Few scientists have lived long enough to publicly doubt their own life’s work. Geoffrey Hinton is one of them.
Geoff Hinton Photo File Toronto
Often called the “Godfather of AI,” the British-Canadian computer scientist spent decades working on an idea most of his peers dismissed, that machines could learn the way brains do, by adjusting connections between artificial neurons instead of following rules written line by line by human programmers. In 2023, at an age when most people are settling into retirement, he walked away from a senior position at Google so he could speak freely about what he believed that idea had unleashed.
To be fair to the question, Hinton did not single-handedly invent artificial intelligence, and he would be the first to say so. The field has many fathers and invenstions, and the modern boom owes just as much to cheap computing power and vast amounts of data as it does to any single theory. But his stubborn work on neural networks, at a time when the approach was unfashionable and underfunded, helped lay the foundation for the deep-learning systems behind today’s chatbots, image generators, and translation tools. The world eventually caught up to him. The Turing Award, often described as computing’s Nobel, came in 2018, shared with fellow pioneers Yoshua Bengio and Yann LeCun. Then, in 2024, came the real Nobel Prize in Physics, which he shared with John Hopfield.
So, does he regret it? The honest answer is yes and no, and that mixed answer is part of what makes his story. In interviews after leaving Google, Hinton said that a part of him does regret his contribution. That is not the same as saying he wishes he had never done the work. He has also leaned on a line that anyone who has ever rationalized a difficult choice will recognize, he consoles himself with the usual excuse that if he hadn’t done it, somebody else would have. It is a comforting thought, and also a slightly unsettling one, because it admits that the technology was probably unstoppable once the pieces were in place.
What seems to weigh on him most is not the technology itself but how wrong he was about the timeline. For years, he assumed that machines smarter than people were something for the distant future, thirty to fifty years away or more. Then the last few years arrived, and the pace of progress shocked him. There is something almost painful in that admission. This is a man who understands these systems better than almost anyone alive, and even he was caught off guard. If the expert was surprised, what chance did the rest of us have to prepare?
His deeper fear is about what happens when digital minds surpass ours. Unlike a human brain, which can share what it knows only gradually through speech and writing, copies of an AI model can potentially share what they have learned almost instantly. One system learns something, and every copy can potentially benefit from that knowledge. Hinton believes this gives digital intelligence an advantage that biology simply cannot match. From there, the worry follows naturally: a system far more capable than its makers might become impossible to control, or might decide that it has no particular need for us.
Whether you find that scenario convincing is up to you, and plenty of respected researchers think it is overblown. But it is worth noticing that the more everyday harms he points to are already here and need no science fiction. Fake images, cloned voices, and fabricated videos are getting cheaper and more convincing by the month, and bad actors have noticed. In a world where anyone can manufacture “proof” of almost anything, ordinary people may simply stop knowing what to trust online. That erosion of shared reality is arguably more dangerous in the short term than any robot uprising. Add the fear of widespread job losses, with customer service, writing, design, and coding all in the line of fire, and it becomes clear why he sees the risks as more than theoretical.
It is also worth saying that his career has not been a story of sacrifice. Google bought his small startup in 2013 for a reported $44 million, and he spent about a decade there as an engineering fellow and researcher before leaving. The prizes brought recognition and some money, with the Nobel’s roughly $1 million prize split with Hopfield. By Silicon Valley standards, though, he is not wealthy. Estimates of his fortune are modest compared with those of the tech billionaires who built businesses on top of the research he helped make possible, and such figures are rough guesses at best. There is a quiet irony there. The people who profit most from a technology are rarely the ones who warn about it.
That may be why his warnings carry weight. He has little to gain from sounding the alarm, and arguably something to lose, since it puts him at odds with the industry that made his work famous. Critics can fairly ask why the warnings came only after he left, or why those who built the engine are now so keen to describe the crash. Those are reasonable questions. But they do not make his concerns any less worth hearing.
Hinton’s regret looks less like a confession and more like a responsibility he has chosen to take on. He cannot un-invent anything, so he is using his reputation to nudge governments toward taking the risks seriously. Whether they will is the open question. What his story offers, perhaps, is a reminder that the people who build powerful things are not always the best prepared for what those things become and that listening to them when they say so is the least the rest of us can do.
