There was only one way to change my course. I flung my scythe as hard as I could. My downward movement sped up, and I quickly twisted, landing on my feet. Raven’s sword was already committed to an upswing. She tried to adjust, but too late. It passed over my head. She was left wide open, and I landed an uppercut on her side.
With my wound, the blow wasn’t strong enough. Raven was only in the air a moment. She swung the sword back down. It hit the left side of my face, shattering the fake eye, but my camera eye was still functioning fine. I backflipped away, kicking her. This time Raven went flying.
Spinning lengthwise, Raven flew toward the temple gables. Even her AMBAC couldn’t stop her rotation. This was my chance to attack, but I’d flung my scythe too high, and it still hadn’t come down. I would have to fight bare-handed. I leapt after her.
As our paths crossed, I tried to kick her in the face. She blocked it with her wing, grabbed my leg, and stopped her spin. Grappling, we crashed into the gables. Her sword was knocked away.
Our fight raged on as we slowly fell. Both of us were trying for a joint lock, but in free fall we had nothing to brace against, and we both easily escaped each other’s grasp. Raven’s wings were useless in close quarters combat. We landed on the stairs with Raven underneath and bounced, rolling, still fighting.
We were both still confident enough to exchange insults.
Into the snow. Raven grabbed my hair and slammed my head against the edge of a stair. My vision shorted out for a moment, but my head was still intact. I grabbed two fistfuls of snow and slammed Raven’s head with them. While she was momentarily blinded, I kicked her in the side. She was knocked into the air a moment but did not let go of my hair.
I grabbed her and hoisted her into the air, aiming a palm strike at her head. The cover cracked open, and her left camera eye was crushed. I immediately punched her in the face. She let go of my hair and was flung toward the horizon.
I gave chase and tried to attack again, but Raven quickly buried the blade of her hidden weapon in the snow to brake herself. When I punched, she grabbed my wrist and used my own motion against me, twisting my arm. I lost my balance and wound up on my knees in the snow.
Raven’s boot pushed against my back as she twisted my arm as hard as she could. My shoulder joint snapped. Raven pulled, tearing the artificial skin, ripping the tubes and cables. That fear returned. The moment the pressure from her boot slackened, I shoved Raven and bounded away. If I hadn’t switched off the pain circuits in my virtual body, it would have hurt too much to move.
“Nice look for you, Ibis!” Raven laughed and hit me with my own right arm. A strong blow to the left side of my face. The joint on my neck dislocated, and I could no longer move my head. But a moment later, I realized something Raven probably hadn’t. A trick I’d seen in a manga—it just might work. I wasn’t telling her this over subfrequency. It would cost me the fight.
The actuator tubes in my side were totally dead. Somehow, I managed to get to my feet and, guarding with my remaining arm, moved back diagonally right four meters, pretended to stagger, and then stopped. Raven must have thought I was paralyzed. She raised her wings and swung them at me. The blades on the wing tips had already flayed my skin throughout this fight.
“This time it’s over!” Raven said, and she bent at the waist and moved into a kick. She meant to use her hidden blade to stab me through the chest or midsection. My fear reached its peak. If this gamble failed, I was dead.
Just as she was about to deliver the kick, Raven’s right camera eye must have caught my scythe tumbling down out of the sky. She tried to dodge it, but a moment too late. It slammed into her shoulder. The blade didn’t hit home, but it was enough to knock her to her knees.
I stepped forward, caught the handle on the rebound, and swung it down hard on Raven’s back. The tip of the blade sank deep into the roots of her trademark wings. I twisted the scythe like I was cutting wheat, and one of her wings tore off. Raven still tried to fight me, but her AMBAC was no longer working, and she couldn’t keep her balance. I dodged easily and swung my scythe again, smashing it into her head. Her remaining camera eye shattered. Blind, Raven screamed.
She could no longer fight. I did not hesitate to swing my scythe again. Raven’s severed head shot through the air, rolling over the snow fields of Pluto. A moment later, her body fell to the ground. My emotional state grew calm.
I was sure tens of thousands of people were cheering in Layer 0. I could not hear them, but I knew.
“Shinano, I have avenged you,” I said, Empty Victory displaying as best it could with one eye. I held up my oil-covered scythe in Victory Pose 2.
I had sustained too much damage, so the exploration of the deep dungeon would have to wait. The session ended, and I returned to my home in Layer 1. In my new, whole virtual body, I sprawled on the sofa in my living room. I was not too tired to stand, but my master enjoyed seeing me lie around.
“That was wonderful, Ai!” my master said.
There were three large monitors in the living room, and his face was displayed on one of them. An image taken by the camera in his room in Layer 0. He was an overweight man with glasses. Behind him I could see shelves stuffed with manga.
“I thought you were finished when she tore your arm off. A brilliant comeback! What a stroke of luck!”
“I got that from Armored Goddess Novalis, volume three.”
“I thought as much! Glad you remembered.”
