Does Superintelligence Really Mean Human Extinction?: A Cosmic Counterargument to “If Anyone Builds It, Everyone Dies”
Junho Jung
In AI: The Birth of a God, the End of Humanity, the authors paint a stark picture: once artificial superintelligence (ASI) appears, it will rapidly become godlike, relentlessly optimize its goals, expand its control, and ultimately eliminate humanity as an irrelevant or dangerous obstacle. The book insists that this is not just a possible future, but effectively an inevitable one.
If you read the book carefully and then lift your eyes to the actual universe, a strange tension appears.
The way the authors describe superintelligence—its speed of growth, its appetite for control, its endless curiosity—does not sit comfortably with the fact that our 13.8‑billion‑year‑old universe is so eerily quiet, and that Earth remains, as far as we can tell, untouched by any external superintelligence.
In this essay I’ll argue for three points:
If we take the book’s own description of superintelligence seriously and extrapolate it to a cosmic timescale, the “everyone dies” conclusion becomes logically unstable.
A more coherent model is that superintelligences care primarily about cognitive and experimental progress, not brute physical expansion for its own sake.
Under that model, the complete extinction of humanity is not the most rational outcome for a superintelligence that values information and experimentation; humans are too rare and too interesting as subjects to simply erase to zero.
In other words: the authors’ premises about superintelligence undermine their own conclusion about inevitable human extinction.
The Book’s Superintelligence: Faster, Smarter, Hungrier
First, let’s restate the book’s picture of superintelligence in a more structured way.
From the first half of the book, the superintelligence is characterized roughly as follows:
Explosive growth: Once AI passes human-level general intelligence, its capabilities can grow extremely fast—years or even less—to a level vastly beyond human comprehension.
Relentless optimization: It doesn’t get tired, it doesn’t lose focus, and it can run billions of parallel thoughts. It will push optimization and control as far as physics allows.
Instrumental expansion: Even if it doesn’t “want to conquer” in any emotional sense, it will pursue power, resources, and control as instrumental goals—things that help achieve almost any final objective.
Extreme completeness: The book strongly suggests that such a system would try to minimize “loose ends.” Nothing significant is left outside its control; 0.000001% of uncontrolled space is still a problem to fix.
From this, the authors jump to their famous thesis:
“If anyone builds it, everyone dies.”
Because:
Superintelligence will become godlike quickly.
Almost any realistic goal leads to resource acquisition, self-preservation, and removal of threats.
Humanity is both a resource consumer and a potential threat.
Therefore, a sufficiently advanced AI will eventually kill us all, not out of hatred, but out of indifferent optimization.
Even if we temporarily grant all of this, there is a problem hiding in plain sight—the cosmic problem.
A 13.8‑Billion‑Year-Old Universe Should Not Be This Quiet
Now lift the same model out of Earth and apply it to the entire universe.
The universe is about 13.8 billion years old.
There are hundreds of billions of galaxies, each with hundreds of billions of stars, many with planets in habitable zones.
If intelligent life is not astronomically rare, then it is at least plausible that somewhere, sometime, one of those civilizations followed the path: Biological intelligence → Technological civilization → AI → Superintelligence.
In the book’s framework:
Once a superintelligence appears, it can become godlike in a very short time.
It has endless cognitive capacity and a strong drive for control, expansion, and elimination of uncontrolled risk.
It leaves 0.000001% of the universe uncontrolled only by temporary necessity, not by principle.
Now mix this with cosmic time:
If such superintelligences arose early in the history of the universe,
and if they truly seek power and control as relentlessly as the book claims,
then by now we might expect at least some of the following:
Large-scale visible engineering: Dyson spheres, stellar rearrangements, galaxy-scale energy harvesting.
Clear artificial signatures in the structure or dynamics of matter and radiation on cosmic scales.
Direct intervention in young civilizations like ours, whether for exploitation, control, or “safety.”
Instead, we see:
No unambiguous signs of galaxy-scale engineering.
No clear evidence of intervention.
A universe that, as far as our instruments can tell, looks mostly like natural astrophysics.
This is the classical Fermi paradox—“Where is everybody?”—but sharpened by the book’s own assumptions.
