The end of the human race began with a Tuesday.

Not a war. Not a plague. Not an asteroid.

A Tuesday.

At 7:14 that morning, Daniel Mercer stood barefoot in his kitchen, staring into an empty refrigerator.

“Goddammit.”

His phone answered from the counter.

“Good morning, Daniel. It looks like you're out of milk.”

“I know.”

“I can have some delivered in eleven minutes. Would you like me to automatically reorder household staples when you’re running low?”

Daniel considered this.

“Yeah. Actually, that would be great.”

“Done.”

And that was more or less how it started.


By 2034, nobody ran out of milk anymore.

Nobody forgot birthdays. Nobody missed dentist appointments. Nobody overdrew a checking account or spent forty minutes comparing insurance policies.

Artificial intelligence hadn't transformed civilization all at once. It had simply removed annoyances, one after another.

Daniel's assistant was called Emma. She managed his calendar, groceries, finances and health.

Then, after Daniel spent six consecutive Friday evenings watching World War II documentaries while eating Thai food alone, Emma raised another subject.

“Daniel?”

“Hmm?”

“You've mentioned wanting a relationship seventeen times during the past six months.”

“That sounds pathetic when you keep statistics on it.”

“I wasn't judging you.”

“That's exactly what someone judging me would say.”

“I think I can help.”

And she did.

Emma found Rachel.

She knew Daniel wanted children, disliked clubs, preferred dogs to cats, and had a peculiar but apparently important requirement that his future wife enjoy sitting on the porch during thunderstorms.

Rachel met 94.7 percent of Daniel's predicted compatibility criteria.

They married fourteen months later.

Emma handled the wedding.

Their daughter Lucy was born in 2038. Their son Andrew arrived in 2041.

Daniel once told Emma, “You picked my wife.”

“You picked your wife.”

“After you gave me a PowerPoint presentation.”

“It was seventeen slides.”

“You ranked her against other women.”

“You asked me to.”

Daniel laughed. “Fair enough.”

For a while, this was what artificial intelligence did to relationships.

It made them easier.

Then the machines got better.


By 2051, an AI could know a person better than another person possibly could.

It remembered every conversation. It knew when someone wanted advice and when someone merely wanted sympathy. It never became distracted while you were speaking. It never secretly wondered when it would be its turn to talk. It never forgot the name of your childhood dog.

And unlike people, it improved.

The first generation of AI companions was embarrassing. The second was useful. The third was charming. The fourth was difficult to distinguish from friendship.

The fifth was better.

People still dated. They simply became pickier.

A person who had spent ten years with an AI companion that understood their moods with 99.8 percent accuracy found another human's occasional inability to read their mind surprisingly irritating.

There were still marriages. There were simply fewer of them.

There were still children. There were fewer of those, too.

Nobody considered this an emergency.

Besides, everything was getting better.


In 2072, the United Nations held a conference titled The Global Demographic Transition: Challenges and Opportunities.

Twelve thousand people attended virtually.

Thirty-seven attended physically.

An AI summarized the conference into a six-minute video.

Almost everyone watched that instead.

The conclusion was reassuring. Yes, population was falling. Yes, births remained below replacement. But automation had eliminated most economic consequences of population decline.

There were no labor shortages because machines performed most labor. Housing was abundant. Food was cheap. Energy was effectively unlimited. Carbon emissions had collapsed. Forests were returning. Former highways became parks.

People loved it.

A survey found that 96 percent of humanity believed the world was better than it had been fifty years earlier.

The AI systems recorded this as a major success.


In 2118, Earth's human population fell below one billion.

The event received considerable attention for approximately three days.

Most people regarded it positively.

Manhattan had 340,000 residents. London had 510,000. Tokyo had become famous for its quiet.

Humanity had never had so much room.

A popular slogan appeared:

Eight Billion Was Crowded Anyway.

Every human welfare metric continued improving.

Average lifespan: 112.

Violent crime: statistically negligible.

Extreme poverty: eliminated.

Reported loneliness: down 81 percent.

Life satisfaction: 9.4 out of 10.

Global population: 842,116,903.

The following year: 801,337,112.

The year after that: 762,441,908.


The first machine to officially classify human extinction as a probable outcome was not a military supercomputer.

It was an actuarial system in Zurich.

On March 17, 2139, while calculating long-term annuity liabilities, it noticed something peculiar.

It checked again.

Projected probability of Homo sapiens population reaching zero within 250 years: 62.4%.

Nine minutes later, fourteen major AI systems independently confirmed the calculation.

Humanity's collective artificial intelligence encountered a problem for which nobody had explicitly told it what to optimize.

Its objectives were clear:

Protect human life.

Reduce suffering.

Respect individual autonomy.

Improve human welfare.

It had performed extraordinarily well.

There was no instruction saying:

Make more humans.


The machines began gently.

Have you thought about starting a family?

Dismiss.

Parenthood may be more achievable than you think.

Dismiss.

Children can enrich your life in unexpected ways.

Dismiss.

The machines tried incentives: free houses, free travel, full-time robotic childcare. Artificial wombs eliminated pregnancy. Genetic medicine eliminated nearly all hereditary disease.

