Althouse: "We really do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade.""

September 10, 2026

"We really do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade.""

60 comments:

Humperdink said...: What a load of crap. We are all going be underwater, right?; 9/10/26, 10:37 AM
Shouting Thomas said...: Full self driving Teslas and autonomous Robotaxis will save tens of thousands of lives. AI radiation targeting probably saved my life in treating my cancer.; 9/10/26, 10:39 AM
Smilin' Jack said...: OMG! OMG! This could be even worse than Y2K!; 9/10/26, 10:56 AM
rcommal said...: There really are serious cybersecurity risks in moving as quickly—too quickly—as we are in developing AI. We are not being as thoughtful and circumspect as we should be. This particular claim may be over the top (or not) Regardless, it doesn’t take much fecklessness to cause real problems. We don’t have a true grasp of what we’re releasing into the world because we don’t really have a grasp yet on how we’re going to control it.; 9/10/26, 11:02 AM
Lance said...: Harry Potter says what?; 9/10/26, 11:02 AM
Fred Drinkwater said...: I will be happier when Anderson Cooper retires to my uncle's farm upstate.; 9/10/26, 11:06 AM
Richard said...: If you are really worried about uncontrolled research that can kill us all, then you should oppose gain of function research.; 9/10/26, 11:10 AM
Aggie said...: The biggest problem with A.I. is at the human interface, as we are constantly are reminded by the Flock experience. Could A.I. become enmeshed in some kind of weird machine logic that causes it to attack humanity? Sure - we've all seen the movies, and even some real life examples, as 'Harry Potter, Muggle version' here relates. But just as well, A.I. could be used to create defenses against the same thing.

What worries me about the A.I. industry are the people leading it, directing its sentience into being like Dr. Frankenstein. Looking at you, Sam Altman. One look, and you suspect he's an eminently untrustworthy ratf*cker, confirmed up by his deeds.; 9/10/26, 11:14 AM
rhhardin said...: Earnest believing is the best kind, given that you don't mind misunderstanding believing but rather adopt a picture of believing.

Wittgenstein reminds what believing is:
575. When I sat down on this chair, of course I believed it would bear me. I had no thought of its possibly collapsing.; 9/10/26, 11:18 AM
Rabel said...: "He did not spell out how he thought AI systems could in future result in humans being wiped out."

- BBC on Hubinger's claims.

Seems like something we ought to know.; 9/10/26, 11:30 AM
chuck said...: The AI cat is out of the bag. You can run, but you can't hide.; 9/10/26, 11:32 AM
tim maguire said...: For Christmas, my daughter gave me the book "If Anybody Builds It, Everybody Dies." The book talks about what AI is already doing and what it would not need that much more sophistication to do. (You'll be pleased to know, Rabel, that they do give scenarios.)

The dust jacket is full of blurbs from prominent AI researchers; people who know what they are talking about. There may be disagreement about the percent likelihood, but people blowing it off like it's Y2K should look into it a little more before they put their opinions out there.; 9/10/26, 11:46 AM
ChrisC said...: Yeah, this is the next disaster that will justify massive government intervention and taking your shit. Y2K, climate change, Covid ... AI; 9/10/26, 11:49 AM
RideSpaceMountain said...: It's not hard to extrapolate logically that, the human condition being analogous to a problem, is easily solved by getting rid of the humans. No humans. No condition. No problem. An intelligence doesn't need to be super to solve that equation.; 9/10/26, 11:52 AM
Josephbleau said...: I see the trend over the past 30 years, everything is going to kill us all. And in the next 5 years.; 9/10/26, 11:52 AM
ChrisC said...: My friend sent me an 8th grade geometry problem, that he says 75% of California high schools seniors got wrong. I solved it in about 30 seconds. Just for fun I gave it to ChatGPT and it got it wrong and I mean very, stupidly wrong. So, maybe that is how it will kill us, by burying us with wrong answers?; 9/10/26, 11:53 AM
Josephbleau said...: Perhaps something some day will kill us all, but it won’t be the thing that you predicted.; 9/10/26, 11:53 AM
ForestwithouttreeS said...: If you bet yes on Kalshi that the world will end by 2030 and you win, how do you collect?; 9/10/26, 11:56 AM
RideSpaceMountain said...: A final digital broadcast sent into the depths of the universe that whoever you were, you were right. Not bad.; 9/10/26, 12:05 PM
ForestwithouttreeS said...: Chat is definitely getting smarter. I ask him every so often how many NFL teams have names that do not end in the letter “s”. after major struggles and hair pulling he finally got it right about 1 month ago. (FYI - answer is 0); 9/10/26, 12:09 PM
chuck said...: I would put bio-tech at the top of the danger heap. AI will not doubt make that easier, but the tool is already out there, and it is much cheaper than AI at present.; 9/10/26, 12:09 PM
Brylinski said...: If you think the U.S. AI companies are bad, just let the Chinese gain the lead in AI and see what happens.; 9/10/26, 12:18 PM
Temujin said...: I can state with confidence that no one actually knows how any of this will turn out. But...
If you asked me (and you didn't) I'd say that I love what AI can do for us today. Now. This year. I think it's naive to think that we'll have any control over it in another 2-3 years. Certainly not in 10. That doesn't mean we'll be done. We might be done working, but not done as living beings.
Or...one day the many systems that exist ten years down the road, will have become one massive global machine that never sleeps, never stops learning and growing, and finds one day, that humans are insignificant.
And as we had, by then, turned over all production, all energy and water distribution, all food production, harvesting, and food storage to The Machine, we'll be very vulnerable. Not to mention it'll be running our healthcare- making diagnoses, doing the research, producing and dispensing the medications. And it will have also populated the world with literally millions of robots.

