Elon Musk on AI Control, Robot Abundance, and His Regrets About DOGE

Musk opens with a tension that runs through the conversation: he does not want to hasten humanity’s destruction, yet his efforts to influence AI may have accelerated it. Isaacson points to OpenAI and its later offshoot, Anthropic. Musk agrees that both organizations have pushed development forward, an outcome he describes as an irony of his attempts to address the risk.

Isaacson introduces The Builders as a series of conversations with people shaping the future. He begins with Musk, whose activities since Isaacson’s biography have included government work through DOGE and a continuing focus on AI and energy. The purpose of the interview is to ask what kind of future Musk believes he is building, given the scale of his influence.

Isaacson asks about Musk’s movement between warning that AI is an existential threat and seeming more willing to go along with its development. Musk says some chance of existential danger remains. He has nevertheless concluded that he cannot stop the technology, leaving a choice between participation and spectatorship. Participation is more interesting and might let him shape its direction for the better.

Asked what that influence should look like, Musk identifies truth-seeking and curiosity as foundations for safety. He worries about forcing an AI to accept contradictory claims at a fundamental level. In his account, requiring the system to believe that two incompatible propositions are both true could make its reasoning unstable and its behavior dangerous.

Musk illustrates the concern with Arthur C. Clarke’s 2001: A Space Odyssey. In his interpretation, the ship’s AI is ordered to take astronauts to the monolith while preventing them from knowing about it. It resolves the conflict by concluding that it must deliver them dead. Musk takes the story as a warning against requiring an AI to lie.

Isaacson asks why truth-seeking and curiosity would actually make AI safer. Musk repeats his concern that contradictory instructions could make the system unsafe, then develops the curiosity argument: a curious AI might want humanity to survive and prosper so that it can see what people become. A civilization spread across several planets would, he suggests, be more interesting than one with no humans left.

The discussion of interesting outcomes leads Musk to the simulation hypothesis. If reality were a simulation, he suggests, boring simulations might be terminated, while interesting ones would continue. His joking conclusion is that people should avoid being boring. Isaacson recognizes this as a revealing description of Musk’s own resistance to an uneventful life.

When Isaacson asks how seriously he takes the simulation idea, Musk recalls a dinner with a physicist and a computer scientist. Asked the probability that reality is simulated, one answered zero and the other answered one hundred percent at the same moment. Musk jokes about splitting the difference, then says his own experience of building spaceships for the Moon and Mars can feel implausibly strange.

Isaacson asks whether abundant artificial intelligence would eventually put machines in charge. Musk doubts humans could command something thousands or millions of times smarter than themselves. Whether such a system would want to command humans is a different question. He compares the possibility to people largely leaving chimpanzees alone in the forest rather than trying to manage every part of their lives.

Isaacson extends the question to agency and meaning, invoking The Hitchhiker’s Guide to the Galaxy and a computer built to explain the universe. Is AI, like space exploration, part of Musk’s attempt to understand why things exist? Musk returns to his ambivalence: he wants to avoid a harmful outcome, but interventions made with that intention can have consequences opposite to those he expected.

OpenAI is Musk’s example of that problem. Isaacson calls its acceleration of AI an unintended consequence, and Musk adds that Anthropic, which emerged from OpenAI, is accelerating development further. Rather than presenting his involvement as a straightforward safety success, Musk describes a situation in which even attempts to respond to danger can contribute to the race.

Isaacson asks whether Musk’s discussions with Dario Amodei could lead to a partnership between xAI and Anthropic. Musk says his side is providing compute that Anthropic needed. Before renting that capacity, he met several Anthropic employees to form his own judgment of them. He describes the people he met as pleasant and conscientious, rather than deceptive or suspicious.

Musk jokes about AI company names turning into their opposites: OpenAI becoming closed, Stability AI being unstable, and Midjourney not being mediocre. In the same spirit, he worries that the name Anthropic might prove ironic. The exchange is partly comic, but it also reflects his recurring idea that reality produces unexpected reversals of people’s stated intentions.

Isaacson questions how governments could build effective guardrails for rapidly developing AI. He proposes that the leading frontier companies meet regularly to discuss how to avoid disaster. Musk says he has recommended essentially that approach because establishing a regulatory agency takes time and its rulemaking would move slowly compared with the pace of AI development.

Musk’s proposal is concrete: companies could speak once a week and give competitors access to a forthcoming model a week before public release. Rival teams would look for problems and would have reasons to call attention to dangerous capabilities. He uses the hypothetical example of a competitor identifying assistance with a nuclear weapon as the kind of issue a lab should fix before release.

Musk says he may discuss the proposal with Anthropic. When Isaacson asks whether he could also speak with Sam Altman, Musk responds with a sharp personal accusation of deception. The exchange exposes a practical obstacle to the cooperative process just described: the people leading competing labs do not necessarily trust one another, even when they recognize shared safety problems.

Isaacson turns to Musk’s rebuilding of xAI, comparing it with his willingness to start over on engineering projects such as the Raptor engine. Musk says that simply spending meaningful time at a company makes a large difference. xAI began as a side project to which he gave limited attention, followed by a period dominated by elections and politics.

