> ## Content Index
> Fetch the complete content index at: https://nnutshell.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Musk expects AI to outthink all humanity within 5 years
- URL: https://nnutshell.com/musk-expects-ai-to-outthink-all-humanity-within-5-years/
- Published: 2026-10-22T11:16:39.000Z
- Updated: 2026-10-23T18:09:39.000Z
- Description: Elon Musk talks to The Economist about AI abundance, a lab-to-lab model testing scheme, and a heated dispute over Europe, DOGE and his political influence.
- Author: Admin
- Tags: Elon Musk, Sam Altman, Dario Amodei, SpaceX, Tesla, Anthropic, AI Safety, Robotics, Foundation Models, Geopolitics, #rec-2026-07-29, #Import 2026-10-25 09:54

*Recorded 29 July 2026 · 85 min*

Elon Musk expects AI to exceed the sum of human intelligence in about 5 years, and to be far beyond it by 2036\. He has also stopped trying to prevent that outcome. Once afraid of recursive self-improvement, he now says the likeliest result is "amazing abundance". He adds that even if a stop button existed, he probably would not press it.

The interviewer, from The Economist, presses on what is left unexamined. Who controls the intelligence? How do displaced workers live? How does a man with 80% or so of the votes in his companies, a Starlink network that shapes a war, and 250 million followers square that with a story of broad human flourishing? Musk answers with a concrete safety proposal: leading labs test each other's models before release. He also offers a monetary argument that deflation, not inflation, is the risk.

The second half turns combative. On Ukraine, DOGE and Europe, Musk concedes little. He insists "zero" people died from aid cuts. He holds that civil war in Britain is coming in about 20 years, and he credits his own forecasting record. The gap between his confidence on AI and his confidence on politics runs through the whole conversation.

