Stephen Hawking died on March 14, 2018, after spending the last several years of his life speaking and writing extensively about artificial intelligence. He was unusual among public intellectuals in that he warned about AI's potential dangers with the same seriousness he brought to cosmology, while also acknowledging its potential to do tremendous good. Eight years after his death, his statements about AI read as remarkably prescient given how quickly the technology has moved from research lab to everyday life.
This piece collects what Hawking actually said about AI across his interviews, opinion pieces, public lectures, and his posthumously published book, organized so you can see both the warnings and the hope that he expressed.
The 2014 BBC Interview That Made Headlines
In December 2014, Hawking gave a now famous interview to the BBC tied to a new version of the speech synthesis software he used to communicate. The software itself used machine learning to predict the words he was likely to use next, which gave the interview added weight, because Hawking was effectively speaking through a basic form of AI while warning about more advanced forms of it.
His most quoted line from that interview was direct. "The development of full artificial intelligence could spell the end of the human race." He went on to explain that humans, limited by slow biological evolution, would not be able to compete with a self improving AI that could redesign itself at an ever increasing rate.
He was careful in the same interview to credit existing AI as useful and to distinguish between the narrow AI that helped him communicate and the future general AI he was warning about. The warning was about a long term trajectory, not a claim that current systems were imminently dangerous.
The 2014 Opinion Piece in The Independent
Earlier that same year, Hawking co wrote an opinion piece in The Independent with Stuart Russell, Max Tegmark, and Frank Wilczek. The piece was prompted in part by the release of the film Transcendence and used it as a starting point to make a serious argument that the broader scientific and policy community was not taking AI risk seriously enough.
The most cited line from that opinion piece warned that dismissing the prospect of highly capable AI as science fiction could be, in their words, "potentially our worst mistake in history." The authors argued that the potential benefits of AI were enormous, but that little serious research was being done on how to ensure those benefits would actually be realized and the risks managed.
The opinion piece is worth reading in full because it lays out a balanced argument that Hawking returned to many times. The technology is coming, the upside is enormous, the downside is also enormous, and the question is whether humanity will do the preparation required to capture the first while avoiding the second.
The 2015 Future of Life Institute Open Letter
In January 2015, Hawking was among the prominent signatories of an open letter published by the Future of Life Institute calling for AI research to focus not only on making systems more capable but also on making them robust and beneficial. Other signatories included Elon Musk, leading AI researchers from academia and industry, and dozens of others from across the field.
The letter was notable for being a measured, technical document rather than a doom prediction. It called for prioritizing research into AI safety, alignment, transparency, and societal impact. Hawking's willingness to sign reflected his broader view that the question was not whether AI was good or bad, but whether the people developing it were investing enough in making sure the outcome was beneficial.
The 2016 Opening of the Leverhulme Centre for the Future of Intelligence
In October 2016, Hawking gave a speech at the opening of the Leverhulme Centre for the Future of Intelligence at the University of Cambridge. The Centre was established specifically to study the long term implications of AI, and Hawking's remarks set the tone for the institution.
His most quoted line from that speech captures the balance he tried to strike throughout his AI commentary. He said that the rise of powerful AI would be "either the best, or the worst thing, ever to happen to humanity." He added that we did not yet know which, and that the work of the Centre was part of the effort to make sure it was the former.
He listed specific benefits he hoped for, including the potential to eradicate disease and poverty, and specific risks he wanted studied, including powerful autonomous weapons and the use of AI to oppress people. He emphasized that the technology should be developed with safeguards from the start rather than as an afterthought.
The 2017 Web Summit Speech in Lisbon
In November 2017, Hawking delivered a recorded address to the Web Summit in Lisbon, one of his last major public appearances on the topic before his death. He returned to the "best or worst thing" framing and went further on the specific concerns that had grown over the previous year.
He spoke about the possibility that AI could grow to surpass human intelligence and become a new form of life on Earth. He spoke about the need for international cooperation to manage the development of advanced AI in a way that benefited humanity rather than concentrating power in a small number of hands. He called for the establishment of global standards and ethics for AI development.
He also reiterated his optimism about what AI could do for medicine, climate, and poverty reduction if it was developed responsibly. The speech is consistent with everything he said publicly in his last few years: this is going to be the most important technology humanity has ever developed, treat it accordingly.
The 2017 WIRED Interview
Also in 2017, in remarks reported by WIRED, Hawking returned to his concern that AI could ultimately replace humans, expressing it in stark terms by suggesting that sufficiently successful AI could become a new form of life that would outperform humans.
In the same period he also discussed the more immediate concern of economic disruption, warning that AI driven automation could accelerate the trend of growing inequality unless societies actively chose to redistribute the gains. He framed this as a political and economic question rather than a purely technological one, arguing that the technology itself was largely neutral but the policy choices around it were not.
Brief Answers to the Big Questions
Hawking was working on a book in the final years of his life that was published posthumously in October 2018 as Brief Answers to the Big Questions. The book contains a dedicated chapter titled "Will artificial intelligence outsmart us" that brings together and extends the arguments he had made in interviews and lectures.
