What do scientists say about AI?

Fifteen scientists and scholars, and two camps. The first fears losing control. Geoffrey Hinton, accepting his Nobel Prize, said that when digital beings more intelligent than ourselves are created 'we have no idea whether we can stay in control'. Stephen Hawking said in 2014 that full artificial intelligence 'could spell the end of the human race'. Stuart Russell, Max Tegmark, Gary Marcus and Joy Buolamwini ask for control, red lines and the option to say no. The second camp says the machine is not what it looks like. Emily Bender says we cannot help reading meaning into form. Rodney Brooks says a program's competence may be 'extraordinarily narrow' where a person's would not be. Neil Lawrence says its intelligence is different in kind. Kate Crawford says it is 'neither artificial nor intelligent'.

This is one of nine slices of the register of what the world's leaders say about AI, which sets out how every quotation was found and checked twice against its source. 15 scientists and scholars, 15 quotations, grouped by region.

UK#

By taking over the routine, AI will empower us to focus on the things that are traditionally human: creativity, critical thinking, emotional intelligence, and compassion.

Sir Tim Berners-Lee, Inventor of the World Wide Web; Professor, University of Oxford, United Kingdom. Bylined piece published by his publisher around the release of his memoir This Is For Everyone, 18 September 2025. panmacmillan.com. He argued that personal AI agents acting in a person's own interest could free people for distinctively human work.

Their intelligence is different. It is based on very large quantities of data that we cannot absorb. Our computers don't have a complex internal model of who we are.

Neil Lawrence, DeepMind Professor of Machine Learning, University of Cambridge; author of The Atomic Human, United Kingdom. Max Planck Lecture, Stuttgart, 'The Atomic Human', 17 October 2024. inverseprobability.com. He argued that machine intelligence differs in kind from human intelligence, which is shaped by the narrow bandwidth of human communication.

The development of full artificial intelligence could spell the end of the human race. It would take off on its own and re-design itself at an ever increasing rate. Humans, who are limited by slow biological evolution, couldn't compete, and would be superseded.

Stephen Hawking, Theoretical physicist, University of Cambridge, United Kingdom. Interview with BBC News, 2 December 2014. TIME. Asked about the speech-synthesis system he used, he was drawn on the longer-term consequences of developing fully general artificial intelligence.

North America#

All AI systems should be under meaningful human control. This is especially true for those that could be used in the taking of human lives.

Max Tegmark, Professor of Physics, MIT; President, Future of Life Institute, United States. Future of Life Institute statement on autonomous weapons and AI-enabled surveillance, 27 February 2026. futureoflife.org. He called for legal red lines on autonomous weapons and mass surveillance rather than reliance on companies' internal policies.

The problem is that moral concepts are some of the most complex, subtle, context-sensitive, and culturally dependent of all human concepts.

Melanie Mitchell, Davis Professor of Complexity, Santa Fe Institute, United States. Essay 'Magical Thinking on AI' on her newsletter AI: A Guide for Thinking Humans, 15 September 2025. aiguide.substack.com. She argued against proposals that would have AI systems apply moral judgement, on the grounds that moral reasoning is beyond current systems.

…when we experience language, we are always experiencing both the form and the meaning. And in fact, it's really hard to see that they're different because of that experience.

Emily M. Bender, Professor of Linguistics, University of Washington, United States. Interview, 'Unsafe AI for Education: A Conversation on Stochastic Parrots and Other Learning Metaphors', 26 August 2025. Journal of Interactive Media in Education. She argued that people read meaning into language model output because human language experience fuses form and meaning, while the models produce form alone.

At the heart of every AI frontier system, there should be one guiding principle above all: The protection of human joy and endeavour.

Yoshua Bengio, Professor, Université de Montréal; Scientific Director, Mila; Chair of the International AI Safety Report, Canada. Blog post announcing the nonprofit LawZero, 3 June 2025. yoshuabengio.org. Announcing a nonprofit safety research organisation, he set out the principle he argues should govern frontier AI development.

Even the CEOs who are engaging in the race have stated that whoever wins has a significant probability of causing human extinction in the process, because we have no idea how to control systems more intelligent than ourselves.

Stuart Russell, Professor of Computer Science, University of California, Berkeley, United States. Opinion essay, 'DeepSeek, OpenAI, and the Race to Human Extinction', 31 January 2025. Newsweek. Writing after the release of DeepSeek's models, he argued that competition to build artificial general intelligence proceeds without any solution to the control problem.

There is also a longer term existential threat that will arise when we create digital beings that are more intelligent than ourselves. We have no idea whether we can stay in control.

Geoffrey Hinton, Nobel laureate in Physics; Professor Emeritus of Computer Science, University of Toronto, Canada. Nobel Prize banquet speech, Stockholm City Hall, 10 December 2024. nobelprize.org. Accepting the Nobel Prize in Physics, he set out near-term harms of AI and then a longer-term risk of losing control of systems more intelligent than people.

