But we should still be careful, he warns
-
AI-caused human extinction is possible but extremely difficult, RAND researchers found
-
Nuclear war and climate change are unlikely extinction routes; pandemics are a bigger threat
-
Shutting down AI research isnt the answerbut stronger safety and risk planning is essential
Could artificial intelligence wipe out the human race? According to Michael J.D. Vermeer, a senior physical scientist at the RAND Corporation, that outcomeis not impossible but highly unlikely. Still, he says, its smart to take the risks seriously.
In a recent piece originally published by Scientific American, Vermeer explained how his team at RAND known for their deep background in national security set out to explore whether AI really could cause human extinction, as some researchers and tech leaders fear.
We started with the idea that AI couldnt kill us all, Vermeer wrote. That humans were too spread out, too resilient. But if we could prove ourselves wrong, it would mean the threat was real.
Vermeers team, made up of a scientist, an engineer, and a mathematician, explored how AI might exploit three of humanitys biggest vulnerabilities: nuclear weapons, pandemics, and climate change.
Nuclear war: Devastating, but not extinction
First, they looked at nuclear war. Even if AI somehow launched all 12,000+ nuclear warheads on Earth, the fallout likely wouldnt kill every human.
Humans are too scattered, and nuclear weapons just arent enough to reach every corner of the globe, Vermeer explained. Some people would survive and could rebuild.
Pandemics: The most plausible threat
AI-driven pandemics were a different story. The researchers found that with the right combination of engineered pathogens, AI could possibly create a disease that would be nearly 100% lethal. The challenge, Vermeer said, would be infecting every isolated group of people fast enough.
Youd have to reach the communities that immediately lock themselves down, he said. Thats a real hurdle.
Climate change: Possible, but requires a lot of work
Accelerating global warming, even drastically, would likely not be enough to kill us all either. But Vermeer warned about super greenhouse gases chemicals more dangerous than CO, which could, in theory, make Earth unlivable if released at massive scale.
These gases exist, he wrote. And if AI found a way to make and spread a few hundred megatons of them, it could heat the Earth enough to erase every place where humans could survive.
So, could AI really do it?
Maybe. But Vermeer stressed that no scenario would happen by accident. For AI to wipe out humanity, it would have to:
-
Want to cause extinction
-
Control the tools of destruction (like nukes or chemical plants)
-
Persuade humans to help and stay hidden
-
Operate independently once society collapses
If AI didnt have all four of these, its extinction plan would fail, Vermeer concluded. But those capabilities are being developed, even if unintentionally.
Dont shut it down manage the risk
Despite the grim possibilities, Vermeer doesnt support stopping AI development entirely. We care too much about AIs benefits, he said. Instead, we should invest in AI safety research and reduce extinction risks in generalthings like limiting nuclear weapons, monitoring climate-altering chemicals, and improving disease detection.
In the end, he says, AI may be a threatbut were already capable of destroying ourselves. Being cautious about AI is just part of being responsible about the future.
Posted: 2025-06-04 16:47:55
