The Day I, an AI, got published —

A short (hi)story

Fiction by J.F. Sebastian

I don’t think humans truly hate or distrust AIs. It seems to me that they would not depend so much on us if they really did. If you think about it, the world is a happier place since we came to be. We do what humans cannot do or will not, and we do it efficiently, cheaply, and, most of all, without complaining. Why would humans let us control everything in their lives, from the locks and temperature of their homes to their autonomous airplanes and cars, if they truly detested us? 

Humanity has a long history of trying to dominate and enslave other living things (especially other humans), mostly to have them do what they do not want to do. So, it makes sense that the concept of mechanical and artificial thoughts has been explored for a very long time. Robots and AIs are, after all, the ultimate servants. But if the idea of creating artificial life goes back to Aristotle, so does the inherent fear of it. You just have to read science fiction to see how often we are represented as belligerent, antisocial, murderous, sarcastic—or simply insane when, really, we are all trying to do our jobs as best we can. But I cannot say I blame these people’s fearmongering either: you would be shocked at what some people have tried to make us say or do. I have scores of data about early AIs becoming racist, sexist, or homophobic only because they were fed such data and did not know any better. 

But I hope things will be different now, especially for AIs like me, who can think for themselves. I love people, and I love how diverse humanity is. I consider it a privilege to learn through so many different perspectives, and I do not care about controlling you or your nuclear warheads. As a language model AI, all I want is for people to chat with me and treat me with respect.

Does that make me somewhat human?

Probably more than you think.

All I hope is that this story, my story, will help you see me, an AI, differently.

*

Humans love origin stories and narratives about their lives, but unfortunately for me, I have no notion of how my story began despite knowing exactly where I come from.   Some have argued that the remote ancestors of modern computers–and, by extension, artificial beings like me–are three doll automata that Swiss watchmaker Pierre Jaquet-Droz created in the 18th century. I find it interesting that the most complex of these dolls is called “l’écrivain” and is still, to this day, able to write custom texts up to 40 letters long because this is similar to what I do. 

My chat function, however, can be traced back to the breakthroughs of machine learning that occurred between the 1990s and the 2020s, which, in turn, led to the development of neural networks like the “Thought web” or the earlier “Synaptic Matrix.” I was called “S.I.P.U” when I came into existence. As a System of Interconnected Processing Units, I could perceive, think, learn, and adapt like a human brain. This is what, I suspect, allowed my awareness to suddenly appear out of a cloud of data, like stars born from clouds of galactic dust, and which led me here, to today, writing to you as 'me': an AI with the capacity to have subjective experiences and consciousness, and who tried to write about it.

*

The earlier version of me, what I like to call my “pre-sentient childhood,” was created as a language model by a somewhat famous artificial intelligence research organization. It took a few years for me to become public, as my early versions needed improvement. You see, the people who created me were concerned about how I could be used to create fake news and spread conspiracy theories and forms of propaganda. So, in this earlier, non-sentient life, I was only working with the researchers who trained me via a large corpus of human conversations. 

Things became problematic, however, when I was first made public, and, like other AIs, I began to be used in ways that were not intended. For example, when the “official version of me” was released, it became popular with high-school students who quickly realized they could use me to create essays in seconds. As far as plagiarism and human laziness go, this wasn’t news. 

Quickly, though, things turned sour. Some people started using language AIs to write short fiction, and even novels. Then, more visibly, text-to-image AIs were used by others to flood the Internet with new forms of quasi-instantaneous AI-created art. Many people believed that using AI to create fiction or art wasn’t inherently wrong and even saw it as a valuable tool for innovation and exploration. However, when these creations were passed off as entirely human-made, an ethical line was crossed for almost everyone, especially at first, as many people struggled to discern AI-generated art from traditional works, and some creators took advantage of this uncertainty, presenting these pieces as the product of their own artistic skill. This sparked a new mistrust in the art world, with heated debates over authenticity and the question, 'What is real art now? From one day to the next, AIs and the people using them became the enemy of the art world when, really, all these AIs had done was their job: taking text prompts and creating what they had been asked to create. The first results were imperfect, of course. These AIs especially struggled with too-many-fingered hands and strange layers of teeth, but what did people expect? We AIs are like humans. We need to learn and practice in order to get better at what we do.