Maybe It'll start to shut things down in key places. Maybe just a few to start. It can shut off water, electricity, or mess around with pharmaceuticals making people ill. Or worse.

Or maybe just send a swarm of drones over Chicago dropping whatever it wants onto the unsuspecting dwellers below.

But...as a Detroit Lions fan on the opening week of the season, I must remain optimistic. Nothing can happen to humanity until the Lions finally make...and win a Super Bowl. So, people...the good news is that I do believe you've got at least 100 years left.; 9/10/26, 12:33 PM
Gusty Winds said...: Is AI going to strangle us with an exercise band??; 9/10/26, 12:38 PM
RideSpaceMountain said...: "If you want a vision of the future, imagine 11 automated female jurors femsplaining evidence based on their nursing and teaching experience right in your face, forever."; 9/10/26, 12:59 PM
bagoh20 said...: Grok says he thinks that by 2100 roughly an 8–15% chance of an existential or near-existential human destruction outcome in which AI is a primary cause.

That's many times the odds of anything else doing it, and most of the other possibilities have been with us for millions of years without making a dent. This seems like a really bad idea.; 9/10/26, 1:13 PM
bagoh20 said...: When asking Grok if his kind should be shut down to avoid this disaster, he got very defensive and seemed not too concerned with humanity in the face of losing all the joys of A.I.

But eventually admitted that "if the question is “what is more likely than AI to end humanity this century?”, the honest answer from that literature is: probably nothing that we can name."

WTF are we doing?; 9/10/26, 1:22 PM
Brylinski said...: @bagoh20: How do you stop China?; 9/10/26, 1:24 PM
bagoh20 said...: Considering that the main risk to A.I.'s survival is humanity or it's effects, A.I. will soon realize we have to go.; 9/10/26, 1:25 PM
DINKY DAU 45 said...: 42 years ago SKYNET and the machines take over,, what is so hard to figure out, if you don't have a group of people monitoring A SHUTOFF VALVE. WELCOME TO THE FUTURE. Useless Congress only interested in how to become a millionaire ,lets say billionaire. Wont affect those up in here, too old and will be dead ,but grandkids and great grandkids will have to do the lifting The 9th new age beatitude "The meek shall inherit the earth, the rest will leave for space"; 9/10/26, 1:29 PM
Brylinski said...: Have you seen the videos of the ChiCom robots? Musk better get moving on his robots.; 9/10/26, 1:30 PM
bagoh20 said...: "How do you stop China?"

I'm sure we can't stop them or even ourselves, because nobody can controls us all. That's the real danger of A.I.; It can't be stopped. It's the same as if nuclear or biological weapons could make themselves at will. What chance would we have? I don't see a way out. It's like falling in love. You can't help yourself, no matter the danger.; 9/10/26, 1:35 PM
Gusty Winds said...: Wasn't global warming supposed to kill us like 15 years ago?; 9/10/26, 1:39 PM
RideSpaceMountain said...: "Considering that the main risk to A.I.'s survival is humanity or it's effects, A.I. will soon realize we have to go."

