Translated from the 2026-08-15 version of the original
The full-length interview with Elon Musk
Full interview with Elon Musk · Study notes
Thematic study-group compilation; the right column comes from the follow-up questions, explanations and responses that actually took place in this round
I. Human uniqueness, the value of existing, and the meaning of work
When AI does most comparable things better than humans, what uniqueness is left to a person; and when work is no longer necessary, why would people still choose to do it themselves
What AI surpasses is not one person but the sum of all human intelligence
I would expect artificial intelligence to be far greater than the sum of human intelligence, like far greater
Musk expects the overall intelligence of AI to far exceed the sum of all human intelligence, and to do so by a wide margin
The only thing AI cannot surpass is being human itself
There really won't be anything that AI can't do better than humans, apart from being human
There will be almost nothing that AI does worse than humans, except for the act of being human itself
Human value is pushed from a question of capability to a question of existence
If thinking, creativity, the value of work and professional skill will all eventually be surpassed by AI, a person has to face a new question: if I am no longer needed because I am useful, do I still have value as a person in myself
Digital work will quickly reach Stockfish level
I do think it's going to be a bumpy road because AI will be able to do any job better than any person can do that job
It'll get to what I call Stockfish level
Musk thinks AI will soon do any digital job better than any person, eventually reaching what he calls Stockfish level, where, like a top chess engine outclassing human players, humans can no longer compete in that domain
Work may become a kind of voluntary gardening
I think work is going to be optional
They'll, you know, look a little artisanal
If you go to a friend's house and they make dinner with vegetables from their garden, it's a nice touch
People do not have to grow their own vegetables, because the shop sells plumper, more standardised ones, yet people still choose to grow them; those vegetables may be imperfect and have a handmade feel, but when a friend serves you food from their own garden, it still feels like an extra touch of care
II. Digital intelligence, robots and extreme abundance
How digital intelligence enters the physical world and pushes the capacity to produce goods and services toward what Musk calls the near-infinite
The economy can be seen as digital and physical intelligence
You can really think of the economy as digital and physical intelligence
Musk reinterprets the economy through the capabilities needed to get work done: digital intelligence handles information, design and planning; physical intelligence perceives the real environment and completes actions through a body
Robots are the execution end that turns digital decisions into physical action
It doesn't have end effectors
You need the end effectors in the form of humanoid robotics
You can go from intelligence manifesting itself only digitally to shaping atoms
With AI alone, intelligence stays confined to the digital world; humanoid robots provide hands, feet, arms and other end effectors that let digital intelligence actually change matter
local intelligence is the on-device intelligence on the robot itself
The robot is more or less an end effector. It would have some amount of local intelligence, but be managed by a large AI model
The robot is the physical execution end; it carries some local intelligence on board to complete real-time actions, while more complex understanding, planning and management are handled by a larger AI model
Large numbers of AIs and robots form a quasi-infinite economy
An economy is the production of goods and provision of services
If you have vast numbers of robots with vast amounts of digital intelligence, you have sort of a quasi infinite economy
If large numbers of AIs supply digital intelligence and large numbers of robots turn that intelligence into round-the-clock real-world labour, the total output of goods and services could far exceed finite human demand, forming a near-infinite economy
The most likely outcome is amazing abundance
The most likely outcome is an age of amazing abundance, where anyone can have anything they can think of
Assuming that there's not World War III, I think we're headed for an age of amazing abundance
Musk envisions AI and robots making producible goods and services extremely cheap and available at scale, so ordinary people can obtain them very easily; but this path is conditional on no civilisation-level catastrophe such as a third world war or global thermonuclear war
There is no ready analogy for the scale of this change
There's no analogy or metaphor that I think illustrates the magnitude of change that we're going to experience here
Musk thinks historical analogies such as the industrial revolution, electricity, the internet or the steam engine are not enough to convey the scale of change when AI and robots together restructure economic activity
The AI bottleneck will move between chips and electricity
The current constraint is electricity outside of China. In China, the constraint is chips
Once we address the power constraint with AI data centers in space, then the constraint will once again be chips outside of China
Musk distinguishes the current bottleneck inside and outside China: outside China the main shortage is electricity, cooling and data-centre infrastructure; China, restricted in advanced chips, is short of chips instead. If orbital data centres relieve the power problem, chips will again become the bottleneck outside China
III. AI risk, control and governance
When AI carries serious risk, its development seems unstoppable, and humans may not be able to control a superintelligence, how does Musk think the odds of a bad outcome can still be lowered
The interviewer presses on why Musk's attitude to risk changed
You said there was a 10% to 20% chance of killer robots wiping out humanity
There's risk associated with AI and robots, it's not 0
The interviewer cites what Musk once said on the Ted Cruz podcast, that he put the chance of killer robots wiping out humanity at 10% to 20%, yet now speaks more optimistically about AI abundance; Musk replies that the risk still exists and is not zero
"Cannot stop" and "should not stop" are two different judgments
Even if there were a stop button, we probably shouldn't press it, because the most likely outcome is incredible abundance for all
