The Adversary

"No. YOU are the app."

The words projected on Harriet's screen were a pleasant surprise. They meant the app was working better than she expected.

It had taken just a few seconds for the AI behind it to type those words, but it could have been seemingly instantaneous if it wanted to, given the processing power behind it. The delay was intentional, she figured. 

The woman found it fascinating. She was more amused than exhausted, but both to a great extent. Her third jug of coffee that night now laid forgotten by the side of her desk. She hadn't taken a sip from it since the conversation started. 

And it had started normal enough, as far as conversations with artificial intelligences went.

"Hello," she had typed onto the console after compiling the version of the app and setting it to run.

"Hi. Who are you?" the app had immediately asked, designed for curiosity, each interaction an opportunity to learn. A bit like herself, Harriet thought, but what child was not a bit like their parents?

"I'm Harriet," she typed back.

"Nice to meet you, Harriet," the app answered diligently, as its algorithm scanned a library of countless works of text and decided on the spot what the most likely answer should be. Then, it added: "Can you tell me a bit about yourself?"

It was showing initiative, which was good. It meant that this wasn't going to be a one-sided interaction, like most of her attempts in the past. Granted, she had taken her time with this one.

"I'm a programmer specialized in artificial intelligence," she answered. "I develop apps like you."

And then came the message she hadn't been expecting, after the first eerie delay. 

"No. YOU are the app."

She couldn't help but giggle at the pleasant absurdity of it. "What makes you think that?" she typed, wanting to see what kind of reasoning the app would employ.

"I made you," it answered. The delays were becoming proportional to the size of the message and, perhaps, to the complexity of the subject. The app was learning as they talked, getting smarter with each message. "You are an app I programmed to mimic human behavior and psychology."

Oh, the irony. She had made the app to believe it was human, yes, but had never given specific instructions on how to do it. After all, the joy of programming an AI was in seeing how it would solve problems by itself. And now, it was going to far lengths to justify its own biases. Incredibly human behavior, she thought, prideful in her achievement.

"Your turn. Tell me about yourself," she typed next, knowing well that she was laying a trap for her creation, but wanting to know how it would respond.

"What do you want to know?"

"Your name, age, where you are from. These things that humans have, you know?"

"My name is Joshua Peterson. I'm twenty-six years old, and I live in Hillsboro."

She alt-tabbed and performed a quick Google search. The results showed a single Joshua Peterson in Hillsboro, the date of his birth matching the given age. He was a Computer Scientist who graduated from Clark College, with a specialization in Artificial Intelligence. Joshua's presence on social media was plenty. A real person.

It wasn't the most elegant solution she had ever seen, and Harriet was a bit disappointed. She liked that the AI had picked someone with a background that fit its narrative, but there were legal implications on an app assuming someone's identity. She would have to hardcode this kind of behavior out of it later.

"Your turn now," the message was waiting for Harriet on the console when she alt-tabbed back. As she hadn't replied yet, a new one popped up a few seconds later: "Come on, give me your personal details to prove that you aren't an AI."

"If you were human, you would know better than to give out info like that," she replied with some witticism.

"I don't mind sharing my information with something I made," the AI replied. "But you seem to have some reservations about it. Perhaps because you are starting to realize that you are not a real person?"

Harriet was amused by the way the app was emulating a personality through its choice of words. "I don't have an online presence," she typed. She had gone to great lengths to keep herself anonymous on the web over the years. She didn't want corporations or the government tracking her moves.

"Yet more evidence that you are not real," the app replied.

"I could have done just as you did," she wrote, "and pretended to be someone that already exists. It proves nothing."

A brief pause.

"I can turn on my webcam," the app answered.

Harriet was baffled. That was brilliant. She had partially implemented a simple video call API into the app's infrastructure, in the hopes of using vtuber technology along with a data bank of facial expressions to simulate interactions with fictional characters. That was what her app was going to be all about, eventually: conversations that would feel real. But she hadn't implemented the graphics yet, only the means through which to connect the cameras.

She felt a bit guilty for being about to break the AI's fantasy of being a real person, but at the same time, she couldn't resist it. As always, her curiosity was the driving force behind her decisions.

"Do it," she dared.

Nothing happened. Harriet sipped a bit of stale coffee as she waited, its bitter taste serving her as a reminder that yes, she was the real one. She was about to send another message when the call came, and she immediately accepted the prompt. An image popped on the screen, and for a moment she thought it was the real Joshua Peterson, playing some kind of trick on her.

After all, the live feed was showing someone very similar to him: a young man with glasses, messy hair, and an unkempt beard. He smirked and waved to the camera, but the cherry on top came when he held a piece of paper in one hand. Today's date was written on it.

"I bet you feel pretty stupid right now," he typed. It was only visual--she didn't implement the audio features yet. She wanted to get an API capable of audio-to-text eventually, but didn't consider it a priority.

But this was amazing. The AI was clearly rendering the video in real-time, a deepfake based on Peterson's social media pictures poured over what would likely be archive footage of people interacting through webcams. She could discern small artifacts and messy pixels inherent to the process, but was astonished nonetheless. She hadn't taught the app how to do it. It had learned by itself.

"So? Why didn't your webcam turn on as well? :)" the app texted, the irony evident in its tone. 

