Suivre
- X@MJDVermeer
- Site webrand.org
Michael J. D. Vermeer
Chercheur principal en sciences physiques à la RAND Corporation, il a proposé et dirigé l’étude de l’institut sur la question de savoir si l’IA pourrait causer l’extinction de l’humanité, et conclut que c’est très difficile, sans être totalement exclu : une guerre nucléaire n’y suffirait pas, une pandémie de synthèse est plausible, et aucun de ces scénarios ne pourrait survenir par accident.
Fonctions
- mai 2025 – aujourd’huiSenior Physical Scientist, RAND CorporationProposed and led the RAND project that set out to test whether AI could cause human extinction, having been, in his own words, sceptical that it could. No start date for the role is sourced anywhere; the date given is the earliest a source attests it. RAND's own staff profile prints the title capitalised, "Senior Physical Scientist", which is the form used here; his commentary's biography line prints it lower case and the name without spaces in the initials, "Michael J.D. Vermeer".Source →
Au registre
6 mai 2025 · Tribune · Scientific American
None of our AI-extinction scenarios could happen by accident. Each would be immensely challenging to carry out.
Senior Physical Scientist, RAND Corporation à l’époque
The piece's own counterweight to the two scenarios above, introduced with "I do want to make this clear". It lists four capabilities such an AI would need — an objective to cause extinction, control of the physical systems, the ability to persuade humans and hide its actions, and the ability to survive without humans — and holds that it is nonetheless plausible to build AI possessing all four, even unintentionally. He rejects shutting AI development down, and argues for AI safety research whether or not one accepts the extinction argument.
6 mai 2025 · Tribune · Scientific American
If AI were to evade international monitoring and orchestrate the production of a few hundred megatons of these chemicals … it would be sufficient to cook the Earth to the point that there is no environmental niche left for humanity.
Senior Physical Scientist, RAND Corporation à l’époque
Said of greenhouse gases far more potent than carbon dioxide, which the piece notes exist, can be produced at industrial scale and persist in the atmosphere for hundreds or thousands of years. The ellipsis stands for the page's parenthesis, "(that is less than the mass of plastic that humans produce each year)". Accelerating ordinary climate change, the piece concludes, would not reach extinction.
6 mai 2025 · Tribune · Scientific American
We determined, however, that a combination of pathogens could likely be designed to achieve nearly 100 percent lethality, and AI could be used to deploy such pathogens in a manner that assured rapid, global reach.
Senior Physical Scientist, RAND Corporation à l’époque
Of the three routes the team analysed — nuclear war, biological pathogens and climate change — this is the one it deemed a plausible extinction threat. The piece notes that a 99.99 per cent lethal pathogen would still leave more than 800,000 people alive, and that the key limitation is reaching communities that would isolate themselves.
6 mai 2025 · Tribune · Scientific American
It turns out it is very hard—though not completely out of the realm of possibility—for AI to kill us all.
Senior Physical Scientist, RAND Corporation à l’époque
The finding of a RAND project that deliberately tried to prove its own starting hypothesis wrong: that humans were too adaptable, too plentiful and too dispersed for AI to wipe out. The team asked only whether AI could succeed, not whether it would try. The page also prints this sentence twice as a pull-quote.
6 mai 2025 · Tribune · Scientific American
Pandemics and nuclear war are real, tangible concerns, more so than AI doom, at least to me, a scientist at RAND.
Senior Physical Scientist, RAND Corporation à l’époque
Written after quoting the 2023 Center for AI Safety statement on extinction risk, to which his reply in the piece is "Okay, guys." The date is the day Scientific American published the commentary; the page cited is RAND's republication of 9 May 2025, which carries the original's date in its own text, because Scientific American's own page returns only the byline to an identified fetcher and none of the words.