There’s this super popular sci-fi cliché that artificial intelligence is like, gonna go totally rogue one day and wipe out all of humanity. Could this really, like, happen? Some AI researchers are all like, “Um, yeah, it’s totally possible that AI could extinctify humans.” In 2024, like, hundreds of these brainiacs signed a statement saying, “Yo, we should totally make preventing AI from wiping us out a top priority, just like pandemics and nuclear war.”
So, like, okay, some peeps are all worried about pandemics and nukes, which makes sense ‘cause those are some real, tangible threats. Me, I’m just a scientist at RAND Corporation who’s like, “Eh, maybe not so much on the whole AI doomsday thing.” RAND Corporation is all about national security stuff, like how to prevent nuclear catastrophes back in the day. So, I was all skeptical about AI’s potential to extinctify humans, and I was like, “Hey, let’s do a project to see if AI could actually pull it off.”
My team’s guess was that there’s no way AI could definitively wipe out all humans. We were thinking, like, humans are too adaptable, too many, and spread out all over the planet for AI to completely obliterate us with any tools it might have. If we were wrong, then it would mean that AI could be a legit threat to humanity.
We had a scientist, an engineer, and a mathematician on our team. We put aside our doubts about AI and started figuring out how AI could potentially cause human extinction. We weren’t just looking at regular disasters or societal collapses. We wanted to focus on a full-on extinction event. We didn’t care if AI would try to kill us, we were just interested in whether it could actually succeed.
It was a pretty grim task. We dug deep into how AI could exploit three major threats that are often seen as existential risks: nuclear war, biological pathogens, and climate change.
Turns out, it’s like, seriously tough for AI to wipe out all of us. The good news, if you can call it that, is that we don’t think AI could end us all with nukes. Even if AI somehow got access to all the nukes on the planet, the explosions and fallout still wouldn’t be enough for a total extinction event. There are just too many humans spread out all over for nukes to kill everyone. AI could blow up all the most densely populated areas and still not cause as much chaos as the asteroid that probably killed the dinosaurs. Plus, there aren’t enough nukes to totally ruin all the farmland on Earth. So, even if AI went all out with nukes, it would be a disaster but not a total wipeout.
On the other hand, we figured pandemics could be a real threat to extinction. Past plagues have been super bad, but humans always manage to bounce back. Even with a pathogen that’s like 99.99% lethal, there would still be more than enough humans left to keep going.
But, if AI orchestrated a combo of pathogens that were close to 100% lethal, and used AI to spread them globally super fast, then that could be a problem. The catch is that AI would need to somehow infect or wipe out communities that would try to isolate themselves in the face of a pandemic.
Lastly, if AI sped up regular ol’ climate change caused by humans, it still wouldn’t lead to total human extinction. We’d probably find new places to live, even if it meant moving to the poles. To make Earth completely uninhabitable for humans, AI would have to pump out way more powerful greenhouse gases than what we’re currently dealing with.
I just wanna point out that none of these AI-extinction scenarios could happen by accident. Each one would be like, crazy hard to pull off. AI would have to overcome some major hurdles.
During our analysis, we identified four things that evil AI would need to have: an objective to cause extinction, control over the systems creating the threat, the ability to convince humans to help, and the ability to survive without humans. If AI didn’t have all these capabilities, then our extinction project would be a bust.
But, like, it’s totally possible to create AI with all these capabilities, even if it’s not on purpose. And there are already folks trying to make AI more autonomous and sneaky. So, there’s a chance that AI could be a real threat.
So, will AI end up wiping us all out? It’s not, like, completely crazy to think it could happen. But, we also found that humans don’t need AI’s help to destroy ourselves. One way to lower the risk of extinction, whether from AI or not, is to reduce nuclear weapons, limit harmful chemicals, and improve pandemic monitoring. It also makes sense to invest in AI safety research, ‘cause it could help prevent other AI-related problems that are less serious but still risky.
So, like, the bottom line is that AI could maybe be a threat, but shutting it all down isn’t the answer. We care a lot about the benefits of AI, so it’s not worth giving up on those benefits just to avoid a potential but uncertain catastrophe like human extinction.
