Expert leaves Google and warns

AI expert Hinton said that New York Timesthat he has “new fears” about the technology he helped develop. He must be free to be able to speak openly about it, he justified his resignation. He even goes so far as to say that a part of him now “regrets” his life’s work.

Hinton is not just anyone. In 2018 he shared the Turing Award, the equivalent of the Nobel Prize in Computer Science, with Yann Lecun and Yoshua Bengio. Fellow awardee Lecun, chief AI researcher at Zuckerberg’s meta, says of Hinton:

“Geoff’s contributions to AI are tremendous. He didn’t tell me he was planning to leave Google, but I’m not too surprised.”

The 75-year-old computer scientist Hinton had split his time between the University of Toronto and Google after Google acquired Hinton’s AI startup DNNresearch in 2013. DNNresearch was a spin-off from a university research group that was then doing cutting-edge work using machine learning for image recognition. Google has used their technology to improve photo searches, among other things.

Hinton had long raised ethical issues in dealing with AI. He was particularly concerned about their use for military purposes. He once explained that one reason he’d spent much of his career in Canada was that it was easier there to obtain research funding that wasn’t affiliated with the US Department of Defense.

Google’s chief scientist Jeff Dean also acknowledges Hinton’s achievements: “Geoff has made fundamental breakthroughs in the field of AI, and we appreciate his decades of contributions to Google”. Anticipating Hinton’s criticism, Dean emphasizes: “As one of the first companies to publish AI principles, we remain committed to responsible use of AI. We are constantly learning to understand emerging risks while boldly innovating.”

Hinton is best known for his backpropagation algorithm. He first proposed it together with two colleagues in the 1980s.

Today, this technique, which enables artificial neural networks to learn, is the basis of almost all machine learning models. Hinton saw backpropagation as mimicking the learning behavior of biological brains.

“In my many discussions with Geoff, I’ve always been the proponent of backpropagation, and he was always looking for another way of learning that he felt would be more biologically plausible and perhaps a better model for how learning works in the brain,” says Lecun.

“Geoff Hinton certainly deserves the greatest credit for many of the ideas that have made today’s deep learning possible,” said co-awardee Bengio, a professor at the University of Montreal and senior research fellow at Canada’s Montreal Institute for Learning Algorithms.

Bengio surmises, “I suppose that’s why he feels particularly responsible for warning the public about the potential risks that come with advances in AI.”