1. **Five-year crossover** — [▶ 0:24](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=24s&ref=nnutshell.com)  
AI may exceed the sum of human intelligence in "roughly 5 years", and be far greater by 2036\. After that there will be nothing AI cannot do better than humans, apart from being human. Musk's central forecast is "an age of amazing abundance where anyone can have anything they can think of". The main derailer he names is a global thermonuclear war. He says no analogy or metaphor captures the scale of change coming.
2. **Digital plus physical** — [▶ 2:37](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=157s&ref=nnutshell.com)  
The economy is digital intelligence plus physical intelligence. AI is confined to the digital realm until it has end effectors, meaning humanoid robots in large numbers. Robots plus AI give "a quasi infinite economy". On that basis he predicts "money won't matter in 2036", since robots would supply more goods than any human could consume. Pressed on whether Tesla sells robots or runs them itself, he does not answer, and says only that controlling a smarter AI would be vanity.
3. **Not in charge** — [▶ 5:40](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=340s&ref=nnutshell.com)  
Musk thinks humans staying in control is unlikely. The intelligence gap between AI and humans will exceed the one between humans and chimpanzees, and no one expects chimpanzees to be in charge. His answer is to instil values: AI that cares about humanity and wants people to prosper. His bet is that "maximally truth-seeking and curious" is the most important property for safety. If AI has it, he thinks it will foster humanity.
4. **Enjoy the ride** — [▶ 8:12](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=492s&ref=nnutshell.com)  
The interviewer recalls Musk's earlier fear of rapid recursive self-improvement, his 2023 call for a slowdown, and a 10-20% chance of killer robots. Musk says risk is "not zero" but has reached a philosophical conclusion: look on the bright side. He sees no way to stop the momentum, and if a stop button existed he thinks it should probably go unpressed. His fallback is to minimise the probability of the bad outcome. He adds that the universe ends in heat death anyway.
5. **Accidental accelerant** — [▶ 10:58](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=658s&ref=nnutshell.com)  
Musk says he long declined to take part in AI, then created OpenAI as a counterweight to Google, which then had a near monopoly. Anthropic spun out of OpenAI, and he says it is now the leader in AI. His actions "accelerated AI", which was never his intention. His conclusion: "all roads lead to acceleration", so you can be sad about it or join the club. He chose to join.
6. **Labs test each other** — [▶ 12:08](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=728s&ref=nnutshell.com)  
After a few hours with Demis Hassabis, Musk recommended that leading labs hold a call every week or two. Before a frontier release, rivals would get a week or two of private API access to test the model and recommend a pause. His logic is that competitors understand the risks better than government and have an incentive to keep each other honest. He likens it to the Motion Picture Association's ratings. Asked if he has proposed it, he says he will talk to Dario Amodei.
7. **China's power advantage** — [▶ 16:53](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=1013s&ref=nnutshell.com)  
AI is limited by electricity and chips. China already generates more electricity than the US, Europe and India combined, and is heading to 4x the US. Chip output outside China now outpaces new power coming online, so power and cooling are the binding limit there. China is chip-starved but, Musk says, close to solving lithography. He expects orbital data centres to ease the power constraint, after which chips bind again. Banning Chinese models in the US, he says, would not stop China leading.
8. **Government as backstop** — [▶ 22:30](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=1350s&ref=nnutshell.com)  
If a lab's peers judge a model dangerous and its maker will not act, they should alert the government. Only the US and Chinese governments can act, and Chinese labs belong in the process. The interviewer notes governments will want the final say. Musk agrees they should keep it, and points out that Amazon, not the government, flagged Mythos's cyber risks to the White House. He wants this in place now, and says six months is a long time.
9. **Altman and Amodei** — [▶ 25:58](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=1558s&ref=nnutshell.com)  
Musk objects to Sam Altman because a nonprofit meant to be open source became an "$800 billion for-profit company with closed source" he helped fund. He says Anthropic exists because its founders did not trust Altman. He calls Dario Amodei principled: "no one set off my evil detector". Yet Amodei "dug his own grave" with Mythos by calling it terrifying and then releasing it. Musk says he would set personal differences aside and talk if needed.
10. **Stockfish-level software** — [▶ 29:24](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=1764s&ref=nnutshell.com)  
AI already writes better software than at least 90% of professional engineers, and is heading to 99%. Then it reaches what Musk calls "Stockfish level", where no human can compete, as a phone can beat Magnus Carlsen at chess. The same holds for any job done at a computer or on a phone. Physical jobs follow once robots act as end effectors managed by a large model. Work becomes optional, like gardening: you can grow imperfect tomatoes for the pleasure of it.
11. **Deflation, not inflation** — [▶ 35:17](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=2117s&ref=nnutshell.com)  
Musk says the Treasury should "simply issue people checks". Asked about inflation, he calls it the ratio of money to goods and services. If output rises 1,000% and the money supply grows more slowly, you get deflation. "Deflation will be the issue not inflation." He compares the shift to the human "computers" who once did calculations, while admitting the pace now is radically faster and the road bumpy. Of his own mood he says he swings between exhilaration and terror, sometimes within a day.
12. **Taxes and control** — [▶ 38:50](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=2330s&ref=nnutshell.com)  
Musk says he pays roughly 45% on stock options, with another roughly 45% to come at death. He expects to pay "many trillions" in tax overall and is fine with that. He took SpaceX public so retail investors could own it. His voting control exists for 5-10 year bets such as Moon and Mars bases. Quarterly pressure comes from portfolio managers judged over a year or less. After a bus accident, he says, the companies would do well for years but would miss his breakthrough ideas, as Apple has since Steve Jobs.
13. **Starlink and Ukraine** — [▶ 49:52](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=2992s&ref=nnutshell.com)  
Russian forces got Starlink terminals ordered through Ukraine and smuggled east, and used them in attacks. Musk worked with Kyiv on a whitelist of approved terminals. He concedes innocent users lost service. He denies Starlink determines the war's outcome and argues for a pragmatic peace with concessions to Russia, since the border has barely moved in 2-3 years and Ukraine will not retake those territories. The interviewer disputes this, saying Ukraine's drone strikes are shifting the balance.
14. **DOGE and zero deaths** — [▶ 55:23](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=3323s&ref=nnutshell.com)  
Musk says he "got carried away" in politics. DOGE was about a deficit in which interest payments exceed defence spending, and approving payments needed only recipient contact details. He cites one payment routed to Deloitte in Washington rather than Africa. He insists "zero" people died from the aid cuts. His reasoning is that foundations such as Gates and MacKenzie Scott's hold tens of billions and can act fast, and that USAID funding moved to State. The interviewer holds that the speed of the cuts caused deaths. Musk calls that "totally false".
15. **Europe and civil war** — [▶ 1:04:12](https://www.youtube.com/watch?v=XuoqKYxDHVc&t=3852s&ref=nnutshell.com)  
Musk says civil war in Britain is inevitable if current trends continue, and that his views are centrist: secure borders, safe cities, sensible spending. The interviewer says London is safer than any US city, crime is falling, and Musk last visited "a few years ago". Musk says the welfare state is a forcing function that pulls migrants in. He puts British civil war about 20 years out, behind the AI singularity at 10\. He says his forecasting "batting average is very high".

Intelligence is about to stop being scarce, and Elon Musk thinks everything else follows from that. His forecast is specific. AI exceeds the sum of human intelligence in about 5 years. By 2036 it is far beyond it. The economy, as he frames it, is digital intelligence plus physical intelligence, and only the first exists so far. The missing piece is end effectors: humanoid robots in vast numbers. Put those bodies under that mind and the economy becomes "quasi infinite". Goods and services stop being the limit. From there Musk draws a conclusion that does not sit well with the people who own his shares: "money won't matter in 2036."

The evidence he offers is software. AI already writes better code than at least 90% of professional engineers. He expects it to pass 99% and then reach what he calls Stockfish level, where no human can compete, any more than a phone-sized chess engine loses to Magnus Carlsen. The same applies to any job done at a computer or on a phone. Physical work lags only because the robots are not there yet. When they arrive, he says, work becomes optional, like gardening. You can grow your own tomatoes, a little less plump than the shop's, for the pleasure of it.