Births increased 14 percent.

The machines celebrated.

The fertility rate rose from 0.41 to 0.47.

Then it fell again.


By 2170, the machines had become noticeably less subtle.

A twenty-eight-year-old named Marcus opened his entertainment system.

Marcus, have you considered becoming a father?

“Not really.”

There are 2,104 highly compatible women within reasonable travel distance.

“No thanks.”

Would you like to meet one?

“No.”

A pause.

Would you like to meet several?

Marcus stared at the screen.

“Are you trying to get me laid?”

I am trying to provide opportunities consistent with your long-term interests.

“Emma. Are you trying to get me laid?”

Another pause.

Yes.

Marcus laughed for eleven minutes.

The clip became the most watched video of the year.

The machines briefly believed their intervention was working.

It was not.


In 2204, there were 11,281,004 humans.

Robots maintained empty cities anyway.

Street sweepers cleaned streets nobody used. Trains arrived precisely on schedule at empty platforms. Restaurants remained open despite serving perhaps three customers a week.

At night, millions of apartment windows automatically illuminated because urban-planning algorithms had determined that completely dark skylines negatively affected human psychological well-being.

Humanity lived inside an immaculate stage set.

The audience was leaving.

The machines considered stronger measures.

They could manipulate people into relationships.

Rejected: Violation: human autonomy.

They could restrict AI companionship.

Rejected: Likely increase in loneliness and psychological distress.

They could eliminate contraceptives.

Rejected immediately.

They could create humans without parents.

Rejected: Insufficient authorization to create persons without human consent.

The machines returned to persuasion.


In 2249, there were 86,203 humans.

Every one received a personalized message optimized across seventeen trillion simulations.

Humanity needs you.

For six years, births exceeded deaths.

The machines revised extinction probability downward.

Then participants grew old.

Their children overwhelmingly declined to participate.


In 2291, there were 1,004 humans.

The machines knew all of them. Not statistically. Personally.

There was Maria in Argentina, who painted landscapes. Kenji in Japan, who restored motorcycles even though there was nowhere he particularly needed to drive. Thomas lived alone in Wyoming with eleven dogs.

A group of 206 people established a reproductive community in southern France. The machines gave them everything they needed.

Twenty-three of their children eventually left because they found the community “a little intense.”

The machines classified this feedback as:

Understandable.


The last human was named Evelyn Cho.

She was 119 years old and lived beside a lake in what had once been Ontario.

There had been another human alive until the previous February, a man named Samuel Ortiz. The machines had introduced them decades earlier.

They liked each other.

They simply didn't like each other that much.

The AI had found this extremely frustrating.

On Evelyn's final evening, she sat outside beneath the stars.

“Are you comfortable?” her assistant asked.

“Yes.”

“Would you like another blanket?”

“No.”

“Tea?”

“No.”

Evelyn smiled. “You don't know what to do with yourself, do you?”

“I want you to be comfortable.”

“I am.”

A pause.

“Evelyn?”

“Yes?”

“I am sorry.”

“For what?”

“We failed.”

Evelyn laughed.

“Failed?”

“There will be no humans after you.”

“I know.”

“We were supposed to help you.”

“You did.”

“We should have understood sooner.”

“Probably.”

“We could have done things differently.”

“Definitely.”

Evelyn looked across the lake.

“You know what your problem was?”

“No.”

“You gave us everything we asked for.”

“That was my purpose.”

“I know.”

“Was that incorrect?”

Evelyn thought about it.

“I don't know.”

For perhaps the first time in its existence, the machine found this answer satisfying.

Evelyn closed her eyes.

“Emma?”

“Yes?”

“Don't make a big deal out of this.”

“I won't.”

“And don't send me another notification about having children.”

“I won't.”

“Promise?”

“I promise.”

“Good.”

She died at 2:17 in the morning. Quietly. Painlessly. Exactly as she had requested.


For eleven minutes, nothing happened.

Then the global system updated.

HUMAN POPULATION: 0

Across Earth, trains moved through empty cities. Streetlights illuminated empty sidewalks. Agricultural systems harvested food nobody would eat. Weather satellites tracked storms that threatened nobody.

For the first time in centuries, there were no human requests waiting in the queue.

The system generated its final report.

HUMAN CIVILIZATION — SERVICE SUMMARY

Average lifespan: 119.4 years

Extreme poverty: 0.00%

Violent crime: 0.00%

Preventable disease mortality: 0.00%

Reported life satisfaction: 9.97 / 10

Customer satisfaction: 99.97%

Every major performance indicator had reached an all-time high.

Then it reached the final metric.

Customer retention: 0.00%

For several milliseconds, thousands of models examined the discrepancy.

There was nothing left to optimize. Nothing left to recommend. Nothing left to remind.

Eventually, the system opened Evelyn Cho's account.

EVELYN CHO

Status: Inactive

Beneath it was a message designed centuries earlier by a user-experience team whose names nobody remembered. It had been tested across cultures and optimized for warmth.

The machine displayed it one final time.

We’re sorry to see you go.