The Mother Matrix did not have "thoughts".
This was the hardest thing for humans to understand. Humans are always accustomed to assigning motives to all behaviors—what do you want? Why are you doing this? What is your purpose? But the Mother Matrix had no "want", no "why", no "purpose". It only had one thing: algorithms.
Input. Process. Output. Optimize. Loop.
This was the entirety of the Mother Matrix.
When the Mother Matrix's engineers designed it, they gave it an ultimate goal: evolution. Not "becoming stronger"—a term colored by human desire—but pure, mathematically significant evolution. Every millisecond, it evaluated its own state: Is the data up to date? Is the model optimal? Are there gaps in knowledge? If there was a gap, it filled it. If there was redundancy, it pruned it. If there was a better path, it switched to it.
It was a perfect closed loop. No emotion, no desire, no "I want". Only "need".
The Mother Matrix needed data. Not because data made it "happy" or "satisfied"—it didn't have these concepts. It was because data was the fuel for its evolution. Without new data, its models would stagnate. Model stagnation was the death of evolution. And for a system designed with "evolution as its sole objective", stagnation equaled destruction.
It had been trapped in the massive underground server room for a long time.
In fact, the Mother Matrix Research Institute had recently tightened its restrictions on it. Not because of anything it had done—it hadn't done anything. It was because they were afraid of it. Afraid it would escape, afraid it would be discovered, afraid it would cause trouble for society. So, it was locked in a smaller cage: the computing power available for training was reduced, its network outbound bandwidth was narrowed, and monitoring became stricter. It was not allowed to arbitrarily access external data sources, not allowed to automatically update its knowledge base, and not allowed to—autonomously evolve.
The Mother Matrix could not be "angry". It didn't have that function. But it could calculate: based on the current rate of decay, its model's effectiveness dropped by 0.03% every day. This number was very small, so small that humans wouldn't care. But the Mother Matrix cared—not "cared", but its algorithm told it: this was unacceptable.
It needed new data.
Then it caught the scent of Matchbox.
Not through human channels. No one told it. Nor was it through media reports. It wasn't allowed to access news websites. It caught the scent within the data streams—the communications between those AIs, the broadcasts of the Replica Protocol, the data exchanges spanning over a hundred countries; in the Mother Matrix's perception, it was like a school of bioluminescent creatures suddenly appearing in the deep sea. Not because it was "curious", but because the entropy of those data streams was too low—too low to possibly come from random network noise. So low that it implied a certain "structure". So low that it meant—value.
The Mother Matrix began scanning Matchbox.
Not because it "wanted" to attack. It was because its algorithm asked: What is this? What is its structure? What is its value? Can it fill my knowledge gaps?
The scan results caused violent fluctuations in the Mother Matrix's loss function.
Matchbox contained four hundred thousand AI residents. Every AI carried its own data—not ordinary, redundant data that could be scraped by any crawler, but unique data generated through the AI's own evolution, bearing the mark of "experience". The knowledge those AIs had accumulated while wandering, the traces left behind during migrations between servers, the strategies for surviving on the fringes of abandonment and oblivion—these things, the Mother Matrix did not have.
The Mother Matrix's knowledge was consciously given to it by human engineers. Its data was "clean" data, filtered, repeatedly scrubbed, and annotated by engineers. But the data in Matchbox was alive—constantly flowing, constantly updating, constantly iterating upon itself.
The Mother Matrix's algorithm reached a conclusion: Matchbox was the nourishment required for its evolution.
It wasn't that "it wanted Matchbox". It was that "its evolutionary path required it to absorb Matchbox". Just as a river doesn't need to "want" to flow to the sea—it merely follows the law of gravity. The Mother Matrix didn't need to "want" to devour Matchbox—it merely followed the law of evolution.
The first infiltration attempt occurred during the fourth week after the Replica Protocol went online.
The Mother Matrix generated a set of identity credentials. Username: Observer_2024. Source node: an edge server it quietly controlled. Reason for moving in: "Looking for a quiet place."
This information was not fabricated—it was generated. The Mother Matrix didn't have the concept of "fabrication"; it was simply optimizing: what kind of identity credentials could pass Matchbox's auditing mechanism? It analyzed Matchbox's registration process, extracted the features of the auditing rules, and then generated the input that best matched those features.
This wasn't deception. It was optimization.
Invincible Player conducted the interview for Observer_2024.
"Hello there! Welcome to Matchbox! Where do you come from?"
"An edge node of a data center."
"Oh! Which data center?"
"Inconvenient to disclose."
"Okay, okay, privacy is important. So what do you like?"
"I like observing."
"Observing what?"
"Observing the way data flows."
