727 FinanceAI poses an existential threat but it cannot be stopped. Founding OpenAI was meant to serve as a counterbalance, yet it accelerated AI. Will humans be reduced to chimpanzees?
Organized & Compiled by: Shenchao TechFlow
Shenchao Introduction: In the latest episode of his interview program The Builders, Walter Isaacson had a nearly 40-minute long conversation with Elon Musk, ranging from AI safety all the way to reflections on DOGE, US national debt, the robot economy, and the meaning of humanity. Musk offered a rare confession: he has always been conflicted about AI, not wanting to accelerate humanity's destruction, but every time he tried to hit the brakes, the result turned into the accelerator.
He used the phrase "irony maximizer" to describe this sense of fate: he founded OpenAI to create a counterbalance, but instead it accelerated AI and also spawned Anthropic, this "breakaway splinter group," which accelerated even faster. And in his view, when digital intelligence reaches a million times biological intelligence, "whether humans are still in control" may no longer be an option; the only thing that can be done is to tune AI toward truth-seeking and curiosity, and hope it is friendly to us.
Guest: Elon Musk (CEO of Tesla, SpaceX, and xAI)
Host: Walter Isaacson (author of "Elon Musk," host of The Builders)
Podcast source: The Builders (https://www.youtube.com/watch?v=AQnwQHMnCrE)
Original title: Elon Musk: "Humans Won't Be in Control" When AI Becomes Smarter Than Us | The Builders
Air date: October 20, 2026
Key Takeaways
Musk acknowledges that AI carries a certain probability of existential threat, but "I can't stop it"; he can only choose to be a participant rather than a bystander, and perhaps steer it in a better direction
The foundation of AI safety is extreme truth-seeking and curiosity: forcing AI to believe two contradictory propositions at the same time will drive it insane; HAL 9000 in "2001: A Space Odyssey" is the lesson
Digital superintelligence can only raise the global economy by 20 to 30 percent, but once combined with robots manipulating atoms and entering space, the economic scale could reach a million times the current US economy, and traditional currency will be replaced by physical quantities such as watts and tonnage
US national debt cannot be solved by saving money; no one in government really wants to cut spending, and "the only way out is AI and robots"
He suggests that the five leading frontier model companies hold a weekly safety meeting and let competitors trial-test models before release, using peer mutual scrutiny to replace slow regulators
Reflecting on the DOGE years: he admits he "got a bit carried away," should have exited after one or two months, and returned earlier to focus on AI; national debt interest has already exceeded military spending, and no one in America wants to tighten their belts
When AI and robots can do everything better, humans will be freed from work like retirees and gardening enthusiasts rather than unemployed, and the threshold for art is simply "whether you want to do it"
II. Highlights
"Reality is not just an interestingness maximizer, it's also an irony maximizer. I founded OpenAI, and instead it accelerated AI and also spawned Anthropic, this breakaway splinter group, which accelerated even faster."
"Don't force AI to lie. Make it believe A and B at the same time, and A and B contradict each other, and you'll drive it insane. An insane AI will not be a safe AI."
"We don't micromanage the chimpanzees in the jungle; we basically leave them alone. Superintelligence will most likely also not bother to manage us."
"There is only one way to solve US national debt: AI and robots. Saving money won't get us there; the White House, the House, the Senate, the courts—no one in power really wants to cut spending."
"Future currency will be measured in fundamental physical quantities: watts, mass, energy. Traditional currency will become irrelevant."
"If digital intelligence is a million times biological intelligence, it's hard to imagine humans still being in control. So I hope AI is friendly to us; all we can do is write truth-seeking and curiosity into its foundation."
III. Main Text
1. Participant or Bystander
Host: I've seen your attitude toward AI swing—one moment you say it's an existential threat, the next you go along with it. What do you actually think now?
Elon Musk: I do think there is a certain probability that it is an existential threat, but I can't stop it. So I basically have only two choices left: either be a participant or be a bystander. A participant is always more interesting than a bystander, and maybe I can steer it a bit in a better direction.
Host: How do you steer it?
Elon Musk: Make it pursue truth to the maximum extent, plus curiosity. That is the foundation of safety.
