The godfather of AI: why I left Google

Ten minutes before I meet Geoffrey Hinton, the ‘godfather of AI’, the New York Times announces he’s leaving Google. After decades working on artificial intelligence, Hinton now believes it could wipe out humanity. ‘It is like aliens have landed on our planet and we haven’t quite realised it yet because they speak very good English,’ he says. He also tells me that he has been unable to sleep for months.

Hinton, 75, revolutionised AI not once but twice: first with his work on neural networks, computer architecture that closely resembles the brain’s structure, and then with so-called ‘deep learning’, which allows AI to refine and extract patterns and concepts on a vast quantity of data. He persevered with neural networks during the 1970s and 1980s when the industry had largely abandoned it. In 1986 he designed the very first chatbot, then called a ‘language predictor model’. Decades later, in 2012, he developed a deep-learning AI, the intellectual foundation for ChatGPT.

Hinton comes from a family of mathematicians and scientists (his grandfather was George Boole, who formulated Boolean algebra, laying the foundations for the entire digital age). He has had his qualms about AI in the past, but only over its application rather than the fundamentals. He loathed the development of autonomous weapons, and left his US job because of the State Department funding of AI. He’s also been worried about the ability of AI to help ‘bad actors’ such as Vladimir Putin pump out fake news. ‘I’d always kind of ignored the existential threat before about it wiping out humanity,’ he says. ‘Now I think there’s a significant chance of that.’

What changed? ‘Google had a big language model called Palm which could explain why a joke was funny. That shocked me. I arrived at the conclusion – this might just be a much better form of intelligence.