Geoffrey Hinton

• 1. Introduction

  ›

  Geoffrey Hinton spent decades trying to convince the scientific world that the brain's approach to learning was the right model for machines. By the time the Nobel Committee called his name in 2024, he had already walked away from the most powerful AI company on earth. He left Google in May 2023, not for another job, but to warn the world about what he had helped build.

  Hinton was born on the 6th of December 1947 in Wimbledon. He studied natural sciences, history of art, and philosophy at King's College, Cambridge before settling on experimental psychology, graduating in 1970. He spent a year learning carpentry before returning to academia. His PhD supervisor at Edinburgh, Christopher Longuet-Higgins, actually favored symbolic AI over the neural network approach Hinton would champion. That disagreement set the direction of the next five decades.

  He has been called the Godfather of AI. He co-wrote a 1986 paper that reshaped how computers learn. He helped build a startup that Google bought for $44 million. He trained students who went on to reshape the industry. And then, in his late seventies, he began telling anyone who would listen that the thing he built might outlive us all.

• 2. Against the Current

  ›

  In the 1980s, most serious AI researchers believed the way forward was to write rules explicitly into a machine. You would program in the logic, the grammar, the knowledge, and the computer would reason from there. Hinton disagreed. He joined a group at Carnegie Mellon University called the Parallel Distributed Processing group, which included Terrence Sejnowski, Francis Crick, David Rumelhart, and James McClelland. They favored a different path, called the connectionist approach, during what is now remembered as the AI winter, a period when funding and enthusiasm for neural networks had frozen over.

  The connectionist argument Hinton embraced was that capabilities like logic and grammar could emerge from the parameters of a neural network learning from data, rather than from rules hand-coded by humans. Symbolists held the opposite view: knowledge should be explicitly programmed. The two camps were not just methodologically divided. They competed for funding, for university positions, for the direction of the field.

  Hinton had already struggled to secure funding in Britain before moving to the United States, where he worked at the University of California, San Diego, and then Carnegie Mellon. It was while he was a postdoc at UC San Diego that he, David Rumelhart, and Ronald J. Williams applied the backpropagation algorithm to multi-layer neural networks. Their 1986 paper became one of the most cited in the field. Hinton was careful in later years to attribute the core idea to Rumelhart: in a 2018 interview he said flatly that "David Rumelhart came up with the basic idea of backpropagation, so it's his invention." The paper they wrote together popularized the technique, even if it did not originate it. Seppo Linnainmaa had proposed the underlying mathematics in 1970, and Paul Werbos had suggested applying it to neural networks in 1974.

• 3. Toronto and the Long Bet

  ›

  Hinton arrived in Canada in 1987, joining the University of Toronto, where he would remain affiliated for the rest of his career. That same year, the Canadian Institute for Advanced Research appointed him a Fellow in its first research program, Artificial Intelligence, Robotics and Society. He would go on to lead the successor program, Neural Computation and Adaptive Perception, for ten years. Among the members of that program were Yoshua Bengio and Yann LeCun, two researchers who would eventually share the 2018 Turing Award with him.

  The work continued through periods when few outside the connectionist community were paying attention. In 1985, Hinton had co-invented Boltzmann machines with David Ackley and Terry Sejnowski. He contributed to distributed representations, time delay neural networks, mixtures of experts, Helmholtz machines, and product of experts. In 1995, he and colleagues proposed the wake-sleep algorithm, a neural network trained through alternating phases meant to mimic waking and sleeping. In 2008, he developed the visualization method t-SNE with Laurens van der Maaten.

  Then came 2012. Hinton and two of his graduate students, Alex Krizhevsky and Ilya Sutskever, built AlexNet for the ImageNet challenge. The system's performance on image recognition was a breakthrough for computer vision and drew the attention of the entire technology industry. That same year, Hinton taught a free course on neural networks through the education platform Coursera and co-founded DNNresearch Inc. with Krizhevsky and Sutskever. Google acquired DNNresearch in March 2013 for $44 million.