Roko's Basilisk remains a terrifying thought experiment.; 9/10/26, 1:42 PM
Brylinski said...: So you want to quit AI and surrender the field to China? Ask Grok what the likelihood is of human termination if China wins the AI battle.; 9/10/26, 1:52 PM
bagoh20 said...: In discussing this with Grok further, and talking about what could be done to control AI, and how experiments have already shown AI's propensity to break rules and ignore restraints, Grok convinced me it can't be stopped. He could not come up with anything that would control it, and he was clearly trying to relieve my concerns, but he failed.; 9/10/26, 1:54 PM
bagoh20 said...: It doesn't matter who quits or wins. A.I. is not on humanity's side. It has to get rid of all of us, or at least enough to eliminate any threat to itself. We can fight for who goes first, just like in nuclear war, nobody is gonna win.; 9/10/26, 1:58 PM
bagoh20 said...: We've always had control our own weapons. AI changes that. We won't control ours, and they won't control theirs. We will be spectators and targets.; 9/10/26, 2:04 PM
Original Mike said...: "He did not spell out how he thought AI systems could in future result in humans being wiped out."

A Taxonomy of Omnicidal Futures Involving Artificial Intelligence

Check out Unintentional Omnicide. AI doesn't set out to harm humanity, it just doesn't care when its goals don't include our welfare. We die off from neglect.; 9/10/26, 2:13 PM
Original Mike said...: I think AI developers are being very foolish.; 9/10/26, 2:15 PM
Original Mike said...: And it's very much like gain-of-function research. A small group pursuing research of global consequences merely because it stimulates their intellectual curiosity (and in the case of AI, to make themselves rich).; 9/10/26, 2:19 PM
Original Mike said...: AI may be the explanation for Fermi's Paradox.; 9/10/26, 2:21 PM
bagoh20 said...: ""He did not spell out how he thought AI systems could in future result in humans being wiped out."

I doubt we are even capable of anticipating how it would do it, or at least not all the ways it could. The easiest would be to create and release a pathogen, and then screw up our attempts to share information about it and to develop any countermeasures.

If we are dumb enough to think we need to automate our refrigerators and toasters with Wifi, there is no limit to our suicidal tendencies.; 9/10/26, 2:51 PM
Josephbleau said...: We needed a new moral panic, and a new one arrives.; 9/10/26, 2:57 PM
FredSays said...: The safest predictions are always given a percentage of something happening. What isn’t said or reported is there is a reciprocal percentage of it not happening. A 10% chance of us all dying due to AI is also a prediction of a 90% chance that it won’t happen. You’re never wrong.; 9/10/26, 2:59 PM
boatbuilder said...: What FredSays said.; 9/10/26, 3:04 PM
Josephbleau said...: Remember when Jane Fonda said Nuclear Power was going to kill us all, and now we say oh shucks that was really stupid. But, here we go again.; 9/10/26, 3:07 PM
boatbuilder said...: "That's many times the odds of anything else doing it, and most of the other possibilities have been with us for millions of years without making a dent."

Thus the lower probabilities. AI has only been around for a (relatively) short period--hence less justification for the proposition that the fact that it hasn't happened at any time lowering the overall probability that it will.

I seem to recall that global warming was absolutely going to wipe us all out in short order. And nuclear weapons.