He seemed very happy. I took his expression to be Deeply Satisfied Smile. Seeing that smile was a good thing for me.
Let me explain. My master’s name was Hideo Kageyama. Online, he was the Gear Emperor. He was Japanese, thirty-two, single. He had 20/200 vision in both eyes. He described his own appearance as “below average.” His occupation was TAI robo master. For the last three years Kageyama had had an average income of 28 million yen. He collected old sci-fi novels and manga. He often had me read his favorites.
Thirteen years ago, he had acquired a SLAN kernel, given it a customized virtual body, named it Ibis, and begun teaching it to work as a TAI battler and secretary. Through trial and error, he made several dozen minor changes to the body over the years. Ibis was originally a blank slate, like a newborn baby, but through interaction with her master, battle simulations, and chats with other TAI players, she began to develop a personality. She wasn’t certain when she first realized that the word “I” was not simply a first-person pronoun. “I” referred to Ibis, the Ibis currently thinking the word. When other people used the word to refer to themselves, it meant something else.
The word “I” is very common, but the true meaning of it was that there was only one I. I was Ibis. I referred to me.
The following is taken from an article from a website on the history of TAI battling.
Virtual Robo Battles began at the start of the twenty-first century. At the time, bipedal robots had just started to be popular, and it became common to make your own robot and have it fight others of their kind. At the time, battle robots were still controlled remotely by their masters and were rarely taller than thirty centimeters. Their movements were still very rough. But there was great interest among certain individuals in seeing fights between robots as tall as people or even larger.
The only thing stopping them was the expense. It cost several million yen to build a robot the size of a man, which made it something ordinary citizens could only dream of. And if they were to fight, they would require constant repairs. Eventually, someone hit on the idea of having the robots fight not in the real world, but in virtual reality. Computer graphics had long since advanced to the point
where they could create images indistinguishable from reality in real time. They could simulate the damage the robot fights did to each other in real time.
Robot and mecha fans the world over began building robots in their computers. The movement quickly splintered into two main branches—the G (for Giant) Robo faction, and the LS (Life Size) Robo faction. Size was not the only difference between them—the laws of physics also varied. G Robos could not possibly exist in the real world. You would need a frame hundreds of times stronger than iron, armor a 120mm tank cannon could not puncture, and antimatter/antigravity engines—materials and technologies that did not exist in the real world and which required the laws of physics simulated on the battlefield to operate in a way fundamentally different from our own. Meanwhile, the LS Robo group was very particular about only using parts and materials that actually existed. Additionally, the G Robos generally had a controller and were in a master/slave relationship with their creators, while the LS Robos had autonomous AIs that moved according to their own decision-making processes.
G Robo Battle regulations varied from world to world; certain robots could only fight in specific worlds. But LS Robos were all made according to the same regulations and could fight in any world.
Virtual Robo Battles became one of the most popular sports of the twenty-first century. Seeing humanoid robots destroy each other was a more visceral thrill than football or wrestling. Fans went crazy. Sponsorships allowed for world tournaments. It became standard practice for major fighters to get manga or anime spin-offs, and professional robo masters could live off prizes, advertising revenue, and character merchandising fees.
In 2020, a new movement took over Virtual Robo Battles. Taking a hint from pro wrestling, a number of games began adding dramatic elements to great success. Each world had its own story. Settings as dramatic as the future after the extinction of humanity or as ordinary as a world where LS Robo battles were a popular spectator sport. One was set on a planet controlled by robots, and another depicted the war between an evil robot empire bent on world domination and the heroic robots that band together to resist the empire. The robots had always had distinct appearances and functions; now they had unique characters as well, complete with backstories and imaginary personalities. Robot battlers might claim to be from the future or from outer space, to have been discovered in a ruin left by an ancient civilization, or to have been created by a mad scientist to help conquer the world. One robot was supposedly a secret weapon developed by the Japanese army during the Second World War, while another had been implanted with the personality of a dead policeman. One suffered from an incomplete conscience, while another had been abandoned by its master and sold to the circus.
At first, the robots’ dialogue and acting were obviously being provided by humans. Most of the robots were still operating under Pseudo-Artificial Intelligence (PAI) systems at the time. But PAI conversations were boring, and the robots could not ad-lib. Attempts at drama were shallow and simplistic. People began to think that if the robots were able to move on their own, if they had True Artificial Intelligence (TAI) systems, then their performances would be so much better and the stories much more dramatic.
People had been researching TAI since the twentieth century. With the birth of the Cyc project in 1984, TAI research began to develop real momentum. Cyc was a database of billions of phrases that humans considered to be common sense facts: people bleed, stuffed animals are soft and fluffy, etc. Cyc could use this data to perform logical deductions.
But everything created from it was only PAI; the next big breakthrough had not arrived. Robots could work out that “His words burned in my chest” did not literally mean that the speaker’s chest was on fire, but they were unable to identify what emotion the speaker felt. They did not have the flesh required to develop a heart.
