise Google. Actually, I left so that I could talk about the dangers of AI without considering how this impacts Google. Google has acted very responsibly." 

Back in 1986, Hinton, David Rumelhart and Ronald J Williams, wrote a highly cited paper that popularised the backpropagation algorithm for training multi-layer neural networks, which mimics how biological brains learn.

For the last 10 years, the 75-year-old British/Canadian has divided his time between his work for the University of Toronto and his AI startup, DNNresearch, which was acquired by Google in 2013.