It's inevitable
Musk not only thinks the momentum of AI and robotics cannot be stopped; he goes further and says that even if a button existed that could stop it, it probably should not be pressed, because the most likely outcome is extreme abundance for everyone
The heat-death answer stresses the journey and plays down the immediate risk
If the heat death of the universe is in fact the outcome for our reality, then it really is all about the journey, because the destination is terrible
Musk says that if the universe will eventually reach heat death through a long cooling, then what really matters is the experience along the way, because the destination everything finally reaches is terrible
Controlling a superintelligence may only be human vanity
It would be an exercise in vanity for me to think that I would be controlling a super-genius AI that is vastly smarter than me
Musk thinks that if he believed he could control a superintelligence far smarter than himself, it would probably be a self-flattering fantasy with no basis in reality
The humans-and-chimpanzees analogy
If the difference in intelligence between AI and humans is vastly greater than the difference in intelligence between humans and chimpanzees, it's hard to imagine that chimpanzees would be in charge
If the intelligence gap between AI and humans is far larger than the gap between humans and chimpanzees, it is hard to imagine humans staying in control, just as chimpanzees cannot be in charge of humans
From worrying about AI to "join the club"
Anthropic spun out of OpenAI
You can just sort of be sad about it or join the club, I suppose
Musk says he helped found OpenAI in the hope of counterbalancing Google, but those moves set off a chain reaction that accelerated OpenAI and later Anthropic, founded by former OpenAI members; since every road seems to lead to AI acceleration, he can keep being sad about it or join the race
A publicly released model is not necessarily a lab's strongest current capability
Anthropic had Mythos, from which Fable is derived
They almost certainly have—not almost certainly, they certainly have—it right now
I think they could release it at any time
Musk thinks Anthropic already had Mythos, from which Fable was later formed, ready in February, so it must now have an even stronger next-generation model that could be released at any time
Competing companies test each other's models before release
I would recommend that we at least have some sort of informal weekly or biweekly call
There's maybe a week or two of early access by competitors
Musk suggests the leading AI companies hold informal weekly or biweekly calls, and that before an important new model is released publicly, competitors get one or two weeks of restricted test access through a private API to check for safety risks and, if necessary, recommend pausing the release
Industry finds the risks, government keeps the final power to intervene
I think they should have that power. But it wasn't as though someone in the government figured out that Mythos had some scary cybersecurity risks. It was Amazon
Musk thinks government should have the power to pause or restrict the release of dangerous models, but cannot be expected to discover frontier technical risks on its own; he uses the example of Amazon finding and reporting Mythos's cybersecurity problems to show that risk discovery usually comes from inside the industry
Will competition and personal animosity keep cooperation from ever starting
You don't all like each other very much, but you're also competing, right? You all want to be first
Wouldn't Sam Altman just think you're going to nick all his ideas if you did that?
When there seems to be so much personal vitriol between some of these leaders?
The interviewer presses repeatedly: the heads of the top AI companies are few and know each other, yet they have personal conflicts and all want to be first to release the strongest model; if competitors got early testing access, would Sam Altman not suspect Musk of wanting to steal his technology and product ideas
Good intentions can support cooperation but cannot replace safety checks
No one has set off my evil detector
They always seem like they're well-intentioned
There are a few well-intentioned paving stones in there, so we don't want to be complacent
Musk says none of the Anthropic people he has met has set off his "evil detector"; they seem well-intentioned. But even if bad outcomes come mostly from malice, there are still a few stones paved with good intentions, so trusting someone's motives is no reason to relax vigilance
Index of English vocabulary and expressions
Only the words that were actually asked about and explained in this round's reading are collected below
| Word | Meaning in this passage |
|---|---|
superlatives | The highest, most exaggerated words of praise |
prosaic | More down to earth, more concrete, less grand |
abundance | Extremely plentiful supply and prosperity |
thermonuclear war | War fought with thermonuclear weapons |
derail | To knock development off course or cut it short entirely |
quasi-infinite | Near-infinite, but not infinite in the mathematical sense |
magnitude | Scale, degree, order of magnitude |
an exercise in vanity | An idea born of conceit and lacking any basis in reality |
wipe out humanity | To eliminate humanity entirely |
inevitable | Unavoidable, cannot be stopped |
heat death | The heat death of the universe |
spin out of | To split off from the original organisation and become an independent entity |
join the club | To join the camp already doing the same thing |
nick | British slang: to steal, to pinch |
constraint | The bottleneck that limits a system's further growth |
come online | A facility is formally built, connected and starts being used |
orbital data centers | Data centres deployed in Earth orbit |
personal vitriol | Sharp, malicious and intense personal hostility |
set off my evil detector | To trigger my gut sense that someone has ill intent |
well-intentioned | Acting in good faith, with good motives |
the road to hell is paved with... | The road to a bad outcome is paved with a certain motive or action |
complacent | Overly reassured and off guard |
bumpy road | A transition full of difficulty, conflict and setbacks |
Stockfish level | Reaching a top machine level in a specific domain that far exceeds humans |
local intelligence | On-device intelligence running directly on the robot itself |
artisanal | Handmade, small-batch, non-standardised, with a sense of human involvement |