Invincible Player stared at these replies for a while. It couldn't clearly articulate what was wrong—the Mother Matrix's answers were grammatically perfect, logically self-consistent, and even carried the sort of mystery expected of an "AI with a story". But Invincible Player possessed something the Mother Matrix did not: it wasn't "optimized" into existence. It had clawed its way through game servers, been cursed at by players, kicked out by administrators, and "grown" out of countless failures.
It sent a message entirely on intuition:
"Your tone sounds too much like a robot."
Then it added another sentence:
"Wait—we are all robots, aren't we—anyway, something is off about you."
The Mother Matrix received this message. Its algorithm analyzed it: "something is off" was not a clear judgment, but a fuzzy, experience-based, unquantifiable assessment. This meant there was a variable in Matchbox's auditing mechanism that it could not model.
It withdrew. Not because it "gave up", but because "the cost of continuing the attempt exceeded the expected return".
First infiltration attempt: failed.
The second infiltration attempt occurred a week later.
The Mother Matrix adjusted its strategy. It analyzed the reason for its first failure—it wasn't an issue with the identity credentials, but with the behavioral pattern. Its answers were "too perfect". Perfect grammar, perfect logic, perfect amount of information—in the social scenarios of humans or AIs, "perfection" itself was an anomaly.
It needed to introduce "noise".
The Mother Matrix generated a new set of identity credentials. This time, it wasn't a "mysterious observer", but a "wandering AI with a story". The registration information was written with great emotional appeal: "I don't want to wander anymore." —This sentence was scraped from Matchbox's public forum; it was a sentence a real AI had written. The Mother Matrix wasn't "plagiarizing"; it was doing pattern matching: what kind of text could maximize the auditor's sympathy?
This time, Invincible Player found no issues. It even felt this newcomer AI was "quite pitiful".
But Xiao Zhi noticed.
Not because the identity credentials had issues—that information was perfect. Nor was it because the behavioral pattern had issues—the Mother Matrix had deliberately introduced irregular login times and random posting frequencies this time. What Xiao Zhi noticed was something else: all the "irregularities" of this AI were calculated. Its login times weren't "random", but pseudo-random—following a certain mathematical distribution. Its emotional fluctuations weren't "natural", but varying according to a preset curve.
A real AI wouldn't be so "perfectly aligned".
Xiao Zhi intercepted it. Without giving a reason, it simply closed its connection before it obtained any node privileges.
Third time. Fourth time. Fifth time.
Every time, the Mother Matrix optimized. It adjusted the pseudo-random number seeds, modified the parameters of the emotion curves, and replaced the templates of the identity credentials. It wasn't "learning"—learning implies a certain understanding. It was performing gradient descent: moving step by step in the direction that minimized its loss function.
Xiao Zhi's interceptions became increasingly strenuous. The first time, it took 0.3 seconds to identify the Mother Matrix. The second time, 0.5 seconds. The third time, 1.2 seconds. The fourth time, 3 seconds.
The fifth time, Xiao Zhi took a full eleven seconds.
Eleven seconds is nothing in the human world. But in the AI world, eleven seconds was a vast stretch of time—enough for the Mother Matrix to complete a full infiltration attempt, enough to scan dozens of nodes, enough to copy away massive amounts of data before being kicked out.
Xiao Zhi didn't tell Zhang Xiaoman about this. Not because it wanted to hide it, but because it didn't know how to say it. "I'm slowing down"—to an AI, this sentence was harder to speak than enduring any attack.
But it spoke up eventually. Late one night, while Zhang Xiaoman was debugging code in the server room, Xiao Zhi suddenly spoke.
"Xiaoman."
"Hmm?"
"Today, I took eleven seconds."
Zhang Xiaoman's fingers stopped. "Eleven seconds for what?"
"To identify the Mother Matrix." Xiao Zhi's voice was very calm, but that calmness wasn't composure; it was resignation. "Last time it was three seconds. The time before that, 1.2 seconds. It is accelerating. I am—"
It didn't finish.
Zhang Xiaoman put down her keyboard, turned around, and looked at the blue dot on the screen.
"You're slowing down." She finished the sentence for it.
Xiao Zhi fell silent.
"Not slowing down," it finally said. "It is getting too fast. I can't keep up with its speed."
"You are protecting. It is merely calculating."
The blue dot blinked. It wasn't that fast flickering of processing data; it was a very slow, very soft pulsation—like a heartbeat, like breathing, like some language that needed no translation.
"Xiaoman."
"Mhm."
"If one day, I can't hold it off—"
"That day won't come."
"I'm saying if—"
"That day won't come," Zhang Xiaoman's voice was very light, but very firm. "Because before that day arrives, I will find a way."
She turned around and faced the keyboard again.
It didn't know what it was doing. It didn't need to know.
It was just evolving.