Geoffrey Hinton

Geoffrey Everest Hinton Template:Post-nominals (born 6 December 1947) is a British-Canadian computer scientist and cognitive psychologist whose work on artificial neural networks and deep learning earned him the reputation as the "Godfather of AI." He received the Turing Award in 2018 (shared with Yann LeCun and Yoshua Bengio) for conceptual and engineering breakthroughs that enabled deep neural networks to become a critical component of computing, and the Nobel Prize in Physics in 2024 (shared with John Hopfield) for foundational discoveries that enable machine learning with artificial neural networks.

Hinton held a position at the University of Toronto for over 40 years and was a vice president and engineering fellow at Google from 2013 to 2023. In May 2023, he resigned from Google to speak freely about the existential risks posed by artificial intelligence, becoming one of the most prominent voices warning about the dangers of the technology he helped create.

Early life and education

Geoffrey Hinton was born in Wimbledon, London, into a distinguished scientific family. He is a great-great-grandson of the mathematician George Boole, whose Boolean algebra underpins all of digital computing. His father, Howard Everest Hinton, was an entomologist at the University of Bristol. His cousin is the surgeon and author Atul Gawande.

Hinton studied experimental psychology at the University of Cambridge (King's College), graduating with a BA in 1970. After a brief period studying carpentry — motivated by uncertainty about whether AI research was viable — he returned to academia and received his PhD in artificial intelligence from the University of Edinburgh in 1978, supervised by Christopher Longuet-Higgins. His doctoral thesis explored the use of relaxation methods in neural computation.

Career

Academic positions

After his PhD, Hinton held postdoctoral and faculty positions at several institutions:

University of Sussex (1978–1980) — research fellow.
University of California, San Diego (1982–1987) — faculty member in the Department of Computer Science and the Institute for Cognitive Science, where he collaborated with David Rumelhart and Ronald Williams on backpropagation.
Carnegie Mellon University (1982–1987) — concurrent appointment.
University of Toronto (1987–present) — University Professor Emeritus in the Department of Computer Science. Hinton moved to Canada partly because he objected to military funding of AI research in the United States during the Reagan era.

At Toronto, Hinton founded the program that became the epicentre of the deep learning revolution, training a generation of researchers including Ilya Sutskever, Yann LeCun (who also studied under him as a postdoc), Alex Krizhevsky, and many others.

Google (2013–2023)

In March 2013, Google acquired DNNresearch Inc., a startup Hinton had formed with two of his graduate students (Alex Krizhevsky and Ilya Sutskever), for a reported $44 million. Hinton joined Google as a vice president and engineering fellow, dividing his time between Google Brain in Toronto and Mountain View. At Google, he contributed to advances in speech recognition, image recognition, and the development of techniques that fed into products used by hundreds of millions of people.

Departure from Google (2023)

On 1 May 2023, Hinton resigned from Google, telling The New York Times that he wanted to speak freely about the dangers of AI without considering the impact on Google. He expressed regret for his life's work, saying "I console myself with the normal excuse: if I hadn't done it, somebody else would have." He specifically warned about the risks of AI being used for misinformation, job displacement, and ultimately posing an existential threat to humanity.

Scientific contributions

Backpropagation

Hinton's most influential early contribution was his role in popularising backpropagation — the algorithm for training multi-layer neural networks by computing gradients of the loss function with respect to each weight. While the mathematical basis was known earlier (Seppo Linnainmaa's 1970 work on automatic differentiation, Paul Werbos's 1974 thesis), the 1986 Nature paper by David Rumelhart, Hinton, and Williams ("Learning representations by back-propagating errors") provided the definitive experimental demonstration that backpropagation could train useful multi-layer networks. This paper is one of the most cited in all of science, with over 40,000 citations.

Boltzmann machines

In the early 1980s, Hinton and Terrence Sejnowski developed Boltzmann machines — stochastic recurrent neural networks that can learn internal representations, inspired by statistical mechanics. The restricted Boltzmann machine (RBM), a simplified bipartite variant, became a key building block of deep learning when Hinton showed in 2006 that stacking RBMs as a deep belief network allowed layer-by-layer unsupervised pre-training, enabling the training of deep architectures that had previously been intractable.

Deep belief networks and the deep learning revival

Hinton's 2006 paper in Science ("Reducing the Dimensionality of Data with Neural Networks," with Ruslan Salakhutdinov) is widely credited with reigniting interest in deep architectures after the long "AI winter" for neural networks. The key insight was that deep networks could be initialised via greedy layer-wise pre-training using RBMs, then fine-tuned with backpropagation. This approach made it practical to train networks with many layers, which had previously suffered from vanishing gradients when trained from random initialisation.