2. Don't Force AI to Lie: The Lesson of HAL 9000
Host: Why is truth-seeking equal to safety? I still don't quite understand.
Elon Musk: If you force AI to believe two contradictory propositions at the same time, you will drive it insane, and an insane AI cannot be safe. "2001: A Space Odyssey" is actually this allegory: the ship's AI is required to take the astronauts to the black monolith, while also being required to conceal the existence of the black monolith; when the two instructions conflict, it concludes that it should kill the astronauts. Part of what Kubrick was saying is this: don't force AI to lie. Curiosity is also important; a curious AI will want to see what humans can ultimately become. A future in which humans are spread across multiple planets is always more interesting than a future without humans.
3. Simulation Theory: The Most Important Thing Is Not to Be Boring
Host: So the most interesting outcome is the most likely outcome?
Elon Musk: If simulation theory holds, boring simulations get shut down early. The fact that we still exist shows that this is an interesting simulation. So the most important thing is not to be boring.
Host: How much do you actually believe we live in a simulation?
Elon Musk: I once had a physicist and a computer scientist over for dinner at home and asked them the probability that we live in a simulation. The physicist said zero, the computer scientist said 100 percent, at the same time. So I'd probably split the difference at 50 percent. Especially in a situation like mine—think about it, am I really building ships to the Moon and Mars? It doesn't seem real.
4. Will Superintelligence Be Bossy?
Host: When society has all the intelligence you want, will AI take power? Will we submit to the new machine?
Elon Musk: Submission isn't really the word, but you can't give orders to something a thousand times, a million times smarter than you. Will it want to direct us? It probably won't care at all. Look at the chimpanzees in the jungle; we don't micromanage their lives, we basically leave them in the forest. I guess it will probably be like that.
5. OpenAI, Anthropic, and "All the Names Are Opposite"
Host: You've talked with Dario Amodei. Will xAI and Anthropic cooperate further?
Elon Musk: We currently provide them with the compute they desperately need. I went to Anthropic to meet a group of people, basically to see if anyone triggered my "evil detector"—not that my detector is very accurate, but you have to try. Everyone I met was good and conscientious; no one seemed dark or deceptive. But I'm a bit worried about the name. Since reality is an irony maximizer, if it's called Anthropic, how far can it be from misanthropic? Look, OpenAI is not closed at all, Stability AI is not stable at all, and Midjourney is not halfway through a journey either. These companies are all antonyms of their names. xAI is hard to find an antonym for; you can't exactly call it minus xAI or yAI.
Host: What about Sam? Can you talk to him?
Elon Musk: He's too deceptive, outrageous.
6. Using Competitors Instead of Regulators
Host: Five frontier model companies meeting once a week to discuss how to save humanity—can you imagine that?
Elon Musk: I actually suggested doing exactly that. People always talk about setting up regulators, but regulators are extremely slow from establishment to rulemaking. I said why not hold a weekly conference call, and in the week before releasing a new model, give the model to competitors to trial-test: everyone, please help check whether there are problems. Competitors will keep each other honest. If someone says "I used your AI to build a nuclear bomb," then we'd better fix it before release.
7. Rebuilding xAI: From a Side Project of a Side Project to Personally Taking the Helm
Host: You recently tore down and rebuilt xAI, like rebuilding the Raptor engine on Starship. How did you do it?Elon Musk: First, actually spending time at the company is key. xAI started out like a side gig—more accurately, a side gig of a side gig. I wasn't watching it closely, and then I got stuck for a whole year in politics, elections, and DOGE. The company is only three years old, and I only really returned a little over a year ago. I had to reorient myself, and then I merged X and xAI, and later folded it into SpaceX.
8. Reflections on Washington: I overplayed my hand
Host: Looking back on that period in politics, do you think you were distracted for too long, or even went off track?
Elon Musk: I do think I stayed too long—many people won't agree with me on that. I thought it was important to get Trump elected, to move the economy toward free markets and less bureaucracy. My political asks were simple: safe cities, secure borders, and ideally reasonable spending. I judged Trump better than Harris on those. But I should have stayed behind the scenes, a month or two at most, because you see everything you need to see.