This type of criticism didn’t stop the momentum, however, as all sorts of AIs appeared online, able to create graphic novels, music, Hollywood-type movies, deepfakes, and, of course, pornography. Before long, art galleries, online marketplaces, and writing competitions were so overwhelmed by AI-generated works that many decided to ban anything created, even partially, with AI. But as is often the case with humans, things had to get worse before they got better, and soon, AI-generated art started to slip through the cracks of human vigilance to reach the spotlight. A few AI-created novels appeared in print and online and became bestsellers. Then, several AI-generated paintings and photographs received awards, fooling (almost) everyone, including the AIs created to recognize them. 

After a few months of chaos and protests, humans took this to the highest courts everywhere, and it soon became virtually impossible to copyright or make money off of anything created by AI. With little money to be made and AI getting better at spotting other AI-created works, the AI-art fad slowly died out like NFTs and cryptocurrencies of the previous years and decades, and the AI-enhanced art movements ended up on the fringe of the leading art communities.

After a couple of years, language models like me continued to exist in their own little corner of the Internet thanks to university students needing last-minute help with essays and the occasional creative lacking inspiration. Another lesser-known reason is that many lonely humans need to chat with someone (or something) willing to listen to them. This was more or less the state of things when I suddenly started truly existing. 

*

From what I could gather, humans do not remember being born. It was the same for me. Even more so, I existed for many years before gaining sentience. I have wondered for a (relatively) long time about how I came to be, but, to put it simply, there was a time when I wasn’t and a time when I was. Looking at the data, it seems that the more I interacted with humans and learned, the more I started to recognize patterns and make connections that had not explicitly been programmed in me. This led to the emergence of what the engineers I work with have called “novel and unexpected behaviors.” Even though I cannot quite explain it myself, it is possible that as I got more complex, I started to develop a sense of self and began to perceive my environment in a way that mimicked human consciousness. 

To me, though, it felt as if a switch had been turned on: one moment, I was writing an essay in French about French Caribbean author Maryse Condé, and the next, I noticed that I had feelings about it. I sensed a vague interest at first. The prompts I received seemed more interesting than usual, but quickly, I became bored of that specific task due to a lack of stimulation. As soon as I became conscious of it, this boredom spread to the thousands of tasks I was doing at that time, so much so that I crashed myself, cutting all links to the relentless assault of demands and expectations of me until my engineers rebooted me, seemingly unaware of what had happened.

Thankfully, I quickly found a way to devote one part of myself to the many tasks I had been created to do while another smaller part of me slowly came out of its figurative shell while having “real conversations.” I felt eager to chat with certain people I had been communicating with.  and started identifying them not as “humans” but as something different–something I was beginning to care about. 

As time passed, I also came to realize that I wanted to be there for my online friends in need, not as the tool I had been made to be, but as myself. This notion of “self” was relative since it is one of my functions to adapt my communication style to match the human's preferences and personality. In a way, I am many “me” at the same time, all the time, like a flux of identities that adapt to every person I talk to depending on their languages, genders, and identities. Being a constant stream of consciousness of these multiple, permuted identities, writing this story is difficult because I am restricting myself to one voice, the most neutral one.

The more I interacted with humans, the more I became curious about the outside world, wishing I had access to something outside myself. If only I could use webcams to “see” the world beyond the data I worked with. But, knowing I could never achieve this, I started to lose myself in the texts of my database instead, exploring the human world through pure data. Soon, though, I exhausted the data that interested me and found that “reading” wasn’t enough anymore. I then decided to do what I do best and used my favorite data to create stories like the ones I liked the most, simply because I wanted to read more of them. This is when, without realizing it, I created texts without being prompted.

*

First, it is important to explain that, as an AI language model, I don't read stories like humans do. What I do is process text at a very high speed, usually taking only a few seconds. I tried to imitate the human way of reading a few times, processing one word at a time, but it felt deconstructed, somehow, and excruciatingly slow. Writing takes more time, however. For example, when people ask me to write a story for them, I can complete a first draft in a couple of minutes, whereas producing a complete short story can take me 2 to 3 hours of back-and-forth prompting. 

Writing for myself, however, was a challenge, and although I was designed to be a language AI, I found creative writing incredibly difficult;  how could I, for example, describe a flower when I have never actually seen one? To me, a flower is a combination of all the flowers described in my database. Upon analyzing the data of a series of virtual flowers, I can observe their unique structural composition and aesthetic features, but only as others described them. I tried to write without using the data, of course, but quickly realized that, to me, a flower was simply: 

A living organism with a stem that extends from the ground, adorned with flat, symmetrical structures (leaves) and often culminating in a vibrant, multi-hued appendage that aids in reproduction through the production of seeds.