As a technologist, I absolutely believe in increased productivity, but that doesn't automatically translate into shared prosperity. And that's a societal level issue.

Fei-Fei Li, Professor of Computer Science, Stanford University; Co-Director, Stanford Institute for Human-Centered AI, United States. Interview, 'AI Is a Tool, and Its Values Are Human Values', Issues in Science and Technology, April 2024. Issues in Science and Technology (Arizona State University). Asked about AI and the economy, she distinguished productivity gains from the distribution of their benefits.

The option to say no, the option to halt a project, the option to admit to the creation of dangerous and harmful though well-intentioned tools must always be on the table.

Joy Buolamwini, Founder, Algorithmic Justice League; author of Unmasking AI, United States. Opinion essay, 'How I accidentally became a fierce critic of AI', adapted from Unmasking AI, 26 October 2023. The Boston Globe. She argued that researchers and companies must retain the ability to refuse or stop building a system.

We have built machines that are like bulls in a china shop – powerful, reckless, and difficult to control.

Gary Marcus, Professor Emeritus of Psychology and Neural Science, New York University, United States. Written testimony to the US Senate Judiciary Subcommittee on Privacy, Technology, and the Law, hearing on Oversight of AI, 16 May 2023. judiciary.senate.gov. Testifying alongside AI industry witnesses, he argued that current systems are deployed without adequate understanding or oversight.

The extent of the program's competence may be extraordinarily narrow, in a way that would never happen with a person.

Rodney Brooks, Panasonic Professor of Robotics Emeritus, MIT; roboticist, United States. Essay 'What Will Transformers Transform?' on his personal blog, 23 March 2023. rodneybrooks.com. He argued that people wrongly infer broad competence in an AI system from a narrow performance, because that inference works for other people.

We have to figure out how to slow down, and at the same time, invest in people and communities who see an alternative future.

Timnit Gebru, Founder and Executive Director, Distributed AI Research Institute, United States. Interview with IEEE Spectrum about founding DAIR, 31 March 2022. IEEE Spectrum. Explaining why she set up an independent research institute, she argued for slowing the pace of AI deployment and funding alternatives.

AI is neither artificial nor intelligent. It is made from natural resources and it is people who are performing the tasks to make the systems appear autonomous.

Kate Crawford, Research Professor, USC Annenberg; Senior Principal Researcher, Microsoft Research; author of Atlas of AI, United States. Interview with the Guardian on the publication of Atlas of AI, 6 June 2021. Slashdot (reproducing The Guardian interview). She argued that AI systems depend on mined materials and on human labour that is hidden by the appearance of automation.

Two camps that need each other#

The control camp and the not-what-it-looks-like camp are usually presented as opponents, and the register shows why they should not be. Brooks's observation, that we infer broad competence from a narrow performance because that inference works for people, is the mechanism behind automation bias, and automation bias is how control is lost one decision at a time inside an ordinary organisation, long before any existential scenario. Bender's point about form and meaning is the reason people trust fluent output. Melanie Mitchell's, that moral concepts are among the most context-dependent of all, is the reason a machine should not be given the moral call. Hinton's fear and Brooks's scepticism describe the same handover from opposite ends.

Prosperity, direction and the option to stop#

Fei-Fei Li separates productivity from shared prosperity and calls the gap 'a societal level issue'. Timnit Gebru asks to slow down and invest in alternatives. Yoshua Bengio, whose institute chairs the international safety report, sets one principle above all: the protection of human joy and endeavour. Buolamwini's line is the practical one: 'the option to say no, the option to halt a project' must always be on the table. Berners-Lee's optimism, that AI taking the routine will free people for creativity, critical thinking and compassion, is the claim the rest of this site tests against the evidence on deskilling.

The decision that stays with you#

Whatever the scientists conclude about the far future, the near one is decided inside organisations: which decisions a machine may make, who can stop each one, what people must remain able to do without it, and how anyone would know if it went wrong. That is the argument of Rules Before Tools, and the measurable risk sits there, in the handover of decisions to machines and what happens to human judgement afterwards.

What these quotations do not show#

Scientists disagree with each other, and the two camps here are a reading, not a taxonomy anyone signed up to. Hinton and Bengio built the field they now warn about, which cuts both ways. Kate Crawford holds a post at Microsoft Research and Fei-Fei Li co-founded a company after the interview quoted; both are included as scholars, and the affiliations are stated. Hawking's remark is from 2014 and Crawford's from 2021, before the current models. Crawford's words are checked against a verbatim reproduction of a Guardian interview rather than the Guardian's page. Thirteen of the fifteen are in North America or Britain; no scientist from Asia, Africa or Latin America is here yet.