Geoffrey Hinton: The Man Who Bet on the Brain
Geoffrey Everest Hinton was born on December 6, 1947, in Wimbledon, England, and nothing about his early path suggested a man destined to reshape modern technology. He was educated at Clifton College in Bristol and arrived at King’s College, Cambridge, in 1967. There, he drifted. He tried natural sciences, then the history of art, then philosophy, before settling on experimental psychology and graduating with a Bachelor of Arts in 1970. He then did something that rarely appears in the biography of a future Nobel laureate: he spent a year working as a carpenter’s apprentice. It is tempting to read that detour as restlessness, but it also fits the man who later spent his career asking how a mind, built from simple parts, comes to understand the world.
He returned to academia and studied at the University of Edinburgh from 1972 to 1975, earning a PhD in artificial intelligence in 1978. The irony is that his supervisor, Christopher Longuet-Higgins, favored the symbolic approach to AI, in which intelligence is built from explicit rules and logic, and had little faith in neural networks. Hinton believed the opposite. He thought a machine should learn from data and experience, adjusting the connections between artificial neurons much as a brain does, rather than being told what to think line by line. Disagreeing with your own supervisor is a lonely place to start a career, and it set the pattern for what followed.
The following years were not easy. After working at the University of Sussex and at the MRC Applied Psychology Unit, he found it hard to get funding in Britain for his kind of research, so he crossed the Atlantic. In the United States, he worked at the University of California, San Diego, and later at Carnegie Mellon University. Neural networks were deeply unfashionable then, and the field was in the long, cold stretch now remembered as the AI winter. At Carnegie Mellon, he joined the “Parallel Distributed Processing” group alongside scientists such as Terrence Sejnowski, Francis Crick, David Rumelhart, and James McClelland. They held to the connectionist view that abilities like logic and grammar need not be programmed by hand because a neural network can learn them from data. In 1985, Hinton co-invented Boltzmann machines with David Ackley and Terry Sejnowski, and in 1986 he co-authored a highly cited paper with Rumelhart and Ronald J. Williams that popularized the backpropagation algorithm for training multilayer neural networks.
It is worth being careful about credit here, because Hinton himself is. The team did not invent backpropagation. Seppo Linnainmaa had described reverse-mode automatic differentiation, of which backpropagation is a special case, in 1970, and Paul Werbos proposed using it to train neural networks in 1974. In a 2018 interview, Hinton said that Rumelhart came up with the basic idea, so it was his invention. What the 1986 paper did was show, convincingly, that such networks could learn useful internal representations of data. That demonstration changed minds, even if the world took years to act on it.
In 1987, Hinton moved to Canada and joined the University of Toronto, where he has been affiliated ever since, apart from a spell at University College London from 1998 to 2001, where he was the founding director of the Gatsby Charitable Foundation Computational Neuroscience Unit. The same year, he became a fellow of the Canadian Institute for Advanced Research, in its first research program. In 2004, he and his collaborators successfully proposed a new CIFAR program, “Neural Computation and Adaptive Perception,” which he led for ten years and which is now called “Learning in Machines & Brains.” Two of its members, Yoshua Bengio and Yann LeCun, would later share a Turing Award with him. Over the years, his lab became a kind of training ground for the field. Former students and postdoctoral researchers include Peter Dayan, Max Welling, Richard Zemel, Brendan Frey, Radford Neal, Yee Whye Teh, Ruslan Salakhutdinov, Alex Graves, Zoubin Ghahramani, and Ilya Sutskever, whose names now run through the history of modern AI.
The long wait ended in 2012. That year, Hinton taught a free online course on neural networks through Coursera, and he co-founded a small company, DNNresearch Inc., with two of his graduate students, Alex Krizhevsky and Ilya Sutskever. The three also built AlexNet, an image-recognition neural network that won the ImageNet challenge in 2012 and became a breakthrough in computer vision. Almost overnight, an idea that had been dismissed for decades looked like the future. In March 2013, Google acquired DNNresearch for a reported $44 million, and Hinton planned to divide his time between his university research and his work at Google. For the next decade, he did exactly that, working with Google Brain while remaining at Toronto. In 2017, he also co-founded the Vector Institute in Toronto and became its chief scientific adviser, and he serves as a strategic adviser to the Schwartz Reisman Institute for Technology and Society.
