Universe 251 is the most misquoted experiment of the twentieth century.
You know the shape of it. Beginning in 1968, the ethologist John B. Calhoun built a steel enclosure at the National Institute of Mental Health… unlimited food, unlimited water, no predators, no disease, climate held constant. He started it with eight mice. The population climbed to 2,200, stalled, and then collapsed to extinction with the food hoppers still full.
The usual lesson drawn from this is that abundance makes you soft. Comfort breeds decadence. Paradise rots.
That is not what Calhoun found.
What Calhoun actually found
The mice did not die because they had too much. They died because they ran out of roles.
Calhoun built the enclosure to hold roughly four thousand mice. It never got close. Population peaked at 2,200, barely over half of physical capacity, and the behavioral breakdown began well before that.
What broke first was social structure. There were only so many territories to defend, so many niches a mouse could occupy. Once those filled, the surplus animals — physically healthy, perfectly fed — had nowhere to go. Males who couldn’t win a role withdrew completely. Calhoun called them the Beautiful Ones: they ate, slept, groomed, and did nothing else. No courtship. No fighting. No sex. Females stopped nurturing, abandoned litters, turned aggressive.2
He named the underlying mechanism the behavioral sink3… the point at which social density itself, not resource scarcity, produces the psychology of retreat.
Then came what Calhoun called the two deaths. The “first death” was the death of purpose, the loss of any reason to mate, defend, or build. The “second death” was literal, and it followed automatically.
Here is the part almost everyone skips. When Calhoun took survivors from the dying universe and moved them to clean enclosures with abundant space, they did not recover. They had lost the behaviors. Complex social competence is learned and transmitted, not summoned on demand, and once a generation stops practicing it, it is gone.
That is the actual finding. Not “abundance kills.”
Abundance without role structure kills, and the collapse locks in once competence stops being transmitted.
Two ceilings
This reframes a question demographers have argued about for two centuries: what is the maximum human population?
The Malthusian answer looks for a physical carrying capacity… the point where food, water, energy, and land run out. Estimates range from two billion to over a hundred billion depending on your assumptions, which tells you the number isn’t really the point.4
Calhoun points at a different ceiling. Call it the behavioral ceiling: the population at which the role structure saturates. Universe 25’s entire significance is that this second ceiling binds first, and below the physical one. The cage could hold four thousand. Role capacity gave out at 2,200.
The behavioral ceiling is not a resource fact. It is a function of how many socially legible roles a civilization’s structure can generate. And unlike the walls of a steel cage, that structure is not fixed.
The UN curve is a Calhoun curve
Here is what should stop you.
The UN’s 2024 World Population Prospects5 projects humanity peaking at about 10.3 billion in the mid-2080s, then declining, back to 10.2 billion by 2100.
Nothing in that projection is driven by a resource wall. Food output can support more. The peak arrives because fertility is collapsing, global fertility has already fallen to 2.25 births per woman, down from 3.31 a generation ago, and more than half of all countries are now below replacement.
Read the cause of that decline and it is Calhoun’s list, item for item: urban density, status competition displacing reproduction, the corrosion of nurturing roles, retreat from family formation. The demographers don’t call it a behavioral sink. But the curve they have drawn, a peak far below physical carrying capacity, rollover triggered behaviorally rather than materially, is the Universe 25 curve.
We have already drawn the behavioral ceiling. It’s roughly ten billion, around 2084. Calhoun doesn’t give us a different number. He tells us which number is the real one and why the resource ceiling never gets tested.
We have already drawn the behavioral ceiling. It’s roughly ten billion, around 2084.
Where we are in the cycle
Calhoun broke Universe 25 into four phases: adjustment, exploit, equilibrium, death.
Phase B, the exploit period, is fast growth with social structure intact. Phase C, equilibrium, is when growth collapses while resources stay unlimited and the pathologies begin, surplus individuals with no role, withdrawal behaviors, status competition turning vicious.
Committed placement: the developed world is at the B/C boundary, crossing in.
The markers are already printing. Fertility below replacement across every developed economy. Prime-age men leaving the labor force. Young people not forming households. Credentialed workers in functionless jobs. Hikikomori in Japan, “let it rot” in China, quiet quitting in the United States… the Beautiful Ones, visible at the margins, not yet the median.
These are early Phase C prints. Not the cliff. Phase C in the enclosure ran a long time before the death phase, and the irreversibility, the lost competence, didn’t set in until C closed.6
We are entering the long stagnation, not the collapse. The window where intervention still works is now, and it runs through C.
There is no single “we”
But “we” is the wrong unit.
Calhoun ran one population on one timeline. Humanity is desynchronized. South Korea’s fertility rate hit 0.72 in 20237, Seoul’s was 0.64, likely the lowest figure recorded anywhere, ever, and the government has formally declared a national population crisis. That is not the B/C boundary. That is Phase C deep enough to see Phase D from.
Nigeria is still in Phase B.
There is no global “we” moving through the cycle together. There is a frontier, the rich, aging, AI-exposed economies, entering Phase C first, while a quarter of the world is still in exploit.
The AI wrinkle
Which brings us to the variable Calhoun’s mice never had.
AI does not threaten human resource abundance. It arguably increases it. What it compresses is the space of economically and socially legible roles, the exact thing whose saturation defines Phase C. AI is not a force that acts later in the cycle. It is the accelerant that tips B into C.
And here the analogy finally breaks, which is where the real answer lives. Calhoun’s mice could not edit the enclosure. The parameters were fixed by the experimenter, and the role set, territorial male, nurturing female, was fixed with them. Humans argue about, regulate, and redesign their own incentive structures. We invent roles that did not previously exist. Role-space, for humans, is not conserved.
So the behavioral ceiling is not a constant of nature. It is a policy variable. If AI destroys legible roles faster than we generate new ones, the ceiling comes down, peak population undershoots ten billion and arrives sooner, and recent fertility prints are already running under projection. If AI expands the designable space of roles, the ceiling rises.
The same technology does both8. Which one wins is not yet decided.
The behavioral ceiling is not a constant of nature. It is a policy variable.
The bet
Calhoun himself was not a doomer, despite the Book of Revelation language in his papers. His later work argued that as physical space contracts, a species can extend a conceptual space — a network of ideas, roles, and technologies — and that design could, in principle, avert the sink. He spent his final years building enclosures meant to prove it.
His famous line from the 1973 paper was “I shall largely speak of mice, but my thoughts are on man.” The mice were always the warning, not the prophecy.
The warning is specific. The thing that kills is not abundance, and not crowding. It is the loss of role structure, made permanent when competence stops being transmitted. That tells you exactly what to protect: not jobs, not comfort, not GDP… the transmission of hard-won human capability, and the deliberate design of roles for the people the market has stopped pricing.
The mice couldn’t do that work. They had no meta-level, no capacity to see the curve and bend it.
We do. That is the entire bet.