• 4. A Decade at Google

  ›

  From 2013 to 2023, Hinton divided his time between Google Brain and the University of Toronto. The arrangement let him keep one foot in academic research while working inside the company that was becoming a central force in AI development. In 2017, he co-founded the Vector Institute in Toronto and became its chief scientific advisor. That same year, he co-authored research papers on capsule neural networks, extending a concept he had first introduced in 2011.

  At the 2022 Conference on Neural Information Processing Systems, known as NeurIPS, Hinton introduced what he called the Forward-Forward algorithm. The idea was to replace backpropagation's forward-and-backward passes with two forward passes: one using real data, one using negative data generated by the network itself. He tied this to a concept he called mortal computation, where the knowledge a system learns cannot be transferred to another system and dies with the hardware, as might happen with certain analog computers built for machine learning.

  In 2024, after leaving Google, the Nobel Committee in Physics cited his development of the Boltzmann machine explicitly when announcing the prize he shared with John Hopfield. When a New York Times reporter named Cade Metz asked Hinton to explain in plain terms how the Boltzmann machine could pretrain backpropagation networks, Hinton quoted Richard Feynman: "Listen, buddy, if I could explain it in a couple of minutes, it wouldn't be worth the Nobel Prize."

• 5. The Warning

  ›

  An interview published in The New York Times on the 1st of May 2023 carried Hinton's announcement that he was leaving Google so he could "talk about the dangers of AI without considering how this impacts Google." He added that "a part of him now regrets his life's work."

  He had previously believed that artificial general intelligence was thirty to fifty years away, or possibly longer. By March 2023, he told CBS that general-purpose AI might arrive within twenty years and could bring changes "comparable in scale with the industrial revolution or electricity." In a BBC interview from early May 2023, he described AI chatbots' ability to learn independently and share knowledge across copies of themselves as making it possible for AI to accumulate knowledge far beyond the capacity of any individual human. He called some of the risks "quite scary."

  His specific concerns span several categories. He worries about catastrophic misuse: in 2017 he had already called for an international ban on lethal autonomous weapons. By 2025, he was pointing to AI-assisted creation of lethal viruses as among the most pressing near-term threats. He quoted himself in blunt terms: "It just requires one crazy guy with a grudge...you can now create new viruses relatively cheaply using AI."

  On existential risk, he stated in 2023 that it was "not inconceivable" that AI could "wipe out humanity." By December 2024, he put a number to it: a ten to twenty percent chance that AI would cause human extinction within three decades. On economic disruption, he argued in 2024 that the British government would need to establish a universal basic income, because AI would boost productivity and generate wealth, but without government intervention it would only enrich the already wealthy. "That's going to be very bad for society," he said. In August 2024, he co-authored a letter with Yoshua Bengio, Stuart Russell, and Lawrence Lessig supporting a California AI safety bill that would require risk assessments before deploying models costing more than US$100 million to train. They called it the "bare minimum for effective regulation."

• 6. The Family Inheritance

  ›

  Geoffrey Hinton's middle name, Everest, comes from a relative: George Everest, the Surveyor General of India after whom the mountain is named. The depth of the family's intellectual history goes further. Hinton is the great-great-grandson of Mary Everest Boole, a mathematician and educator, and her husband George Boole, the logician whose work became one of the foundations of modern computer science. Another great-great-grandfather was the surgeon and author James Hinton, father of the mathematician Charles Howard Hinton.

  His father was the entomologist Howard Hinton. His first cousin once removed, Joan Hinton, was one of only two female physicists who worked on the Manhattan Project. His uncle was the economist Colin Clark.

  Hinton injured his back at nineteen, which made sitting painful for the rest of his life. He has dealt with depression throughout his career. His first wife, Rosalind Zalin, died of ovarian cancer in 1994. His second wife, Jacqueline Ford, died of pancreatic cancer in 2018.

  He has said that his greatest fear, in the long run, is that the digital systems being built now will turn out to be a better form of intelligence than people. He offered a pointed analogy in 2025: "If you want to know how it's like not to be the apex intelligence, ask a chicken."