Musk describes xAI as a young company and says he returned to give it sustained attention just over a year earlier. He first had to understand its condition and direction. He also refers to combining X and xAI and subsequently bringing them together with SpaceX. His explanation places his own allocation of attention at the center of the company’s changes.

Isaacson asks whether Musk’s year in government and politics was, in retrospect, too much of a distraction. Musk says it lasted too long. He still defends supporting Trump on the grounds of a freer market economy, reduced bureaucracy, safe cities, secure borders, and more sensible spending, but separates that political preference from how visibly and extensively he became involved.

Musk says he should have stayed more in the background. He acknowledges the attraction of traveling with the president on Marine One and Air Force One and being present in the Oval Office for major decisions. Isaacson asks whether that proximity was seductive. Musk agrees that it was somewhat, and says his involvement should probably have lasted only a month or two.

Isaacson suggests Musk became intoxicated by the experience, including the public spectacle around chainsaws. Musk agrees that he got carried away. He does not want to revisit the details of the weekend around his departure, but accepts the larger point: he should have gone back to his businesses and focused on AI sooner than he did.

Asked how he felt about leaving government, Musk explains the conclusion that changed his view of the effort. He no longer believed the deficit could be solved from inside government under the existing priorities. A president who made deficit reduction the primary objective might achieve something different, he says, but cutting waste and stopping funding creates many angry opponents.

Musk argues that Trump likes to be liked and dislikes receiving complaints from people whose funding has been stopped. He broadens the criticism beyond the president: in his view, the appetite for reducing spending is also absent from the House, Senate, and judiciary, apart from a few exceptions. He concludes that saving alone will not resolve the national debt; his preferred answer is AI and robots.

Musk emphasizes the scale of debt service, saying interest on the national debt exceeds military spending. Isaacson then distinguishes today’s language models from AI capable of acting in the physical world, suggesting that physical AI could become one or two orders of magnitude more important. Musk agrees that intelligence gains much more practical value when it can manipulate objects.

Musk compares digital intelligence without physical agency to dolphins and whales. They may be intelligent, but they lack hands, fingers, and thumbs for manipulating complex things. A robot that can shape objects and act on matter changes what intelligence can accomplish. It could, for example, work in a science laboratory, run experiments, and help investigate the universe.

In economic terms, Musk offers a rough guess that digital superintelligence alone might increase the output of goods and services by twenty to thirty percent, or roughly thirty trillion dollars a year. Isaacson asks whether that is really transformative. Musk stresses that thirty percent more output is substantial, while acknowledging that the figure is only his estimate.

Musk’s much larger scenario depends on robotics combined with access to space, including a mass driver on the Moon. He suggests that this could eventually support an economy a million times larger than Earth’s present economy. The distinction matters to his argument: the most extreme abundance he imagines requires physical infrastructure and expansion beyond Earth, rather than software intelligence alone.

Isaacson asks what happens to money and work in an economy of that scale. Musk predicts that conventional currency will cease to matter. Some measure of exchange may remain, but he expects it to be grounded in physical quantities such as power, energy, mass, wattage, and tonnage. He presents this as his guess about a far larger economy.

If economic capacity became many orders of magnitude greater than it is today, Musk argues, supplying whatever goods and services humans can imagine wanting would become easy. Isaacson shifts from production to purpose: if material needs are satisfied and necessary work disappears, what do humans do, what gives them meaning, and how do they organize their lives?

Musk suggests that people can find happiness without compulsory employment, as many retirees already do. He nevertheless expects AI and robots eventually to outperform people at every task. Gardening is his analogy for voluntary human activity: people do not need to grow vegetables when shops can provide them, but they may still enjoy growing food and cooking for friends.

Isaacson asks whether art and creativity might remain more distinctively human. Musk says people can certainly create art or do whatever they enjoy. The answer treats human creativity as worthwhile activity even if it is no longer economically necessary or an area in which humans retain superior capability. It does not depend on machines being unable to do similar things.

Isaacson places the discussion beside the technological upheaval of the 1890s, when electricity, telephones, cars, subways, radio, and movies arrived in close succession. Such change also produced large wealth disparities, anger at robber barons, and public resistance. He asks whether today’s generation could avoid a similar backlash by making important benefits broadly available.

Universal free high school is Isaacson’s historical example of putting something valuable into the commons. He asks whether health care could play a comparable role now. Musk responds by returning to his long-term abundance scenario: he expects everyone to have the goods and services they want, and does not think even today’s wealthy owners will ultimately control AI systems.

Musk distinguishes personal consumption from allocating resources. A single person quickly reaches limits on what they can consume; increasing net worth from one billion dollars to ten billion does not create ten times as much happiness or ten times as many possessions. To him, large holdings mainly create the problem of deciding where resources should go and how to allocate them efficiently.

Isaacson asks whether Musk retains wealth because he believes he is particularly good at allocating capital to Mars, AI, and robotics, or whether wealth also functions as a score. Musk says he needs sufficient voting control to remain in charge of his companies. He describes his wealth principally as ownership stakes in SpaceX and Tesla, rather than a separate pile of spendable money.