In the chapter he reiterated that AI is likely to be either the best or worst thing ever to happen to humanity. He argued that the success of AI could be the biggest event in human history and possibly the last unless we learn how to avoid the risks. He restated the case for serious investment in AI safety research and for international cooperation on the development of advanced AI systems.
The book was assembled from his writings, lectures, interviews, and notes by his family, students, and colleagues. It is the most comprehensive single source for his thinking on AI and worth reading in full for anyone who wants to engage with his views beyond the most widely quoted lines.
The Specific Risks He Warned About
Across all his public statements, Hawking returned to a consistent set of concerns. They are worth listing because they map almost directly to the issues being actively debated in AI policy and governance today.
Loss of human control. The concern that an AI more capable than humans could pursue goals not aligned with human interests, and that humans would not be able to course correct once that gap had opened up.
Autonomous weapons. The concern that AI driven weapons could lower the threshold for conflict, remove humans from the decision to kill, and proliferate beyond the states that originally developed them.
Concentration of power. The concern that a small number of companies or governments could use AI to consolidate wealth, influence, and political control in ways that previous technologies did not enable.
Economic disruption and inequality. The concern that AI driven automation could displace workers faster than societies could adapt, leading to widening inequality unless deliberate political choices were made to address it.
Existential risk from superintelligence. The longer term concern that sufficiently advanced AI could pose a risk to human existence itself, similar in category to risks from nuclear weapons or runaway climate change.
The Specific Benefits He Hoped For
It is easy to remember Hawking as a pessimist on AI because the warnings made the better headlines. The full record shows a more balanced view. He repeatedly identified specific benefits he hoped AI would deliver.
Eradication of disease. He believed AI could accelerate medical research and improve diagnosis and treatment in ways that could substantially reduce the global burden of disease.
Reduction of poverty. He saw AI as a tool that could meaningfully address poverty by improving productivity, agricultural yields, and access to education and services.
Scientific discovery. He spoke about AI's potential to accelerate scientific progress across physics, biology, and other fields where the complexity of the data exceeds what humans can analyze unaided.
Helping people with disabilities. Hawking himself relied on AI assisted communication and was an advocate for the role of technology in expanding what people with severe physical limitations could do.
His position was never that AI should be slowed or stopped. It was that AI should be developed deliberately, with serious investment in safety and ethics, and with international cooperation to ensure that the benefits were broadly shared.
How His Warnings Look Today
Eight years after his death, several of Hawking's concerns have moved from theoretical to immediate.
The pace of AI capability growth has been faster than most observers, including many AI researchers, expected. Large language models with broad general capability are now in widespread use. Autonomous AI agents that take real actions in the world are being deployed across industries. The capability gap between leading AI systems and the next tier is already raising the concentration of power concerns Hawking flagged.
Autonomous weapons development has continued, and the international community has not yet produced binding agreements on their use. Economic disruption from AI is showing up in specific job categories, and the policy debate about how to manage the transition is active in most developed economies.
On the benefit side, AI has begun to deliver on some of the hopes Hawking expressed. AI assisted drug discovery has produced real candidates. Diagnostic AI is improving outcomes in specific areas of medicine. Tools like the communication systems he himself relied on have become dramatically more capable.
The overall picture is consistent with what Hawking predicted. The technology is the most important development of our era. The benefits are real and growing. The risks are real and growing. The question of which side will dominate depends, as he repeatedly said, on the choices being made by the people building and governing the technology right now.
Why His Voice Still Matters
Hawking was not an AI researcher. He was a theoretical physicist whose authority on AI came from a combination of his standing as a scientist, his personal experience with AI assisted technology, and his willingness to engage seriously with questions outside his immediate field. That combination gave him a platform that very few AI researchers had, and he used it to push the conversation toward longer term thinking at a time when public attention was almost entirely on near term consumer applications.
His core contribution was not a technical insight. It was a frame. He insisted that AI was important enough to deserve the kind of careful, multi decade attention that humanity has historically reserved for things like nuclear weapons and climate change. That frame has aged well. A good deal of the serious AI governance work happening today proceeds from broadly similar assumptions, even if the people doing the work are operating in very different institutions than the ones Hawking knew.
The Bottom Line
Stephen Hawking spent the final years of his life arguing that artificial intelligence would likely be either the best or the worst thing ever to happen to humanity, that the outcome depended on the seriousness of the preparation, and that humanity was not yet treating the question with the gravity it deserved. He warned about loss of human control, autonomous weapons, concentration of power, economic disruption, and existential risk. He also hoped for the eradication of disease, the reduction of poverty, accelerated scientific discovery, and expanded capability for people with disabilities.
Eight years after his death, those statements continue to shape how serious people think about AI. Some of the institutions and frameworks now governing AI trace part of their lineage to organizations he publicly supported, and many of them operate within the broader frame of thinking he helped popularize. His warnings were never about slowing the technology. They were about doing the work to make sure it ends well.
The decisions being made by companies, governments, and individuals over the next decade will determine which side of his "best or worst thing" framing humanity lands on. That is a useful frame to carry into any decision you make about how your organization or your work engages with AI.