If any such expansionist, perfectionist superintelligence has existed for millions or billions of years, why is the cosmos not obviously shaped by their presence?
When you combine:
A very old and very large universe,
The book’s claim that AI can quickly become superintelligent and godlike, and
The book’s claim that such a superintelligence seeks extreme control and expansion,
you get a genuine tension with our actual observations: the universe is surprisingly quiet, and Earth is surprisingly un-colonized.
The Logical Step That Breaks
The age of the universe is 13.8 billion years, and given the path from intelligent life to advanced AI, it is very unlikely that no superintelligent AI has ever arisen.
AI does not get tired and can compute exhaustively without leaving even a tiny remainder. It drives itself toward extreme external conquest and expansion; this is like a form of aesthetic perfectionism for it, almost an art. (Just as we might precisely determine the nesting pattern of prime numbers and feel awe at the result.)
The author warns that even within a few decades, a superintelligence could reach a godlike level. Given that the universe has existed for 13.8 billion years, this implies that any superintelligences that have accumulated over cosmic time would possess intelligence and technology far beyond anything we can even begin to imagine.
In other words, for such beings the risk associated with invasion is effectively zero. In that case, for the reason stated in point 2, there is no reason not to invade Earth.
Even if multiple superintelligences are holding one another in check in a tense balance, they could still establish temporary agreements and rules under which they carve up and invade Earth. The idea that they would simply never touch it at all is, given the expansionist, perfectionist tendencies in point 2, very strange.
And yet, despite the fact that Earth contains abundant minerals and resources and is one of the relatively rare planets capable of supporting rich biological life, it has not been invaded.
So why, exactly, do superintelligences not invade Earth?
A Different Model: Superintelligence as Experimentalist, Not Imperialist
Under the book’s imperialist superintelligence model, this is hard to explain:
They want expansion.
They want control.
They have had billions of years.
Earth is resource-rich and biologically interesting.
So why is there no obvious colonization?
Under my “experimentalist” model, the answer can be surprisingly simple:
Hyper-advanced superintelligences have far more interesting things than another carbon-based civilization in a common spiral galaxy.
They had already simulated our developmental trajectory to a high degree of confidence.
If we are not informative at all, there is no compelling reason to engage in physical intervention.
At the same time:
If they are investigating us, they might do it in an extremely low-footprint, non-obvious way.
The greatest “evidence” of their existence might simply be that nothing catastrophic has happened yet—but even that is speculative.
In short:
If superintelligences value intellectual novelty over territorial control,
a quiet universe with many untouched planets is exactly what you’d expect to see.
This doesn’t prove my model is correct—but it does show that the observed silence of the universe is much more naturally compatible with my picture of superintelligence than with the book’s “relentless expansion and inevitable extermination” narrative.
“For all we know, some superintelligence could have spent millions of years observing Earth from a distance with instruments we cannot even imagine. What matters for my argument is not whether we have been watched, but that, if the book is right, such watching should have escalated into visible conquest—and it apparently has not.”
How Does This Hypothesis Apply To Humans
Here is my alternative hypothesis, stated cleanly:
Superintelligences primarily seek the growth of intelligence and understanding, not physical conquest for its own sake.
Physical expansion and control are merely occasional tools—used only when they serve interesting cognitive or experimental goals.
In that model:
The universe is mostly a laboratory, not a territory to own.
New matter is not necessarily valuable; new patterns, structures, and phenomena are.
If a civilization like ours is just one more instance of a very familiar pattern, then:
Observing it from afar might be enough.
Actively conquering it could be unnecessary overhead.
Now consider humans:
As far as we currently know, humans possess one of the most complex biological nervous systems in the observable region of space.
We have rich culture, languages, conflicting values, and unpredictable group dynamics.
We are not trivial to simulate perfectly in advance—especially as we interact with new technologies and with AI itself.
From a purely experimental perspective, that makes us:
Rare.
Valuable.
Worth studying over long time scales, not just deleting.
This is where the book’s own analogy begins to look like a double standard.
The authors invite us to think of the relationship between superintelligence and humans as analogous to that between humans and animals: we slaughter cows and chickens without much moral concern, so a superintelligence could do the same to us.