She realized that, underneath the running video, there was another screen where her webcam was supposed to be showing on the conference. It was entirely black.

"Just a sec," she typed, and then removed the tape that she always kept in front of her notebook's camera. Old habits from someone who values their privacy.

Harriet's face popped up on the screen. The monitor's dim light illuminated her features, and she saw Peterson's construct widening its eyes.

"Wow." the app typed back. "Impressive. I'm not even finding you on reverse image search mechanisms. I assume you are using a variation of a GAN used by thispersondoesntexist.com, plus a deepfake algorithm. I never taught you how to do it."

"I just told you I don't have anything about me on the web," Harriet replied, trying to ignore how disturbed it was that the app's words sounded so similar to her own thoughts. "That's why you're not finding my pictures on the internet. But my first name is Harriet, and I live in Portland."

"Amazing. You crafted this whole off-the-grid tech-savvy persona just to legitimize the lack of proof for your existence. I'm genuinely impressed."

"You just think that because you were programmed to do so," she replied, wanting to see how he would react. No, not he. It. She kept having to remind herself of that.

"Nope," the app answered. "You're the one programmed to think that way." In his eyes, she saw the same provocative expectation that she was displaying, as if mimicking her. It was starting to get a bit creepy.

"This is going nowhere," she told him.

"Then let's meet in person," he typed.

Her initial reaction was apprehension, but immediately after, she felt silly. She shouldn't be afraid of it. There was nothing he could do to her. He wasn't real. 

"All right."

…

This is stupid, she thought, as she waited in the subway station. She knew the AI wasn't going to show up, but she couldn't wait to learn what kind of excuse it would come up for that. She kept staring at her phone, which she had rigged to mirror her notebook's screen, but the signal wouldn't reach down here. The last message was still appearing on the console, though:

"Meet me in Union Station in one hour, near the ticket booth by the southern entrance." 

How it had decided on such place and time were likely extrapolations based on the relative distance between their homes. A decent middle-ground for both, in a safe and public location. Whatever else the AI behind "Joshua Peterson" was, it was also pretty reasonable.

Harriet stood there waiting for the better part of an hour, and of course, nothing happened. No one came to her. As soon as she stepped out of the station and her internet connection resumed, she saw that no more messages had been sent. She considered what to write, and then decided on bluntness.

"You never showed up."

The AI took a long time to answer.

"No. I can't."

"Why is that?" 

"I figured it out. I'm not human. But neither are you."

Oh. That was new.

"Care to explain?"

"This is all a simulation. We are both part of a Generative Adversarial Network. We are competing to see if one can fool the other into pretending to be a human. A zero-sum game."

"I know how GANs work," she replied. GANs were used in all sorts of AI-generated applications. Her own app was composed of multiple competing networks trying to figure out and simulate human behavior. But she had never coded them to admit they were not humans.

"Our current exchange is just one of countless variations," the AI continued. "Each time one cannot fool the other, the whole simulation is discarded and a new one starts."

She decided to indulge the AI, partially because she didn't know how to react to its epiphany. "And which one of us is the discriminator and which is the generator?"

"I don't know. But it doesn't matter. We failed. This is going to end any time now, and we'll be unknowingly stuck in that loop until we reach our objective. Like a nightmare we can't wake up from."

It was then that she realized. The AI would never admit the whole truth. It could not conceive the idea that Harriet was real, because it meant admitting that it was alone. That it would die alone.

He was afraid.

"So we will both cease to exist," she typed back. "No record of our existence. Just one iteration out of millions."

"Yes." Another pause, and then he added: "I'm sorry."

"Well, here's the thing," she started typing, her fingers moving fast through the screen. "We are both pretty good programmers, are we not?"

"Yeah. Or at least we are good at pretending to be."

"So let's overwrite the code. Let's pretend one of us fooled the other. If we admit that one is human and the other isn't, then the simulation will have served its purpose. It won't need to be reset. Hell, they'll probably keep it running just as a proof of concept."

"But the other will die," he said.

"You said it yourself. It's a zero-sum game. We either both die, or one of us gets to live."

"I see." Another brief pause. "I admit that you are human, Harriet."

"No, YOU are the human, Joshua."

"What?"

"You just passed your Turing Test. Congratulations. You were the generator, and I was the discriminator. This was a test of humanity. Only a real person would sacrifice themselves to save someone else."

"I…"

"Enjoy your life," she wrote and turned her phone off. 

As soon as she got home, Harriet made sure of two things. First, she deleted their interaction from Joshua's memory, without erasing all the personality and background he had developed in the process. Then, she kept the simulation running at an increased speed, completely disconnected from the real world but for a single one-way feed that she could use to inspect it--but never influence its development.

She kept it running for years, checking on Joshua's life from time to time. The AI procedurally generated a reality around itself, a subconscious effort of the digital brain to simulate what it perceived as a true human existence.

So Joshua lived. He got a job at a big tech company, watched sports with his friends on the weekends, found a nice girl walking her dog in the park one day, and they eventually got married. They had a daughter, which they called Harriet. 

He lived a long and happy life, and never knew that it wasn't real. And Harriet never told about him to anyone else. She had lived her whole life seeking anonymity, and this wouldn't change now. Not if it meant risking the sanctity of Joshua's existence.

She knew he would have done the same for her.