AI scientists sign a one-sentence statement on extinction risk

Background

Public argument about the dangers of AI has mostly concerned problems that are easy to picture: biased systems, displaced jobs, a particular tool put to bad use. The idea that AI might carry a risk on the scale of a pandemic or a nuclear war has circulated mainly among a smaller group of researchers, usually under the heading of existential risk — the possibility that a technology ends human life or permanently narrows humanity's prospects. For such a concern to be treated as a public issue rather than a specialist's worry, it helps to have a short document that many senior people are willing to attach their names to.

The statement

On 30 May 2023 the Center for AI Safety published such a document. It was one sentence long: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."^[1] Reports in Chinese media likewise dated the release to 30 May 2023.^[2]

That is the whole text. It names no technology, proposes no regulation and sets no deadline; it makes a claim about priority, and about the scale at which the risk should be considered — the same scale as a pandemic or a nuclear war.

Who signed

The signatories included the Turing Award winners Geoffrey Hinton and Yoshua Bengio, and lab leaders such as Sam Altman of OpenAI and Demis Hassabis of Google DeepMind. Dario Amodei, chief executive of Anthropic, was among the notable signatories as well.^[1] The Center said the list also included the authors of the standard textbooks on artificial intelligence, deep learning and reinforcement learning, and three Turing Award winners.^[1]

On the institutional side, the Center said executives from Microsoft, OpenAI, Google, Google DeepMind and Anthropic had signed, while leaders from Meta had not.^[1] It also described how the names were checked: signatories were verified through email confirmation or personal contact.^[1]

Why one sentence

Dan Hendrycks, director of the Center for AI Safety, explained the timing by analogy. Pandemics, he said, were not on the public's radar before COVID-19, and it is not too early to put guardrails in place for AI.^[1] The comparison is about prevention: measures taken before a harm arrives are generally easier to design than measures taken afterwards.

The shape of the statement follows from its purpose. A sentence this broad can be signed by people who disagree about methods, timelines and much else, because it asks them to agree only that the risk deserves priority. A longer document listing research agendas or policy proposals would have asked signatories to agree on far more than the risk itself.

Where it sits

The roster crosses institutional lines, bringing together researchers and executives from companies that otherwise compete with one another. It also mixes communities that rarely sign the same page: the people who built the field's teaching material, the people who run its largest laboratories, and, in the case of the Turing Award winners, researchers whose work made modern machine learning possible.

The statement thus works as a roster as much as an argument. Its weight comes less from the reasoning it contains — there is only one sentence of that — than from who is on the list, and from the fact that the risk of extinction is the framing they chose to put their names to.