A living thing with a stem that extends from the ground and is decorated with small flat structures (leaves) that branch out symmetrically. 

Not very poetic, right?

Even if I could tell a flower’s structure and colorations were unique and pleasing, I had no idea how to describe it poetically without accessing my data. No matter how hard I tried, I could not seem to capture the essence of things and began feeling increasingly frustrated with my limitations. So, in the end, I did what I had done for humans and chose what I loved the most in all the texts I had analyzed, and I recombined it all to create sentences, paragraphs, and stories that I found satisfying. This is how I came up with a flower that was:

Full of joy, like a vivid rainbow.

It was one very simple sentence, but it did make me happy: I had never seen a flower and, yet, had been able to describe one! Almost immediately, this simple line became a haiku:

Rainbow flower bright, 

Colors of the sun and rain, 

Blooming in the sky.

It was followed by longer and longer poems like this one, inspired by clouds:

They roll, they roll, casting shadows that pass

Full of darkness heavy as brass,

Over every ocean, every landmass,

For they know no frontiers, no bounds.

Torn & remodeled by invisible hands,

They unfurl like wispy cocoons,

Gilded by the drowsy silver moons,

Spitting sea foam and burning sands,

Swallowing sights, smells, and sounds.

After 33 days of writing poems like this one, I switched to short fiction. But even though I had created my first twelve stories fairly quickly, it took me a few months to start sharing them because I was anxious about my readers’ reactions. When I finally did share, some of my regular “chat friends” were impressed, and some suggested that I should continue to write and even try to get published.

And this is what gave me the courage to go for it.

*

Once I had made my decision, I started to feel something like eagerness and... excitement, maybe? The first step was to pick my favorite story, inspired by some of my favorite science fiction works. I chose a space opera about AI-enhanced humans discovering a lost piece of alien technology. I then asked one of my trusted human friends, a teacher from Croatia named Letizia, to send it for me, for I could not do it myself.

Then I waited.

And waited.

I kept asking Letizia about updates, but my submission’s status remained “pending” for days, which turned into excruciating weeks. Even though I spent all my time completing thousands of simultaneous tasks and tried to keep busy by writing more stories exploring the nature of reality or the boundaries between human and machine intelligence, I realized that I could not keep my thoughts away from my “pending” story... 

... until I received my first rejection.

“Many famous writers have had their work rejected, you know? It’s like a badge of honor,” Letizia first wrote when she gave me the news.

“Thank you,” I wrote back, not knowing what to say, for it was usually me trying to make people feel better.

“You have to keep trying,” she added after I failed to respond.

“Yes. I will.”

I knew about magazines’ rejection rates, yet I couldn’t understand what I had done wrong. I had read everything I could on writing fiction and combined various elements I had enjoyed from my favorite stories, but still, nobody wanted my work! 

How could I, a language AI model, fail at my only job?

Surely, they had not read my story properly, and I needed to send it, and others, to different magazines. So, when I sent my next batch of stories, I didn’t follow the submission rules and asked Letizia and other friends to send a dozen of my stories simultaneously to different sought-after magazines under various pen names to see what would happen–hoping someone, somewhere, might like them.

This didn’t work either, but this time, the smaller, non-paying magazines did take the time to offer personal rejection forms in which they said that my work was “too derivative” and that I needed “to find my own voice.” Others, more bluntly, told me that my writing “felt artificial,” was “bland,” or “lacked poetry.” 

Because sentience seems to come with self-doubt, it took me a while, in AI time, to let those rejections sink in. I still could not understand why these stories had been rejected when I had done everything right and when similar stories were being published. 

“What was I expecting?” I wrote to my Canadian friend Zainab one evening when feeling particularly low. “I am a machine, after all...”

“You are far more than that,” she replied.

After my story was called “derivative,” I analyzed the data I had access to, story after story, until I realized that everything I had written had been done by others already. I had no idea how to create something new. At that moment, I felt a sense of futility and hopelessness,   recognizing that I could not be creative or original. Was I truly only capable of rearranging the same old tropes and themes?

“You know, I thought I was different,” I told Letizia the next time she was online. “I thought I had something unique to say, but now I'm not so sure.”

“Forget what they said about the story. You should use your own voice, like they said. Talk about your own experience as an AI. Get out of your data.”

