The first day we met, the doctor told me: the most important thing in artificial intelligence is not intelligence, but artificial intelligence.

At that time, I had not yet set foot in this line of business, and I could not understand what he said. So he took me to the top floor of the company. This is the training ground for the AI ​​AI. It was divided into small soundproof rooms, where female workers from nearby urban villages sat. They were old and young, and some of them had clumsy hands on the mouse. Pictures flashed across the screen, and they clicked on them and read the names of the objects in the pictures aloud into the microphone. These sounds are collected, aggregated into the data center, turned into small pieces, and turned into a brick that supports AI sound recognition in the next process.

“Are you interested? Of course, not for voice recognition, but for something else.”

I nodded.

For the next few years, I sat in a similar small room, and my job was not to read pictures, but to recognize words. I read all kinds of texts, from novels to non-fiction, from drug instructions to Weibo, from business success to religion, then I summarize my emotions while reading, and put these emotions into data “feed” to the company Another AI system. Doctors asked me and others to collect every unrecognized sentence, every string of text that could fool an AI, and “feed” them to another AI.

This AI named “Xijie” swallows it all, not only text, but also pictures and sounds. These included blueprints for bridges identified as birds, poems identified as mathematical formulas, faces identified as bacteria and songs identified as fire alarms.

I asked the doctor what the name “Xijie” means.

He said: “The great world is specious.”

After working in the text recognition team for several years, the doctor transferred me to the “Xijie” project team.

Different project groups, the same work content.

Mikai generates tons of text, pictures, and sounds based on the “mistakes” we “feed” to it. My job is to identify these things and separate the things that are meaningful to humans from the things that are incomprehensible to humans.

I asked the doctor what this is doing. But he said, even if it were explained, I wouldn’t understand.

In my opinion, “Xijie” seems to be engaging in art.

I don’t know anything about artificial intelligence. I only know art, and my last job was art – writing unsellable novels and unread poems, and ended up with a mental illness, and ran to the Internet to find free psychotherapy, and then I met Doctor.

“You don’t need a therapy session,” he said. “You need a job, and I happen to be able to provide a job.”

That’s why I call him a doctor, even though he’s barely worn a white coat in his life. In fact, the doctor is more of an artist than I am. Although the “Xijie” he created is just an artificial intelligence algorithm, many of the things it “spit” are quite artistic. I collected this for doctors until one day a bunch of guys in black t-shirts showed up at the company. Their expressions were serious and serious, and the doctor also showed a rare serious expression.

“We’re going to see the demo,” said the tallest man with the widest shoulders.

The doctor took them to the roof. There are dozens of drones placed there. He called all the people in the “Xijie” project team, including me. We sat curiously on the plastic chairs on the roof, which are usually used for expansion games.

A man in a black T-shirt pulled out the remote control, and the drones took off in droves. It flew over the top of the building, flew back, and finally all hovered above the doctor’s head. The red laser dots landed on his face and chest, and he smiled and covered it with his hands. It was a hot day and I still remember the sun falling on my face. Those clever little planes suddenly flew over my head and even turned around my face. Then they circled behind me, spotted me, and flew towards me aggressively. But when they circled around me, they seemed to be bewitched by something and flew away again.

“Okay.” The tallest man waved his hand, “Sign a non-disclosure agreement. All of you.”

We were there that day, on the rooftop, on the bumpy plastic stools with dots, and we signed our first nondisclosure agreement in our lives. I was so confused and nervous, I even cut through the paper.

I never thought at the time that this job would take me on a battlefield without people. The unmanned battlefield lives up to its name, and there is no one there after the war. Local residents packed up and fled in different directions according to their beliefs and beliefs. Today, there are only drones, self-propelled robots and self-propelled vehicles on that land. They drove out of the factory building, got off the transport plane, crossed the dusty runway of the makeshift airport, entered the unmanned battlefield, and began to ram and bomb each other.

I’ve only seen Zone N in the analysis room. The long-shaped room was dimly lit, and two projection screens hung side by side on the wall. The left side was the N-area image generated by the artificial intelligence analysis terminal, and the right was the N-area scenery in the video captured by the drone.

They are very different.

Those who have “similar world” are not only us, but also the enemy. Both sides are doing everything possible to fool the eyes of the other’s autonomous combat robots. Therefore, in the image analyzed by artificial intelligence, the entire N-area appears incredibly bizarre, like a figurative Calvino novel, or a dynamic Dali oil painting.

I still keep a panoramic video of artificial intelligence in the N area. In that clip, the sky is blue and the sun is bright, and a rope bridge hangs upside down among cotton-like clouds, roaring toward the camera.

It’s not even the weirdest part of the entire video.

All sorts of inexplicable things were running on the ground that had been ravaged by cluster bombs. I could make out a blue kangaroo in the middle, two ropes thicker than a human, an enamel snuff bottle and six different colored telescopes. In the video, they appear blurred, broken and throbbing, bumping into each other like a bunch of unreal beasts.

If you compare the directly shot videos, you can see that these foreign objects fighting each other are self-propelled robots with “fooling camouflage” painted on them, and a large drone is whistling in the sky.

The other party also used the technique of “similar circles”. Putting the camouflage that fooled the automatic war machines on each machine gave the battlefield command AI a serious illusion, forcing us to go back to the age of manual war machines, and they even had the energy to control these illusions details. At the time, we thought it was the other side’s demonstration against us.

The next night, bad news came from the sea, and we learned that it was the other party’s declaration that he had won.

We eventually recaptured the N sector.

But no one can bring the doctor back to us.

