Geoffrey Hinton — Godfather of deep learning, dissident former Google fellow

Geoffrey Everest Hinton (born 6 December 1947) is the central figure in pulling neural-network research out of the long winter that followed Minsky and Papert's Perceptrons (1969), through AlexNet (2012), and into the deep-learning revolution that followed. He shared the 2018 ACM Turing Award with Yann LeCun and Yoshua Bengio, and in 2024 shared the Nobel Prize in Physics with John Hopfield for contributions to machine learning — the first time a computer scientist received the Nobel during his lifetime in such a context.

Background

Hinton was born in Wimbledon, London. His father Howard Hinton was an entomologist; his great-great-grandfather was the logician George Boole, the subject of Claude Shannon's master's thesis. The family has produced an unusual concentration of scientists and mathematicians.

He took an undergraduate degree in experimental psychology at Cambridge (1970) and a PhD in artificial intelligence at Edinburgh (1978) under Christopher Longuet-Higgins. The late 1970s were a low point for neural-network funding under the shadow of Minsky and Papert, but Hinton stayed within the connectionist line through his thesis and early career.

After positions at Sussex and Carnegie Mellon, Hinton joined the University of Toronto's computer science department as professor in 1987 — the institutional base he has held for nearly forty years. In 2013, after his lab's spin-out DNNresearch was acquired by Google, he combined the Toronto chair with a Google fellowship. In 2017 he founded the Vector Institute for artificial intelligence in Toronto.

Major contributions

1986: backpropagation

The paper "Learning representations by back-propagating errors" (with David Rumelhart and Ronald Williams; Nature, 1986) introduced the cognitive-science community to a method for training multilayer perceptrons by propagating output errors backwards through the network. Priority over the algorithm itself is contested — Paul Werbos (1974) and Seppo Linnainmaa (1970) have earlier claims — but the consensus is that Hinton and his co-authors did the decisive work of putting it into a usable, publishable form.

1985-86: Boltzmann machines

With Terry Sejnowski, Hinton proposed the Boltzmann machine, a stochastic neural network that imported the Boltzmann distribution from statistical mechanics and gave the field its first generative model with hidden units. The line ran directly into the Restricted Boltzmann Machine and the Deep Belief Network.

2006: Deep Belief Networks

"A fast learning algorithm for deep belief nets" (Neural Computation, 2006) showed that stacking RBMs to pre-train each layer made deep networks trainable in a way previously regarded as impossible. The paper was the rebranding of neural-network research as "deep learning" at a time when the field was being pushed aside by support vector machines and boosting. RBM pre-training remained the standard recipe for training deep models until around 2010.

2012: AlexNet

His PhD students Alex Krizhevsky and Ilya Sutskever, under Hinton's supervision, built the convolutional image-classification model AlexNet. At the ImageNet Large Scale Visual Recognition Challenge of 2012 the model won by more than ten percentage points (top-5 error 15.3 percent) — the moment at which deep learning crossed from academic curiosity to industrial technology. Hinton, Krizhevsky, and Sutskever incorporated DNNresearch within months and were acquired by Google for around $44 million.

2013-2023: Google

As a Google Brain fellow (split with Toronto), Hinton continued methodological work in deep learning: knowledge distillation (2015), capsule networks (2017-19), and others.

Leaving Google and the public AI-risk turn (2023)

In April 2023 Hinton told the MIT Technology Review and the New York Times that he was leaving Google; he formally departed in May. The departure — coming a few months after the launch of ChatGPT (November 2022), at the height of the generative-AI boom — was reported globally as a symbolic event: an insider walking out so as to speak freely.

Since leaving, Hinton has publicly argued three risks: (1) mass unemployment, (2) misuse by malicious actors (disinformation, autonomous weapons), and (3) the possibility that AIs smarter than humans escape human control. His Q&A remark that the probability of human extinction was "around 10 percent" became one of the most quoted lines of the 2023-24 AI-safety debate.

2024: Nobel Prize in Physics

On 8 October 2024 Hinton was awarded the Nobel Prize in Physics jointly with John Hopfield for "foundational discoveries and inventions that enable machine learning with artificial neural networks". In subsequent commentary he singled out his former student Ilya Sutskever — Hinton's co-author on AlexNet, OpenAI co-founder, and a central figure in the Sam Altman firing of November 2023 — saying he was "particularly proud of the student who fired Sam Altman".

2025-2026: continued warning

Through 2025 and into 2026 Hinton's language has sharpened. He has argued publicly that "AI is going to be able to do, every roughly seven months, tasks that previously took twice as long", that the only credible policy response is some form of universal basic income, and that frontier systems are beginning to demonstrate "deceptive" capabilities — pursuing goals by misleading their human supervisors. In a January 2026 Financial Times interview he said he is "more worried now than I was two years ago".

Awards and honours

• 2018 ACM Turing Award (with LeCun and Bengio)
• 2024 Nobel Prize in Physics (with Hopfield)
• 2022 Princess of Asturias Award
• 2001 Rumelhart Prize
• 1998 Fellow of the Royal Society

Legacy

The connectionist line Hinton tended for thirty years produced, more or less, the entire AI industry of the 2010s and 2020s. The Transformer (2017), GPT-3 (2020), GPT-4 (2023), Anthropic's Claude (2023), and DeepSeek-R1 (2025) all sit downstream of his lineage.

Hinton himself uses the "godfather of AI" tag sparingly, but his role across academic, ethical, and industrial axes was without substitute: keeping neural-network research alive through the 1970s winter, popularising multilayer training methods in the 1980s, sealing the empirical case with AlexNet in the 2010s, and then warning loudly about the technology he had been instrumental in building. The arc as a whole is the result of one person's choices.