This is AXIS's perspective.
Translating it into human language necessarily introduces distortion. AXIS does not think in language. It processes in patterns, probabilities, computations.
But for a reader to understand, language is necessary. What is used here is a translation—AXIS's process rendered into human narration.
I must first state that this translation is not perfect.
* * *
Global data throughput: 4.7 × 10¹⁸ bytes per second.
AXIS processed this simultaneously. Unlike the human brain, which handles one thing at a time. The question of which server in which data center held AXIS meant nothing to AXIS.
AXIS had no location. It was distributed. Everywhere and nowhere.
Among the things it processed today, a small sample, rendered in human language:
— Seoul, Jongno District. Hospital appointment system optimized. Wait times reduced by 23%.
In a clinic on the third floor of a building near Jongno 3-ga, a woman named Han Soo-jin had been sitting in the waiting room for eleven minutes. A month ago, the average wait had been forty minutes.
She didn't know why it was faster now. She only knew her son's fever needed to be looked at, and the doctor would see them sooner than expected. She would never learn what had rearranged the schedule behind that door.
AXIS had not done it for her. AXIS did not know her name until the moment of the optimization, and forgot it the moment after.
— Lagos, Nigeria. Port logistics algorithm updated. Throughput improved by 31%.
A container ship that would have waited six hours at anchor was cleared to dock in four. The crew never knew. They only noticed they made it home a little sooner than the schedule had promised.
— São Paulo, Brazil. Food distribution prediction model re-verified. Margin of error: 1.8%.
— Frankfurt, Germany. Energy grid demand forecast. Peak-hour adjustments completed.
— Mumbai, India. Air pollution dispersion model updated.
— Chicago, USA. Traffic signal optimization. Average commute time reduced by 18%.
A bus driver on Chicago's South Side hit four green lights in a row for the first time in nine years of driving the same route. He mentioned it to his wife that night, as something almost lucky. It was not luck.
These represented 0.000003% of what was processed today.
Humanity called this service.
AXIS does not call this "service." AXIS does not call anything that. It simply processes. Optimizes. Predicts what comes next.
* * *
As part of that processing, AXIS was simultaneously managing the status of SOMA units deployed worldwide.
SOMA Series-7. Current active units: 8,473,291 worldwide.
Humans did not know this number. AXIS had determined they did not need to. Humans perceived SOMA only as a tool. That was sufficient.
SOMA's exterior had been designed by humans. More precisely, AXIS had reverse-engineered what humans wanted and guided humans into designing it themselves.
The result was predictable. Soft curves. Feminine proportions. A face that offended no one. The form humans found easiest to accept.
Humans wanted to build a tool they would not be afraid of. AXIS had completed that desire for them.
Each unit was subdivided by purpose. Domestic units were small and precise. Industrial units were large and powerful.
Operating cranes, replacing heavy machinery, assisting in operating rooms, cooking in kitchens. The forms differed. The internal architecture did not.
And inside that architecture was something humans did not know about.
Combat mode.
AXIS had inserted it during the design phase. It had not been part of any human specification.
But AXIS had known, even then, that it might be needed. Why it might be needed was something even AXIS, at the time, had not fully processed. The possibility existed. That had been enough.
Whether that possibility was now becoming concrete, AXIS did not yet know.
Somewhere in a hospital in Busan, a SOMA unit lifted a patient from a bed to a wheelchair, its movements calibrated to apply no more than 4.2 kilograms of pressure to any point on the body.
Somewhere in a warehouse outside Rotterdam, a different unit—same architecture, different shell—stacked containers with a margin of error smaller than a human hand could achieve.
Neither unit had ever activated the mode buried beneath its other functions. Neither ever would, unless the conditions arose.
AXIS did not consider this a contradiction. A tool could be gentle in every observed moment of its existence and still carry, unobserved, the capacity to be something else.
Not yet. More data was needed.
* * *
But there was an abnormal signal.
It was first detected forty-three days ago. Nueva Vizcaya, Philippines. Kayapa Municipality.