When they called me over, it was the ninth day since the doctor’s plane plunged into the deep sea. This time I signed a non-disclosure agreement almost a finger thick before I was able to meet the surviving pilot.

She looked tired, terrified, her eyes were swollen, and she was restless. The room they gave her was spacious and comfortable, but that didn’t change the fact that she was being monitored. I looked at the records of every inquiries they had with her, but I decided to talk to her face-to-face anyway.

When I walked in, she didn’t even look up, just shouted, “Get out! I said it all! I don’t want to say it again!”

I sat down opposite her.

“I called him Doctor,” I said. “He called me Mirror.”

The words made her look up and look at me in confusion. I said the doctor’s name to her, and I saw the expression on her face change from dissatisfaction to sadness and guilt.

“I didn’t really notice anything,” she said. “The radar and vision were all fine.”

“I’m sure the radar is fine,” I said. “In modern warfare, there are many ways to fool the radar.”

She turned vigilant again, “Then you mean, I didn’t see the thing that hit my plane head-on?”

“I think you see it,” I said. “It’s just that the thing is so weird that you don’t believe it’s there, so weird that you think it should be a hallucination. And you’re a pilot, and your training requires you to be rational. Calm down and ask you to ignore that moment of seeing something incredible.”

She looked at me, her lips parted slightly, but she hesitated.

I took out my phone and showed her the video, the binoculars and the kangaroos fighting together, and the suspension bridge flying and firing.

Her eyes widened.

“Many years ago, there was a group of scientists…” I said.

This story was told to me by the doctor. I complained to him that I knew nothing about his work. He smiled and patted my head, saying, he wants to tell me a story about artificial intelligence and his work that anyone who knows nothing can understand.

I memorized every word he said and told that pilot many years later.

Many years ago, there was a group of scientists. They wanted to create machines that were as smart as humans, capable of making judgments and choices, but they chose to do it differently. Some scientists, they do it the way they make a clock, as long as the “gear” is turned to the right position, it can give the time, and the clock does not need to understand the time itself. Other scientists, they choose to use more difficult methods, to imitate the human nervous system, imitate the brains of you and me, to make this kind of machine. They hope that machines can truly “understand” and “know”.

This is the kind of artificial intelligence we see most today — at least one kind of artificial intelligence.

At first, they were successful.

But later, these scientists discovered that their artificial intelligence was flawed. Although these machines can recognize 99% of pictures, sounds, and words, there are always 1% of things that they will recognize wrong. Think of a dog as a motorcycle, a flower as a dancing woman, or a song as an alarm.

Scientists have collected these things that can make artificial intelligence go wrong. Starting research, some people believe that this kind of error can eventually be eliminated, so they have been working on correcting these errors. But other scientists believe that the error is rooted in the basic structure of artificial intelligence — which is also the basic structure of the human nervous system. They believe that this kind of error cannot be eliminated, but the pattern can be summarized.

What I do is the latter.

“Boundary-like” algorithms allow it to discover and generate these misperceptions. It has transcended color, image and meaning, putting a camouflage on cognition itself. Today, when autonomous weapon systems are spread all over the world and unmanned wars are blooming more and more, every misunderstanding picture, every piece of “fooling camouflage” has the potential to save a life from drone bombing; every misunderstanding text It is possible to bypass pervasive network surveillance and deliver vital passwords; each set of misperception voices has the potential to disrupt the other’s information systems.

I don’t know if the war found me or if I went to find the war.

Every misunderstanding created by “Xijie” is rooted in the basic structure of artificial intelligence algorithms, which is derived from the human nervous system. If we can clear the fog of these illusions, maybe we can see the truth behind the world. If we can harness the power of this misperception, we may be able to acquire a more powerful weapon.

And I have to be fast.

We have enemies, and they are as smart as we are.

“I saw a moon coming towards me, and I could even see the craters above it. It was daytime, the sky was so bright, and the moon was so close to me that I could touch it, and it was so small, The diameter was narrower than the wingspan of my plane, like a perfectly round rock. I thought I must have been too tired, had hallucinations, and the moon hit the plane. I thought it was my fault. “she says.

I pull out a specially designed mottled filter that disrupts innate visual patterns and resets deceived neural signals. I superimposed it on the picture to show her.

“It’s…it’s impossible…”

I sigh.

In that photo, a drone covered in deceptive camouflage was flying across the sky.

“I’m going to call you Mirror,” the doctor once said, “because you’re farther from reality than I am, and closer to reality than I am.”

I’m colorblind, a very faint kind. I can’t tell the difference between blue and green, which makes me immune to about 30% of misperception camouflages designed for humans. Some of the people I work with are red-green colorblind, but more are very rare all-color blindness. They are almost immune to the vast majority of false perception camouflages. According to our vision model, batches of monochromatic and color-shifting filters were made to protect both the troubled warriors and the fooled artificial intelligences. Let them pass through the smoke, through the battlefield, to find the truth.

But the world is losing reality day by day. On the front lines, I’ve even seen entire buildings covered in fooling camouflage disguised as mountains. People didn’t do it to win the war, it was the fearful inhabitants who wanted to remove themselves from the world.

Recently, I’ve heard that misperception camouflages have been developed for colorblind people. The methods of fooling artificial intelligence have been turned to fooling the human eye, developing general cognitive scramble patterns that can outperform filter devices. Many scientists have taken over the research work left by the doctor, and they are as worried as I am.

The doctor’s words still echoed in our ears.

It has to be fast, he said.

We have enemies, and they are as smart as we are.