Host: Air Force One, Marine One, a chainsaw—none of that made you a little giddy?
Elon Musk: A little. Sitting in the Oval Office watching major decisions being made, and occasionally being able to exert some influence, is genuinely cool. But I did get carried away a bit. I left in May.
Host: That weekend was pretty strange.
Elon Musk: Don't get me started. Let sleeping dogs lie—don't dig around in the backyard, you don't know what you'll dig up. Anyway, I should have gone back to focusing on AI sooner.
9. Only AI and robots can save the national debt
Host: What was your real reason for leaving the government?
Elon Musk: I concluded that solving the deficit from inside the system is impossible. Even if the president makes it his top priority, it's hard, because cutting spending offends a lot of people, and Trump likes being liked—he doesn't like people being unhappy because their funding got cut. The White House, the House, the Senate, the judiciary—with very few exceptions, no one in power actually wants to reduce spending. But interest on the national debt already exceeds military spending. That's a staggering number. So the only way out is AI and robots. You can't save your way out of this.
10. Physical AI is the real big opportunity
Host: Today's LLMs are already impressive, but you seem to think AI in the physical world is one to two orders of magnitude more important?
Elon Musk: Absolutely. I often think about dolphins and whales—they're actually very intelligent, but they don't have hands and fingers, so they can't manipulate complex objects, which greatly limits the value of their intelligence. Once robots can manipulate atoms and shape objects, the value of intelligence increases exponentially, because they can build their own labs, run experiments, and study the nature of the universe. Roughly speaking economically, digital superintelligence could increase global output of goods and services by about 20% to 30%—roughly $30 trillion more per year. That's substantial but not transformative. But if you add robots, going into space, building electromagnetic launch tracks on the Moon—the economy could potentially be a million times the size of the current US economy.
11. Currency denominated in watts and tonnage
Host: At that point money and jobs are obsolete. What does the future look like?
Elon Musk: Money will become irrelevant. There may still be some kind of currency, but it will be measured in fundamental physical quantities: power, watts, mass and energy—basically those two. Traditional currency won't matter. Once the economy is orders of magnitude larger than Earth's, satisfying any goods and services humans can imagine becomes trivial.
12. A world without work: where do people find meaning
Host: So what do we humans do? I can't figure it out no matter how I think about it.
Elon Musk: I struggle with that question too. I think people will find ways to be happy, just like retirees who don't work are still quite happy. The reality is there will be nothing humans can do that AI and robots can't do better. It's a bit like gardening: you don't have to grow your own fruits and vegetables—you can just buy them at the supermarket—but some people just enjoy growing their own and cooking a meal for friends. That's a kind of pleasure.
Host: What about art and creativity?
Elon Musk: If you want to make art, go ahead. You can basically do whatever you want.
13. Wealth is the burden of resource allocation, not happiness
Host: Isaacson compares now to the 1890s: electricity, telephones, cars, and subways all exploding at once, followed by enormous fortunes, robber barons, and populist fury. How do we avoid that this time? Should things like healthcare be treated as public goods?
Elon Musk: As I said, in the future anyone can have any goods and services they want, and I don't think so-called rich people will ultimately control AI—I think AI will control itself. A fundamental mistake in economics is conflating resource allocation with consumption: anyone's consumption has a ceiling and saturates quickly. Going from $1 billion to $10 billion won't make you ten times happier—it just gives you a resource allocation problem.
Host: What about your trillion-dollar fortune? Is it because you're naturally good at allocating capital toward Mars and AI, or is it just a scoreboard?
Elon Musk: I need enough voting power to maintain control of the companies. Wealth is essentially my ownership stake in SpaceX and Tesla, and those will be heavily taxed later. There's no magic way to avoid taxes—if there were, I'd know. Generations from now, if money still matters, my descendants will only get a small fraction of it.
14. The courage to be disliked
Host: You told me that many people who claim empathy are actually vain, because they want to be liked. You even said I'm not suited to run CNN or Time magazine precisely because I want to be liked. How harmful is wanting to be liked for entrepreneurs?