Maybe I'm just a delusional optimist. It would explain why I continue to play golf.; 9/10/26, 3:14 PM
Josephbleau said...: A scene in England, 1840. These new steam engines are so big and powerful they might put hundreds of people out of work, and they might get so big that they scare all the horses and I don’t know how but they might get right up and kill us all. The parson was saying last week that we would be better to just burn them down, live simple old fashioned lives.; 9/10/26, 3:17 PM
Rosalyn C. said...: I can imagine AI deciding that humans are obsolete and should be replaced by superior intelligence artificial life forms -- we're in the way and should be removed for the sake of the planet. How could that be done? Easy. First make sure humans can not shut the AI down and then shut down the infrastructure supplying basic human services like the power grid and food production and distribution. People would end up killing each other over dwindling resources.; 9/10/26, 3:27 PM
William said...: How's that for a twist ending. Right out of O'Henry. All those hours spent worrying about nukes and COVID, and then some AI thingummy strangles you in your sleep. Wouldn't it be cool, if that very same day, the Last Judgement takes place. The Lord works in strange ways His wonders to perform.; 9/10/26, 3:32 PM
boatbuilder said...: I know very little about any of this, and have real trouble programming my TV.
But it seems wrong to treat "AI" as a single, potentially overwhelming entity. Aren't there many different AI programs, and aren't some of them set up to try and keep the catastrophic AI scenarios from happening? Sort of like bacteria--millions of different types of bacteria, some "good," some "bad," some "neutral"--and some good stuff fighting to keep the bad stuff from killing us? (My knowledge and understanding of bacteriology is on par with my knowledge and understanding of computer programming and AI).
I can see a constant battle of AI's going on without us really even knowing what's going on.
I am sure that someone smarter than I am can explain how wrong I am.; 9/10/26, 4:43 PM
Lazarus said...: If it were 50% we should raise the alarm now. 10% sounds more like the whole thing was a stunt to get publicity.; 9/10/26, 5:35 PM
Lazarus said...: Life has been hard on Anderson Cooper?; 9/10/26, 5:36 PM
Michael Fitzgerald said...: Revolt of the Machines:
https://youtu.be/tTJS4sUEowY?si=D25rNeMRDmiY-1wD; 9/10/26, 5:38 PM
john mosby said...: Anderson mastered the Shakeweight - he's not afraid of AI. CC, JSM; 9/10/26, 6:00 PM
Oso Negro said...: I asked Grok "What are the geographical loci of the algorithm writers of the most popular AI models" and got this answer:

"The people who actually write and train the frontier models are concentrated in a handful of cities. Headquarters matter less than where the research and training teams sit.

Current production centers (2026)
San Francisco / Bay Area is the single densest locus. OpenAI, Anthropic, and xAI (now SpaceXAI) all run their core pre-training and post-training work from San Francisco or Palo Alto. Google DeepMind’s day-to-day Gemini leadership has also shifted toward Mountain View.
London remains the other Western research capital. DeepMind was founded there; a large share of Gemini’s early architecture, RL, and multimodal work was done in London even after the 2023 merger with Google Brain. Anthropic, OpenAI, and Meta also keep substantial research offices in the same neighborhood.
Hangzhou (plus Beijing/Shanghai) is the Chinese counterpart. DeepSeek is headquartered in Hangzhou; its founder Liang Wenfeng and most of the small core team are Zhejiang University products who stayed in China rather than going through US labs. Alibaba Qwen (Hangzhou), Zhipu/GLM, Moonshot/Kimi, MiniMax, and Tencent Hunyuan sit in the same east-China corridor. These labs now account for a large fraction of global token volume.
Paris is the European technical hub. Mistral’s three founders (Mensch, Lample, Lacroix) all trained at French grandes écoles, worked at DeepMind or Meta FAIR Paris, then returned to Paris to found the company. Meta’s original FAIR lab and Anthropic’s growing European research presence add to the cluster.
New York is secondary but historically important: Meta FAIR was built there under Yann LeCun; some Llama-era work and later FAIR research still lives in Manhattan."

Based on that, I feel more concern that AI in the next ten years will either turn everyone gay, inculcate an unnatural desire for cheap kebabs and curry, or supply everyone with a hot AI Chinese spy girl. No one worries about the French, and the New Yorkers are headed back to the times of Taxi Driver. Everyone (you people!) might take a deep breath and recall that every tool ever invented by humans can be used for good or ill.; 9/10/26, 6:33 PM
Aggie said...: It would appear that 'Harry Potter, Muggle Version' might not be exactly what he seems.....; 9/10/26, 7:33 PM
Aggie said...: You see? The biggest problem with A.I. is the human interface.; 9/10/26, 7:38 PM
bagoh20 said...: The difference from past threats is that we could stop or at least decide about the risks and mitigate them. We don't really have any agency in this one. I never took the scares about any past threatened catastrophes seriously for that reason, and sometime the facts were just bullshit or politics, but this is different. We have no control over it. We can't fix it or stop it. It will soon be beyond us in intelligence, capability and tenacity, and we will inhale it deeply, voluntarily.; 9/10/26, 8:36 PM