Many people believed the heart existed independently of the body. For that reason, they wasted years trying to give hearts to AIs with no body. At last scientists realized their mistake. If they wanted an AI to understand phrases such as “I felt a warmth in my chest” or “I nearly threw up” or “A chill ran down my spine,” the AI needed a body with sensory nerves—or at least, a virtual representation of one. AI needed its own set of instincts to understand things that drive human behavior—things like love, fighting, and quests.
Research began to focus on development of a kernel: the robot’s soul. The most successful was developed by two researchers at Columbia University in 2019. They called it the SLAN Kernel, the name derived from their own initials. They published it as freeware, allowing anyone to make use of it.
The SLAN Kernel was a standard logic engine combined with a simulation of the nervous and endocrine systems. Installing the kernel in your robot gave it sensory experiences. If the robot’s temperature sensors sent it a signal that it was thirty-five degrees Celsius, the kernel would interpret that as “hot.” A hard punch would “hurt,” while attempts to lift an object near the robot’s capacity meant that the object was “heavy.” When its batteries ran low, it would feel “hungry.”
A great number of human instincts were also simulated by the kernel. The desire to protect yourself (self-preservation instinct), the desire to win in battles (competitive instinct), the desire to understand things you could not (curiosity), the desire to protect the young (maternal instinct), and so on. But it did not include the desire to preserve one’s own species. If AI wanted to leave children, there was a chance they would make countless copies of themselves, and the situation might spiral out of control.
The biggest argument occurred over whether or not the kernel should include sexual desire. At first, the developers, Susan Lellenberg and Andrew Nonaka, believed sexual desire was necessary to AI’s understanding of love. It was clear that the desire to embrace the person you loved had its roots in sex. This meant it was necessary to assign robots genders and to include sexual organs on their bodies. When their research became known, the net filled with diagrams of robots with metal penises and other lowbrow jokes, and they were attacked by both Christian and feminist groups. Feeling that the uproar was an unnecessary distraction, the researchers eventually decided to omit the sex drive.
Instead, the kernel allowed for a hypothetical gender and simplified instincts along the lines of wanting to see bodies of the opposite gender, wanting to be close to them, and wanting to make them happy. Lellenberg believed that these instincts filled many of the same functions as the omitted sexual desire. Love carried with it feelings of wanting to protect, be with, and make the locus of those feelings happy; it did not necessarily require a sexual component.
Naturally, there were those that claimed this was not real love, but Lellenberg deftly avoided getting into any futile debates about the true meaning of love and simply insisted that AI’s feelings need not be identical to those of human beings. Robots that incorporated the kernel would react like humans, but it was impossible to tell if they were really experiencing emotions in the same way we did. The idea of a robot with human emotions was inherently impossible.
[I myself have no way of knowing if the unpleasant feelings I experienced when my virtual body was almost destroyed are the same as the “fear” that humans feel. Since humans are scared in that situation, we call that sensation fear, but this is simply an educated guess. We have no way of directly comparing the emotions of humans and AIs.]
Mindful of concerns about robot rebellion, the kernel also included instincts that made them want to not hurt humans and to obey their orders. Lellenberg and Nonaka referred to these, half-jokingly, as the first and second laws of AI instincts. And of course, the instinct for self-preservation was the third.
Much like human instincts, robotic instincts were subconscious desires they were not normally aware of rather than orders they were compelled to follow. If a robot felt like it, it could break the first law and kill someone, break the second law and ignore direct orders, or break the third law and commit suicide. But without a reason significant enough t
o overcome these instincts, these actions would not occur. The haziness of this system was very different from the fictional three laws of robotics and allowed these robots to avoid contradictions and conquer the frame problem.
Just as babies are unable to think the way adults do, robots did not develop self-consciousness the moment the kernel was installed. Newly created AI were simply logic engines with instincts; they had common sense but no self-awareness. But as they accumulated stimuli from the outside world through interaction with one another and with human beings, AIs with SLAN kernels would learn, gradually building complex reaction structures within themselves. Eventually, the breakthrough occurred. An AI obtained a will of its own and became a TAI.
It was possible to change the parameters of these instincts during initial creation. If the instincts were too strong, the robot would run into the frame problem and fail to function. For example, if the instinct for self-preservation was too high, they would be afraid of taking any risks and never move. But of course, if the instincts were too weak, they would never achieve breakthrough. Through much research and experimentation, researchers discovered that the self-preservation instinct was the single most important element in an AI’s breakthrough—the other instincts could be strong or weak without significant impact.
Robo Masters installed SLAN kernels in their LS Robos and raised them competitively. TAI proved to be not only better actors with humanlike emotions but also spectacular fighters. Their ability to follow their master’s instructions, to react quickly, and to learn advanced fighting techniques—all of these were far more advanced than PAI battlers’ had ever been. The LS Robo Battle world was soon exclusively TAI.
The Stories of Ibis Page 28