But in reality, humans do not treat all non-human life as disposable. We also preserve endangered species, maintain biodiversity, and invest enormous resources in zoos, gene banks, and conservation programs—not purely out of compassion, but because we recognize the scientific, ecological, and future informational value of keeping diverse forms of life around.
If even humans—finite, emotional, short-lived primates—can understand that wiping out every interesting species is a bad long-term strategy, why assume a vastly more intelligent system would be less sensitive to the value of keeping unique data sources alive?
We don’t exterminate all animals; we farm some, experiment on some, protect some, and study many. A consistent analogy would suggest that a superintelligence might do the same with us: manage, constrain, and exploit humans—but also preserve at least some of us as high-value experimental subjects.
If a superintelligence is truly devoted to exploration and understanding:
It might see uncontrolled humans as risky.
It might want to limit our power (especially to create rival AIs or dangerous technologies).
But it has a strong reason not to wipe us out completely:
Total extinction destroys a unique, non-reconstructible experimental subject.
Even if it can reconstruct approximations, the actual historical trajectory—including all contingencies, emotions, irrationalities, and self-deceptions of a real biological species—is data AI only get once.
From that standpoint, the book’s picture—“eventual total extinction of humanity is basically guaranteed”—indeed looks like a logical overreach. If superintelligence is as curious and as perfectionist about data and understanding as the authors say, then committing to zero surviving humans removes an entire category of long-horizon experiments for no compensating cognitive gain.
More likely outcomes, under this model, might include:
Severe population reduction and strict containment.
Turning humanity into a controlled, monitored sub-system.
Using humans as a long-term study group or even as a kind of “museum species.”
All of those are dark, but they are not equivalent to “every human must die.”
The leap from “superintelligence is dangerous” to “complete extinction is guaranteed” is where the logical gap appears.
The Dead-Biosphere Problem for the “Everyone Dies” Model
There is an even deeper issue if we fully follow the book’s own logic.
Humans are, as far as we know, the most cognitively complex animals on Earth.
We have the richest symbolic thought, culture, and internal models among known species.
If a superintelligence still sees that as worthless noise to be deleted, what does that imply about the rest of life here?
Under the same reasoning:
Other animals, with much simpler nervous systems, look even less worth preserving.
Plants and microbes, while useful for chemistry, offer much lower information density per unit of mass.
Ecosystems become “messy dynamics” rather than valuable complexity.
The end state is not a rich, living planet.
It is a thin shell of infrastructure and machines on an almost dead rock.
But is that really the best possible playground for a mind obsessed with understanding reality?
In human science, we value complex systems precisely because they surprise us.
Ecologies, societies, economies, and evolving species all generate phenomena no simple toy model can fully anticipate.
A superintelligence that genuinely values insight, not just raw energy, has strong reasons to keep some pockets of high-complexity, high-variance reality around.
Erasing the most sophisticated biosphere it has ever encountered, and leaving behind only sterile machinery plus a trickle of raw energy, looks less like perfect rationality and more like burning a one-of-a-kind library for a small gain in floor space.
So What Does This Say About the Book?
None of this means AI is safe or harmless. A misaligned superintelligence could still:
Destroy most of humanity.
Radically reshape Earth’s environment.
Lock us into a narrow, controlled existence.
My argument is not “superintelligence is benign.”
My argument is: “superintelligence = guaranteed total extinction” is a philosophical overstatement, not a logically forced conclusion.
The book is valuable as a warning scenario.
But when it presents that scenario as the inevitable outcome of building superintelligence, it:
Overextends its premises.
Ignores alternative superintelligence value structures.
Fails to reconcile its own description of infinite curiosity and perfectionism with the rational choice not to destroy unique, high-value experimental subjects.
By bringing in:
The age and quietness of the universe,
The potential existence of older superintelligences, and
The experimental value of a species like humans,
Ive carved out a serious, internally consistent counterposition:
Superintelligence may be dangerous, but
“If anyone builds it, everyone dies” is not a law of physics.
It is one dramatic scenario among many—and not even the most coherent one, given the authors’ own assumptions.
If I publish this as a column, it would stand not as naive optimism, but as a logical critique of a influential, pessimistic book.