Through all this, he never stopped tinkering. His research has covered distributed representations, time-delay neural networks, mixtures of experts, Helmholtz machines, and products of experts, and he has written or co-written more than 200 peer-reviewed publications. In 1995, he and his colleagues proposed the wake-sleep algorithm, which trains a network with alternating “wake” and “sleep” phases. In 2008, he developed the visualization method t-SNE with Laurens van der Maaten. In 2017, he co-authored papers on capsule networks, which aim to better capture how parts fit into wholes within an image, and in 2021 he presented GLOM, a speculative architecture with a similar goal. That same year, he co-authored a widely cited paper on contrastive learning in computer vision. At the 2022 NeurIPS conference, he introduced the “Forward-Forward” algorithm, which replaces the forward and backward passes of backpropagation with two forward passes. It is suited to what he calls “mortal computation,” where the knowledge a system learns cannot be copied to other machines and dies with its hardware. Given his later worries about digital minds that can share everything instantly, the idea seems almost philosophical.
Recognition arrived in force. In 2018, Hinton received the Turing Award together with Bengio and LeCun for their work on deep learning, and the three are sometimes called the “Godfathers of Deep Learning.” In 2024, he and John Hopfield were awarded the Nobel Prize in Physics for foundational discoveries and inventions that enable machine learning with artificial neural networks. For a man who had spent so many years on the margins of his own field, it was a remarkable vindication.
But the story does not end in celebration. In May 2023, Hinton announced his resignation from Google, saying he wanted to speak freely about the risks of AI and adding that part of him now regrets his life’s work. He has voiced concerns about deliberate misuse by malicious actors, technological unemployment, and the existential risk posed by artificial general intelligence. He has also pointed out that safety will require cooperation among the very groups competing to build these systems, since competition alone could lead to the worst outcomes. After receiving the Nobel Prize, he called for urgent research into how humans might control AI systems that are smarter than they are.
Today, Hinton is University Professor Emeritus in the Department of Computer Science at the University of Toronto. His life makes an unusual arc a restless student, a stubborn minority voice, a celebrated pioneer, and now an anxious messenger. He spent decades convincing the world that machines could learn like brains, and he now spends his time asking whether the world is ready for the consequences. Whether history judges him as the architect of a miracle or the author of a warning, it will almost certainly say he was right about the thing that mattered most: that these networks would work.
Has the “Godfather of AI” Geoffrey Hinton Come to Regret Creating Artificial Intelligence?
Few scientists have lived long enough to publicly doubt their own life’s work. Geoffrey Hinton is one of them.
Often called the “Godfather of AI,” the British-Canadian computer scientist spent decades working on an idea most of his peers dismissed, that machines could learn the way brains do, by adjusting connections between artificial neurons instead of following rules written line by line by human programmers. In 2023, at an age when most people are settling into retirement, he walked away from a senior position at Google so he could speak freely about what he believed that idea had unleashed.
To be fair to the question, Hinton did not single-handedly invent artificial intelligence, and he would be the first to say so. The field has many fathers and invenstions, and the modern boom owes just as much to cheap computing power and vast amounts of data as it does to any single theory. But his stubborn work on neural networks, at a time when the approach was unfashionable and underfunded, helped lay the foundation for the deep-learning systems behind today’s chatbots, image generators, and translation tools. The world eventually caught up to him. The Turing Award, often described as computing’s Nobel, came in 2018, shared with fellow pioneers Yoshua Bengio and Yann LeCun. Then, in 2024, came the real Nobel Prize in Physics, which he shared with John Hopfield.
So, does he regret it? The honest answer is yes and no, and that mixed answer is part of what makes his story. In interviews after leaving Google, Hinton said that a part of him does regret his contribution. That is not the same as saying he wishes he had never done the work. He has also leaned on a line that anyone who has ever rationalized a difficult choice will recognize, he consoles himself with the usual excuse that if he hadn’t done it, somebody else would have. It is a comforting thought, and also a slightly unsettling one, because it admits that the technology was probably unstoppable once the pieces were in place.
READ OTHER ARTICLES:
What seems to weigh on him most is not the technology itself but how wrong he was about the timeline. For years, he assumed that machines smarter than people were something for the distant future, thirty to fifty years away or more. Then the last few years arrived, and the pace of progress shocked him. There is something almost painful in that admission. This is a man who understands these systems better than almost anyone alive, and even he was caught off guard. If the expert was surprised, what chance did the rest of us have to prepare?
