creators

Geoffrey Hinton — backpropagation, AlexNet, Nobel laureate

Geoffrey Hinton wrote the 1986 backpropagation paper, supervised AlexNet, shared the 2018 Turing Award and the 2024 Nobel Prize in Physics.

Geoffrey Hinton spent four decades making neural networks work and now spends his time warning about what they can do. In 1986 he published “Learning representations by back-propagating errors” in Nature with David Rumelhart and Ronald Williams, showing that a network trained by backpropagation invents its own internal “hidden” features; it is still how nearly every model is trained. With David Ackley and Terrence Sejnowski he had built the Boltzmann machine (Cognitive Science, 1985), a network that learns the statistics of its data rather than a fixed input-output mapping — the work the Nobel committee singled out. In 2012 two of his Toronto students, Alex Krizhevsky and Ilya Sutskever, built AlexNet under his supervision, almost halved the ImageNet error rate, and computer vision changed hands. He kept trying to better his own algorithm: capsule networks (Sabour, Frosst and Hinton, NIPS 2017) encoded pose and part-whole structure; the forward-forward algorithm (2022) replaces the backward pass with a second forward pass, from his long search for something more brain-like. He read experimental psychology at Cambridge, graduating in 1970, took a PhD in artificial intelligence at Edinburgh in 1978, and after postdocs at Sussex and UC San Diego spent five years on the Carnegie Mellon faculty. He moved to the University of Toronto in 1987 on a Canadian Institute for Advanced Research fellowship, unwilling, he has said, to take Pentagon money, then the main funder of American AI. Toronto has been his base since, apart from three years from 1998 founding UCL’s Gatsby Computational Neuroscience Unit; he co-founded the Vector Institute there in 2017. He, Krizhevsky and Sutskever incorporated DNNresearch in 2012; Google bought it in March 2013 and he joined as a vice-president and engineering fellow, splitting his time with the university for a decade. He left Google in May 2023 so he could speak about the risks freely. “I actually want to say some good things about Google,” he told the BBC on 2 May 2023. “And they’re more credible if I don’t work for Google.” He told the New York Times that week that part of him regretted his life’s work, and has since put a number on it: asked on BBC Radio 4’s Today programme in December 2024 whether AI still carried a one-in-ten chance of ending humanity in thirty years, he raised it to 10 to 20 per cent. He is University Professor Emeritus at Toronto and takes no students. His safety work is based at the university’s Schwartz Reisman Institute for Technology and Society, funded since January 2026 by a US$700,000 Good Ventures gift. On 16 September 2026 he briefed members of Congress on Capitol Hill at Bernie Sanders’s invitation and said afterwards that Congress had “maybe a year, but not much more than a year” to regulate AI.

Specialization

deep learning, neural networks, AI safety, backpropagation

Country

Canada