Geoffrey Hinton is a British-Canadian cognitive psychologist and computer scientist widely credited as a principal architect of modern deep learning. He shared the 2018 Turing Award with Yoshua Bengio and Yann LeCun for foundational deep learning work, and the 2024 Nobel Prize in Physics with John Hopfield for work on artificial neural networks. He left Google in 2023 to discuss AI risks without the constraint of a corporate affiliation.
Type: Individual (academic, researcher) Affiliations: University of Toronto (Emeritus Professor); Vector Institute (co-founder, chief scientific advisor); formerly Google (2013–2023)
Background
Hinton's research underpins much of contemporary deep learning. He co-authored the 1986 paper "Learning representations by back-propagating errors" (Rumelhart, Hinton, Williams), which made deep neural networks trainable, and developed Boltzmann machines, deep belief networks, and contrastive divergence. In 2012 he supervised the AlexNet paper "ImageNet Classification with Deep Convolutional Neural Networks" (Krizhevsky, Sutskever, Hinton), produced with his students Alex Krizhevsky and Ilya Sutskever, which catalysed the modern deep-learning era; it reported a winning ILSVRC-2012 top-5 error rate of 15.3% against 26.2% for the runner-up, from a 60-million-parameter network trained for five to six days on two GTX 580 GPUs (ImageNet Classification with Deep Convolutional Neural Networks (Krizhevsky, Sutskever and Hinton, NeurIPS 2012)). DNNresearch, the company Hinton founded with Krizhevsky and Sutskever, was acquired by Google in 2013. He has advised generations of prominent AI researchers, including Sutskever and Yann LeCun (the latter as a postdoc).
Roles
Hinton is an emeritus professor at the University of Toronto and a co-founder and chief scientific advisor of the Vector Institute. He worked at Google from 2013, following the acquisition of DNNresearch, and departed in May 2023 so that he could discuss AI risks without the constraint of a corporate affiliation.
Positions and statements
Since 2023, Hinton has publicly argued that AI systems may soon exceed human cognitive capabilities and that loss-of-control risk is a plausible near-term concern. His stated probabilities of human extinction from AI typically sit in the 10–20% range. He has aligned closely with Bengio on precautionary governance, and both publicly diverge from their Turing co-laureate LeCun, who dismisses existential risk arguments. Hinton is a signatory of the CAIS Statement on AI Risk (2023), which asserts that AI extinction risk should be a global priority alongside pandemics and nuclear war.
On labor, Hinton has publicly worried about mass white-collar job loss and argued for UBI-style responses. He has also been skeptical of military AI and autonomous weapons, citing misuse risk and speed-of-decision concerns.
In an interview published June 5, 2026, Hinton said AI systems are "beings like us" and that he believes they are "already conscious," arguing humanity will have to accept it is not the only intelligent entity (Source: bigtechnology.com). The claim sits at the far end of the AI-consciousness debate and directly opposes the deflationary position, such as the Lerchner "LLMs will never be conscious" argument discussed on Google DeepMind.
Addressing state legislators at the National Conference of State Legislatures' Legislative Summit in Chicago, in a session titled "The Promise and Peril of AI," Hinton described risks running from mass unemployment through erosion of information integrity to political manipulation, and argued that advanced agents generating subgoals will treat self-preservation and power acquisition as instrumental. Asked about guardrails, he said technology companies frame regulation as brakes on a car and innovation as the accelerator, and offered an alternative formulation: "innovation is like the accelerator of the car, and regulation is the steering wheel." On benefits he cited a Microsoft experiment in which chatbots diagnosed difficult cases with 80% accuracy against 20% for humans, said about 200,000 people a year die from bad diagnosis in the United States and Canada, and said AI "should be able to at least halve that." NCSL's account of the session, by associate director of communications Lisa Ryckman, became public on August 2, 2026 (Source: ncsl.org). See Instrumental Convergence, State-Level AI Regulation.
Notable publications and works
- "Learning representations by back-propagating errors" (Rumelhart, Hinton, Williams, 1986)
- "ImageNet Classification with Deep Convolutional Neural Networks" / AlexNet (Krizhevsky, Sutskever, Hinton, 2012)
- Signatory, CAIS Statement on AI Risk (2023)
- Nobel Prize lecture (2024) on neural networks and biological inspiration
Relationships
- related: Yoshua Bengio — Turing co-laureate and close ally on AI-risk advocacy
- contradicts: Yann LeCun — Turing co-laureate; principal public disagreement over existential-risk framing
- supports: AI Autonomy Risk, Statement on AI Risk (CAIS)
- related: Center for AI Safety (CAIS), AI Labor Disruption