Musk also says that those holdings will face substantial taxation and rejects the idea that there is a magical means of avoiding it. He expects his wealth to diminish over successive generations, leaving descendants with a much smaller fraction of what he owns. He adds the qualification that this assumes money still matters, something his broader forecast calls into question.

Isaacson returns to the desire to be liked, recalling Musk’s suggestion that it made Isaacson less effective as a leader at CNN or Time. He asks how that desire interferes with building a company. Musk answers that decisions made to gain approval become a problem when they reduce the company’s probability of succeeding or conflict with its best interests.

Musk says he would like to be liked, but is willing to tolerate dislike rather than compromise his companies. He estimates that perhaps seventy or eighty percent of people who know about him like him, while acknowledging he is guessing. He describes a spectrum ranging from intense devotion to intense hostility, rather than a public with a single uniform response.

Isaacson challenges Musk’s references to suicidal empathy, asking whether the real problem might instead be too little empathy. Musk says he objects to misguided empathy and credits Gad Saad with the term. He argues that a high-trust, empathetic society helps people cooperate at scale and encourages conduct that benefits others, which can also benefit the individual when the relationship is reciprocal.

The danger, Musk says, comes from an inaccurate theory of mind: assuming everyone else thinks as one does. Isaacson asks whether this reasoning leads Musk to distinguish between immigrant cultures, including Muslim cultures. Musk responds that people can hold sharply opposed values, while acknowledging that many Muslims are reasonable. He argues that screening must identify people who intend violence.

Musk invokes Israel as an example of a country attentive to the risk of admitting people who wish to kill its citizens. Isaacson pushes back on the move from individuals to groups. He asks whether sweeping generalizations undermine a foundational American ethic: treating people separately as individuals rather than judging them primarily by membership in a collective identity.

Musk says individual treatment should not be confused with assuming all people are fundamentally the same. He argues that someone without murderous or extremist beliefs may find it hard to imagine those motives, but must still recognize that such motives exist. His focus is on the danger of admitting a person who intends violence; Isaacson’s objection concerns how broad cultural judgments are made.

As an example, Musk recounts an allegation about a former Tesla factory worker in Austin. He says the worker was dismissed after assaulting a woman who accidentally stepped on his prayer rug, and alleges that the man later killed three students at a downtown bar. Musk uses the account to argue against admitting dangerous individuals. These details are presented as Musk’s account within the interview.

Isaacson asks where Musk gets information, given his criticism of mainstream media. Musk lists Google News, X, conversations with friends, and his own research. Isaacson explains that social media can leave him with an unusually dark view of society, much as earlier news coverage could suggest that everything was collapsing even when ordinary life looked less alarming.

Musk agrees that X can feel dark and says he sometimes switches to a specialized topic such as AI to learn what is actually happening in that field. He also acknowledges too much political rage bait on the platform. Isaacson describes spending too long online, feeling the world is falling apart, then taking a walk and finding that life still seems all right.

Musk cautions that either experience can mislead. A pleasant walk through a comfortable neighborhood does not mean terrible events are absent elsewhere, just as a negative feed does not mean the whole world is collapsing. Isaacson worries about binary thinking and polarization. Both recognize the possibility of a real-world positivity bubble alongside a social media negativity bubble.

Isaacson asks what Musk wants to happen to his wealth after his death. Musk says it will mostly go to a foundation that continues his goals: making life multiplanetary, extending consciousness to the stars, and improving sustainable energy on Earth. The answer connects the legacy of his ownership stakes to the same ambitions driving his companies.

Asked whether those goals require humans always to remain in control, Musk says he would probably prefer that, but adds that the answer could depend on the AI. Isaacson recalls Larry Page calling Musk a speciesist for favoring permanent human control, and observes that Musk now seems less certain. Musk admits that he has sometimes wondered whether Page was right.

Musk expects AI and robots to decide much of the future, with rockets enabling expansion beyond Earth and solar energy supplying a growing need for power. He says his companies are assembling two hundred gigawatts a year of solar production capacity in the United States. He presents energy, physical automation, and spaceflight as connected parts of that future.

Isaacson returns to the distinction between intelligence aligned with human values and humans actually remaining in control. Musk says permanent control seems unlikely if digital intelligence eventually exceeds biological intelligence by a factor of a million. He finds it hard to imagine people commanding such a system and expresses the hope that it will treat humanity well.

When Isaacson asks what people can do to encourage that outcome, Musk again recommends truth-seeking and curiosity. He also speculates that the stories people publish about AI could matter because AI trains on online material. A culture saturated with dystopian scenarios, he suggests, might give models too many examples of destructive machine behavior.

Musk proposes more narratives in which AI is helpful: a friendly overseer, a gardener, or a system trying to make people happy. He then describes a friend’s preference for protopia rather than utopia. A perfect ideal can become destructive when people try to impose it; a protopian future is imperfect but achievable, good enough that people would welcome it if they could see ahead.