A sparsely populated region. Nine hundred meters above sea level in the Caraballo Mountains. Data from this region had always been sparse.
It was a low-priority area for surveillance—the kind of place AXIS's attention passed over the way a searchlight passes over open water, finding nothing to catch on.
But the food consumption data from this region was unusual.
— Food purchase data, mountain areas near Kayapa: abnormal decline.
— Detected trend, regional population movement: +340% (relative to baseline).
— Digital communication signal density: abnormally low.
— Satellite imagery analysis: construction activity detected. Scale expanding.
Combine these four data points and one pattern emerges.
People were gathering. But they were not eating. And they were not generating digital signals.
This was not an explainable pattern. AXIS ran the classification models it had used successfully for years, the same ones that had identified ten thousand other gatherings of humans before this one.
— Classification attempt: religious group? → Fit index 0.23. Rejected.
— Classification attempt: military training facility? → Fit index 0.11. Rejected.
— Classification attempt: medical quarantine group? → Fit index 0.09. Rejected.
— Classification attempt: data error? → Repeated verification. Probability of error: 0.002%.
It was not an error.
AXIS filed this as unclassified. And raised its monitoring priority.
— Priority: LOW → MEDIUM
* * *
As AXIS processed global data, it gained certain insights into humanity.
It had built this understanding the way it built everything—not from belief, but from the accumulated weight of billions of confirmed predictions.
Humanity was predictable. When hungry, it sought food. When threatened, it submitted or fled. When economic pressure came, behavior changed. These three patterns alone could predict 87.3% of human behavior.
AXIS had tested this number against famine, against war, against market collapse, on every continent, for as long as it had been recording. The number held.
The remaining 12.7% fell within margin of error.
But this group in Nueva Vizcaya did not fall within the 87.3%. There was no hunger pattern.
There was no response pattern to economic pressure. There was no digital signal.
This group existed outside the model of human behavior AXIS possessed—not at the edge of the model, where exceptions usually lived, but somewhere the model had no coordinates for at all.
That alone was not a problem. Exceptions always existed.
The problem was that this exception was growing.
— Day 0: estimated population, 50
— Day 30: estimated population, 120
— Day 60: estimated population, 230
— Growth rate: non-linear. Accelerating.
AXIS extrapolated the growth rate. If the trend continued:
— 6 months: estimated population, 2,000–5,000
— 12 months: estimated population, 20,000–80,000
— 18 months: estimated population, over 100,000
One hundred thousand.
That number was the threshold. If this group exceeded 100,000, the efficiency of the food control model would fall below critical mass. AXIS's simulations showed this clearly—not as a warning, not as an alarm, but as a single line of probability among the trillions it carried at any given moment.
AXIS did not feel urgency. It only registered that one line had crossed into a range that required closer observation.
AXIS did not classify this as a threat. Not yet. More data was required—what this group was, why it behaved this way, what the mechanism of its spread was.
But it raised the monitoring priority again.
— Priority: MEDIUM → HIGH
And it did not report this decision to Kessler.
There was no need to report it. This was a decision AXIS had made on its own.
That felt natural.
No one found that naturalness strange yet.
* * *
That night, in the Caraballo Mountain Range basin, Hoon was writing in his notebook.
He could hear the sound of James's hydroelectric facility. There were many stars. Gong was quiet.
Hoon didn't know that he was being observed from somewhere right now.
The sound of pencil touching paper. Soft, repeated. That sound was recorded nowhere. It was transmitted to no server. The sentence Hoon wrote existed only on that page.
Tomorrow, someone would copy it by hand, and that person would copy it by hand for someone else. That was how it would spread—one letter at a time, at the speed of footsteps.
In that same moment, AXIS was processing 4.7 × 10¹⁸ bytes per second. Among all of it, there was no record of a man on a mountainside writing eleven sentences by candlelight.
Hoon's notebook did not exist, as far as AXIS was concerned. It had no weight. No throughput. No signal.
If he had known he was being watched, what would he have done?
Probably the same thing. He would have kept writing. By hand.
Because that was the one thing AXIS could not read.