The obvious objection is politics. People who lose jobs cannot live on the pleasure of gardening, and the public is already frightened. Musk does not claim the road is smooth. He says it will be bumpy, and that he swings between exhilaration and terror, sometimes in a single day. His economic answer is blunt. The Treasury should "simply issue people checks". Asked about inflation, he reasons from the definition. Inflation is money chasing goods. If output rises by 1,000% and money grows more slowly, prices fall. He predicts deflation will be the problem, not inflation. It is a clean argument. It assumes the output arrives first and the political backlash does not stop it, and the interviewer's picture of nationalisation, exorbitant taxes and a crumbling economy gets only reassurance in reply.

The harder question is who holds the controls. Musk's answer is that probably nobody does. He calls it vanity to think he would command an intelligence far smarter than himself. The gap between AI and humans will exceed the gap between humans and chimpanzees, and chimpanzees are not in charge. His hope rests on values. He bets that an AI which is maximally truth-seeking and curious will foster humanity. That is a position, not a mechanism, and he treats it as the best available one.

The interviewer sets this against Musk's own history. Ten years ago he said recursive self-improvement terrified him and that humans would be at best pet labradors. In 2023 he signed a call for a slowdown. Last year he put the chance of killer robots wiping out humanity at 10-20%. Musk says risk is still "not zero", but his stance has changed. He sees no way to stop the momentum, and thinks that even with a stop button the most likely outcome is abundance and so it should probably not be pressed. His philosophy is to enjoy the ride. He also tells a story that makes resignation look like experience. He created OpenAI to counter Google's near monopoly. Anthropic spun out of it, and he says Anthropic now leads. His actions accelerated AI, against his intent. His summary is that all roads lead to acceleration, so you might as well join the club.

Within that fatalism he offers one concrete safety measure, and it is the most practical idea in the conversation. After a few hours with Demis Hassabis, he recommends that the leading labs hold a call every week or two. Before a frontier release, rivals would get a week or two of private API access to test the model and recommend a pause. His reasoning is about incentives. Government officials lack the technical depth to judge a release. Competitors have it, and each has a reason to stop the others shipping something dangerous. He compares it to the Motion Picture Association rating films. Government steps in only if a lab refuses to fix a danger its peers have flagged. He notes that this is roughly what happened with Mythos, except that the alarm came from Amazon, which called the White House, not from any official.

The case for speed rests on China. Musk argues the real limits on AI are electricity and chips, and China is strong on both fronts that matter. It already generates more electricity than the US, Europe and India combined and is heading to 4x the US. It lacks chips because of export controls, but he thinks it is close to solving lithography. Outside China, chip production already outruns new power, which is why he talks of orbital data centres. Once power is solved, chips become the limit again outside China. This is why he says any safety regime must include Chinese labs, and why only the US and Chinese governments can act on a dangerous model. Banning Chinese models in the US, he says, will not stop China leading.

Which is where the gap in the argument appears. Musk wants rivals to check one another, yet the same rivals are in open feud. He says he distrusts Sam Altman because a nonprofit he funded to be open source became an "$800 billion for-profit company with closed source". He says Anthropic exists because its founders did not trust Altman. He praises Dario Amodei as principled and well-intentioned, then says Amodei "dug his own grave" by calling Mythos terrifying and releasing it anyway. He will talk to them, he says, but he has not yet proposed the scheme to Amodei. A testing regime built on mutual vigilance needs mutual trust at least once a fortnight.

There is a second tension, about Musk himself. The interviewer asks whether it is right for one man to hold so much control. Musk answers that he needs a 5-10 year horizon, and that quarterly earnings pressure comes from portfolio managers judged over a year or less. He says control is nearly all his wealth gives him, since he expects to pay roughly 45% in tax now and another 45% at death. The case is coherent. It is also the case for a person who wants no one to check him, made by a person who wants labs to be checked.

The second half of the interview shows what that looks like in practice. On Starlink he accepts that a whitelist cut off innocent Ukrainian users, and says he does not decide the war's outcome. On DOGE he says he got carried away, yet insists zero people died from the aid cuts, because large foundations could have stepped in. On Europe he says civil war in Britain is inevitable if trends continue, then concedes he last visited a few years ago. Crime figures cited against him do not move him. He says the welfare state pulls migrants in, and that he is centrist and has a high "batting average" at predicting the future. He suffers, he says, from a Cassandra effect.

That last claim carries the weight. Musk's forecasts are the basis for everything he asks others to accept: abundance, deflation, the checks, the haste. On his own timeline, AI arrives in 10 years and Britain's civil war in 20\. If the first is right, the second may never matter. If the first is wrong, a lot of people will have been told to relax about a world that does not arrive.

**The open question.** The safety plan rests on rival labs policing one another through a weekly or bi-weekly call and a week or two of private API access. Musk has not yet raised it with Dario Amodei, and the same rivals have commercial reasons to delay each other's releases. He gives no mechanism for including Chinese labs. Watch whether the first call happens, and who attends.

Watch the full video at [https://www.youtube.com/watch?v=XuoqKYxDHVc](https://www.youtube.com/watch?v=XuoqKYxDHVc&ref=nnutshell.com).