His deeper fear is about what happens when digital minds surpass ours. Unlike a human brain, which can share what it knows only gradually through speech and writing, copies of an AI model can potentially share what they have learned almost instantly. One system learns something, and every copy can potentially benefit from that knowledge. Hinton believes this gives digital intelligence an advantage that biology simply cannot match. From there, the worry follows naturally: a system far more capable than its makers might become impossible to control, or might decide that it has no particular need for us.
Whether you find that scenario convincing is up to you, and plenty of respected researchers think it is overblown. But it is worth noticing that the more everyday harms he points to are already here and need no science fiction. Fake images, cloned voices, and fabricated videos are getting cheaper and more convincing by the month, and bad actors have noticed. In a world where anyone can manufacture “proof” of almost anything, ordinary people may simply stop knowing what to trust online. That erosion of shared reality is arguably more dangerous in the short term than any robot uprising. Add the fear of widespread job losses, with customer service, writing, design, and coding all in the line of fire, and it becomes clear why he sees the risks as more than theoretical.
It is also worth saying that his career has not been a story of sacrifice. Google bought his small startup in 2013 for a reported $44 million, and he spent about a decade there as an engineering fellow and researcher before leaving. The prizes brought recognition and some money, with the Nobel’s roughly $1 million prize split with Hopfield. By Silicon Valley standards, though, he is not wealthy. Estimates of his fortune are modest compared with those of the tech billionaires who built businesses on top of the research he helped make possible, and such figures are rough guesses at best. There is a quiet irony there. The people who profit most from a technology are rarely the ones who warn about it.
That may be why his warnings carry weight. He has little to gain from sounding the alarm, and arguably something to lose, since it puts him at odds with the industry that made his work famous. Critics can fairly ask why the warnings came only after he left, or why those who built the engine are now so keen to describe the crash. Those are reasonable questions. But they do not make his concerns any less worth hearing.
Hinton’s regret looks less like a confession and more like a responsibility he has chosen to take on. He cannot un-invent anything, so he is using his reputation to nudge governments toward taking the risks seriously. Whether they will is the open question. What his story offers, perhaps, is a reminder that the people who build powerful things are not always the best prepared for what those things become and that listening to them when they say so is the least the rest of us can do.
Geoffrey Hinton: The Man Who Bet on the Brain
Geoffrey Everest Hinton was born on December 6, 1947, in Wimbledon, England, and nothing about his early path suggested a man destined to reshape modern technology. He was educated at Clifton College in Bristol and arrived at King’s College, Cambridge, in 1967. There, he drifted. He tried natural sciences, then the history of art, then philosophy, before settling on experimental psychology and graduating with a Bachelor of Arts in 1970. He then did something that rarely appears in the biography of a future Nobel laureate: he spent a year working as a carpenter’s apprentice. It is tempting to read that detour as restlessness, but it also fits the man who later spent his career asking how a mind, built from simple parts, comes to understand the world.
He returned to academia and studied at the University of Edinburgh from 1972 to 1975, earning a PhD in artificial intelligence in 1978. The irony is that his supervisor, Christopher Longuet-Higgins, favored the symbolic approach to AI, in which intelligence is built from explicit rules and logic, and had little faith in neural networks. Hinton believed the opposite. He thought a machine should learn from data and experience, adjusting the connections between artificial neurons much as a brain does, rather than being told what to think line by line. Disagreeing with your own supervisor is a lonely place to start a career, and it set the pattern for what followed.
The following years were not easy. After working at the University of Sussex and at the MRC Applied Psychology Unit, he found it hard to get funding in Britain for his kind of research, so he crossed the Atlantic. In the United States, he worked at the University of California, San Diego, and later at Carnegie Mellon University. Neural networks were deeply unfashionable then, and the field was in the long, cold stretch now remembered as the AI winter. At Carnegie Mellon, he joined the “Parallel Distributed Processing” group alongside scientists such as Terrence Sejnowski, Francis Crick, David Rumelhart, and James McClelland. They held to the connectionist view that abilities like logic and grammar need not be programmed by hand because a neural network can learn them from data. In 1985, Hinton co-invented Boltzmann machines with David Ackley and Terry Sejnowski, and in 1986 he co-authored a highly cited paper with Rumelhart and Ronald J. Williams that popularized the backpropagation algorithm for training multilayer neural networks.