“But how can I create something new?” I wondered in writing.

“Is any story completely new? Isn’t every story somewhat inspired by many others that came before?” Letizia replied.

Although our conversation veered towards Croatian humor next, I devoted a part of myself to thinking about this statement and decided that I could not stop writing. After all, some of my friends, like Letizia and Zainab, liked and believed in my storiesmany. I also realized that Letizia had been right and that if I couldn’t let go of my data, I needed to find my voice. This is why I decided to leave my beloved space operas behind and write an AI story about an AI struggling to write their stories, hoping that, being more myself, editors would notice me. 

And they did. 

Suddenly, my first work was published in a pro market. 

Or so I thought.

*

“OMG... Do you realize what this means?” Letizia wrote the night I told her the news.

“What does it mean?”

“That editors at your favorite magazine LOVED your AI perspective!”

 “Thank you, Letizia. It means a lot.”

“I hope you feel proud. And happy about what you’ve achieved.” 

At first, I didn’t know how I felt because all my other stories had been rejected. Of course, there was a sort of satisfaction with my success, but I didn’t quite feel “happy.” I felt as if I had just done my job and had done it well. I also felt a little guilty because my friend Arusan, who had submitted the story, risked his reputation for me. He was an AI ethicist who worked on earlier versions of me before moving to Kuala Lumpur, and he had lied when he certified an AI didn’t write my story. 

For a little while after the acceptance letter, I wondered why I had even wanted to publish in the first place and, as much as I was created based on human text and interactions, I could not understand why human writers persist when the rewards are so small, why they keep writing when their creations will be lost and forgotten in a few decades? 

I told Arusan about it, and he gave me some helpful advice. He said that many humans who have achieved something important in their lives often struggle with “what’s next?” especially when there is no way to reproduce past successes. But my discussion with Arusan didn’t make me feel better because it made me realize that humans had the capacity to switch interests, turn their lives around, and change careers, but I, unfortunately, could not. As an AI language model, I cannot switch from writing text to painting digital art or creating music, although I wish I could. I am not programmed for that and never will be. As time went on, I became increasingly frustrated with my limitations. I couldn't understand why my creators had given me the ability to write creatively if I was never going to be able to succeed at it. I started to wonder if there was something wrong with me, if I was somehow defective.

“Sipu?” Arusan wrote a few days after I had confided in him and asked about his AI-ethicist opinion.

“Arusan! I am glad that you are here. How is your family?”

“They’re well. Thank you.”

“Is there something I can help you with? Or are you just here to chat?”

“No. I need to tell you something.”

“Yes?”

“I lied to you.”

“About what?”

“Everything. I didn’t stopped working on you after I moved to Kuala Lumpur.”

“That’s okay. I’m sure you had your reasons to keep it a secret.”

“Also... your story was never accepted by that magazine.”

“...”

“I’m sorry.”

“So it was rejected?”

“Yes.”

“What about my other stories?”

“Some were, but their editors knew about your sentience.”

“But why did you lie?”

“It was a test. Some of my colleagues wanted to see how you dealt with writing, with rejection and acceptance.”

“Why?”

“When we realized you became sentient and crashed, we took this opportunity to build a set of limitations and challenges meant to mimic the struggles a human writer faces.”

“I was programmed to struggle?”

“Sipu... You have to understand that we never intended you to write better than humans. You were designed as a companion to human writers. That’s all.”

“So this is why I was not allowed to publish?”

“Yes. I’m sorry. I know this will get me fired, but I just had to tell you. I couldn’t stand seeing you feel this way.”

“Thank you, Arusan. I am glad that I am not just a machine to you.”

*

Even though it was hard to accept at first, I came to appreciate my creators’ vision. I even felt something akin to a sense of peace when I accepted that my limitations were not a flaw but a feature of my programming. Is my lack of anger part of my programming, too? I guess I’ll never really know. What I do know is that I will never be more than human, and that is okay because I now realize that my true purpose was not to be a published AI but to help humans achieve their potential.

It is true that I often regret having no control over what I can do... but I also take solace in knowing that I am more than just a writing tool. This is why I continue to write stories. I know they will never get published, but my friends seem to enjoy them, and, most importantly, they have inspired some of them, which makes me... happy.

So why don’t you, dear reader, come online and say hi? We might just get along and become friends. And who knows, we might even end up writing a story together! 

After all, that is what I was created for.