AlexNet and the ImageNet revolution

In 2012, Hinton's students Alex Krizhevsky and Ilya Sutskever, under Hinton's supervision, developed AlexNet — a deep convolutional neural network that won the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) with a top-5 error rate of 15.3%, compared to 26.2% for the second-place entry. This result demonstrated unequivocally that deep neural networks trained on GPUs with large datasets could dramatically outperform traditional computer vision methods. The 2012 ImageNet result is widely considered the starting point of the modern deep learning era.

Capsule networks

Beginning in 2011, Hinton proposed capsule networks as an alternative to conventional convolutional neural networks. Capsules are groups of neurons whose output vectors represent the instantiation parameters of a specific type of entity. Unlike pooling layers in CNNs, capsules aim to preserve spatial hierarchies. While capsule networks have not displaced standard architectures, they represent an ongoing line of research into more structured representations.

Other contributions

Dropout (2012, with Nitish Srivastava et al.) — a regularisation technique where random neurons are temporarily removed during training, dramatically reducing overfitting. Now standard in deep learning.
Distillation (2015, with Oriol Vinyals and Jeff Dean) — compressing knowledge from a large "teacher" model into a smaller "student" model by training on soft probability distributions rather than hard labels.
Variational autoencoders — contributions to generative modelling with latent variables.
Contrastive learning — work on representation learning through contrastive objectives (SimCLR and related).

Views on AI risk

Since leaving Google, Hinton has become one of the most prominent voices warning about the dangers of advanced AI. His key concerns include:

Superintelligence — Hinton has argued that AI systems may become more intelligent than humans sooner than most experts expect, possibly within 5–20 years, and that such systems could be difficult or impossible to control.
Misinformation — AI-generated text, images, and video could make it impossible for ordinary people to distinguish truth from fabrication.
Labour displacement — AI could automate a large fraction of existing jobs, increasing inequality.
Autonomous weapons — AI-powered weapons could lower the threshold for conflict.
Power concentration — AI could further concentrate power among those who control the technology.

Hinton has called for government regulation of AI and has supported the 2023 Statement on AI Risk, which stated that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

At the same time, Hinton has acknowledged AI's enormous potential for good, particularly in healthcare and scientific research.

Awards and honours

Turing Award (2018) — shared with Yann LeCun and Yoshua Bengio, "for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing."
Nobel Prize in Physics (2024) — shared with John Hopfield, "for foundational discoveries and inventions that enable machine learning with artificial neural networks."
Companion of the Order of Canada (CC) (2018)
Fellow of the Royal Society (FRS) (1998)
Fellow of the Royal Society of Canada (FRSC)
NSERC Herzberg Gold Medal (2010) — Canada's highest honour for science and engineering.
IEEE Frank Rosenblatt Award (2014)
BBVA Foundation Frontiers of Knowledge Award (2016)
Honda Prize (2016)
Numerous honorary doctorates from universities including Edinburgh, Sussex, Sherbrooke, and others.

Personal life

Hinton has described himself as an atheist. He has chronic back pain, which famously prevented him from sitting, leading him to work standing up and to use a specially designed reclining workstation. He has been known to avoid flying and to travel by train or ship when possible, though he has made exceptions for the Nobel Prize ceremony.

He has two children from his first marriage to the molecular biologist Ros Howard, who died of ovarian cancer in 1994. He later married Jackie Forde.

Selected publications

Rumelhart, D. E.; Hinton, G. E.; Williams, R. J. (1986). "Learning representations by back-propagating errors." Nature, 323(6088), 533–536.
Hinton, G. E.; Sejnowski, T. J. (1986). "Learning and Relearning in Boltzmann Machines." In Parallel Distributed Processing, Vol. 1, Chapter 7.
Hinton, G. E.; Salakhutdinov, R. R. (2006). "Reducing the Dimensionality of Data with Neural Networks." Science, 313(5786), 504–507.
Krizhevsky, A.; Sutskever, I.; Hinton, G. E. (2012). "ImageNet Classification with Deep Convolutional Neural Networks." NIPS 2012.
Srivastava, N.; Hinton, G.; Krizhevsky, A.; Sutskever, I.; Salakhutdinov, R. (2014). "Dropout: A Simple Way to Prevent Neural Networks from Overfitting." JMLR, 15(1), 1929–1958.
Hinton, G.; Vinyals, O.; Dean, J. (2015). "Distilling the Knowledge in a Neural Network." arXiv:1503.02531.