It is worth being careful about credit here, because Hinton himself is. The team did not invent backpropagation. Seppo Linnainmaa had described reverse-mode automatic differentiation, of which backpropagation is a special case, in 1970, and Paul Werbos proposed using it to train neural networks in 1974. In a 2018 interview, Hinton said that Rumelhart came up with the basic idea, so it was his invention. What the 1986 paper did was show, convincingly, that such networks could learn useful internal representations of data. That demonstration changed minds, even if the world took years to act on it.
In 1987, Hinton moved to Canada and joined the University of Toronto, where he has been affiliated ever since, apart from a spell at University College London from 1998 to 2001, where he was the founding director of the Gatsby Charitable Foundation Computational Neuroscience Unit. The same year, he became a fellow of the Canadian Institute for Advanced Research, in its first research program. In 2004, he and his collaborators successfully proposed a new CIFAR program, “Neural Computation and Adaptive Perception,” which he led for ten years and which is now called “Learning in Machines & Brains.” Two of its members, Yoshua Bengio and Yann LeCun, would later share a Turing Award with him. Over the years, his lab became a kind of training ground for the field. Former students and postdoctoral researchers include Peter Dayan, Max Welling, Richard Zemel, Brendan Frey, Radford Neal, Yee Whye Teh, Ruslan Salakhutdinov, Alex Graves, Zoubin Ghahramani, and Ilya Sutskever, whose names now run through the history of modern AI.
The long wait ended in 2012. That year, Hinton taught a free online course on neural networks through Coursera, and he co-founded a small company, DNNresearch Inc., with two of his graduate students, Alex Krizhevsky and Ilya Sutskever. The three also built AlexNet, an image-recognition neural network that won the ImageNet challenge in 2012 and became a breakthrough in computer vision. Almost overnight, an idea that had been dismissed for decades looked like the future. In March 2013, Google acquired DNNresearch for a reported $44 million, and Hinton planned to divide his time between his university research and his work at Google. For the next decade, he did exactly that, working with Google Brain while remaining at Toronto. In 2017, he also co-founded the Vector Institute in Toronto and became its chief scientific adviser, and he serves as a strategic adviser to the Schwartz Reisman Institute for Technology and Society.
Through all this, he never stopped tinkering. His research has covered distributed representations, time-delay neural networks, mixtures of experts, Helmholtz machines, and products of experts, and he has written or co-written more than 200 peer-reviewed publications. In 1995, he and his colleagues proposed the wake-sleep algorithm, which trains a network with alternating “wake” and “sleep” phases. In 2008, he developed the visualization method t-SNE with Laurens van der Maaten. In 2017, he co-authored papers on capsule networks, which aim to better capture how parts fit into wholes within an image, and in 2021 he presented GLOM, a speculative architecture with a similar goal. That same year, he co-authored a widely cited paper on contrastive learning in computer vision. At the 2022 NeurIPS conference, he introduced the “Forward-Forward” algorithm, which replaces the forward and backward passes of backpropagation with two forward passes. It is suited to what he calls “mortal computation,” where the knowledge a system learns cannot be copied to other machines and dies with its hardware. Given his later worries about digital minds that can share everything instantly, the idea seems almost philosophical.
Recognition arrived in force. In 2018, Hinton received the Turing Award together with Bengio and LeCun for their work on deep learning, and the three are sometimes called the “Godfathers of Deep Learning.” In 2024, he and John Hopfield were awarded the Nobel Prize in Physics for foundational discoveries and inventions that enable machine learning with artificial neural networks. For a man who had spent so many years on the margins of his own field, it was a remarkable vindication.
But the story does not end in celebration. In May 2023, Hinton announced his resignation from Google, saying he wanted to speak freely about the risks of AI and adding that part of him now regrets his life’s work. He has voiced concerns about deliberate misuse by malicious actors, technological unemployment, and the existential risk posed by artificial general intelligence. He has also pointed out that safety will require cooperation among the very groups competing to build these systems, since competition alone could lead to the worst outcomes. After receiving the Nobel Prize, he called for urgent research into how humans might control AI systems that are smarter than they are.
Today, Hinton is University Professor Emeritus in the Department of Computer Science at the University of Toronto. His life makes an unusual arc a restless student, a stubborn minority voice, a celebrated pioneer, and now an anxious messenger. He spent decades convincing the world that machines could learn like brains, and he now spends his time asking whether the world is ready for the consequences. Whether history judges him as the architect of a miracle or the author of a warning, it will almost certainly say he was right about the thing that mattered most: that these networks would work.