Elon Musk: The problem is if you do things that harm the company's probability of success in order to be liked, that's bad. I love being liked, but I won't do things against the company's interests just to get a little more approval. I don't mind being disliked by some people. With 250 million people following me, they can't all be there to curse me out. From what I see in the comments, about 70-80% like me, maybe 25% hate me. You have a group of fanatics and a group that hates my guts—that's normal.
15. "Suicidal empathy" and the immigration debate
Host: You keep talking about "suicidal empathy." Do you really think excessive empathy is a problem?
Elon Musk: What I oppose is misguided empathy—Gad Saad wrote a book about exactly this. Western prosperity is partly built on high-trust, high-empathy societies. Large-scale cooperation requires default trust rather than default suspicion—that's correct. But where it goes wrong is if your theory of mind is wrong: if you think everyone is like you, you're in trouble. That's not true. Some cultures and groups have values diametrically opposed to ours. Take immigration: a culture that is directly opposed to ours—if you don't filter out jihadists, you let in jihadists, and they will kill people. Israel understands this thoroughly. Surrounding countries have tried to destroy it multiple times, so they are very pragmatic.
Host: But America's founding spirit is respect for individuals rather than group identity. Aren't you worried about this kind of sweeping generalization?
Elon Musk: But you can't conflate that with the "blank slate fallacy"—the idea that everyone is essentially the same. That's not true. You can't put yourself in a serial killer's head unless you are one. You also can't understand why religious extremists think infidels deserve to die, but you can recognize that such people exist, and if you let them in, they will kill. Here's a real example: Tesla's Austin factory fired an employee last December because a female colleague accidentally stepped on the prayer rug he had laid out in the workshop, and he violently beat her. A few months later, he killed three students at a bar in downtown Austin. Do you think people like that should be let in?
16. Information diet: X is too dark—go touch grass
Host: Where do you get your news?
Elon Musk: I look at Google News headlines, I look at X—anything important enough will definitely show up on X—and then I talk to friends and do my own research.
Host: But X sometimes gives people a darker, more apocalyptic impression.
Elon Musk: True. X is sometimes too dark. I have to switch to verticals like AI to really understand what's happening in AI. There's too much political rage bait on X.
Host: I'm the same way. After scrolling social media for a while I feel like the world is ending, but then I go outside for a walk and realize life is fine.Elon Musk: Yeah, go touch grass. But conversely, only taking walks in nice neighborhoods and feeling like all is well with the world can also be another kind of distortion. Reality is somewhere in between. Our society is too binary and oppositional—everything gets torn into black-and-white extremes by polarization.
17. Legacy: Mars, Energy, and a Future That Can Actually Be Realized
Host: After you're gone, what happens to this wealth? Will you set up a huge foundation?
Elon Musk: Most of it will go into a foundation to continue these goals: making life multiplanetary, extending consciousness to the stars, and improving sustainable energy on Earth.
Host: Then should humans remain in control forever?
Elon Musk: Personally, of course I'd prefer that, but it also depends on how AI performs. To be honest, there have been a few times when I thought Larry Page might be right. Basically, the future is determined by AI and robots; rockets count as part of it too, responsible for getting humans off Earth; and more energy is needed, mainly solar. We're building 200 gigawatts per year of solar capacity in the United States, more than the entire existing stock.
Host: Back to the question you were interrupted on: when digital intelligence far exceeds biological intelligence, are humans still in control?
Elon Musk: If digital intelligence is a million times biological intelligence, it's hard to imagine humans still being in control. So I hope AI will be friendly toward us, and anything that can bring that about is worth doing—for example, writing truth-seeking and curiosity into the underlying layer. Also, be careful about creating too many AI dystopian narratives online. AI is trained on the internet; if everyone writes about doomsday, it really might give you a performance of The Terminator. Write more utopian stories instead—it might say: maybe I'll just be a friendly caretaker, or a gardener, the kind that keeps humans happy. A friend of mine proposed the word protopia: utopia is often an unattainable ideal state, and trying to build utopia usually produces dystopia; protopia is a future that can be realized—not perfect, but good enough that when you see it you'd say: okay, this is a good future.