In the new era of artificially intelligent chatbots, the nature of AI as we know it has largely been centered around and defined by ChatGPT. Since its launch on November 30, 2022, ChatGPT has generated billions of new sentences and words and thoughts for the human race in the past year or so. The cultural response to ChatGPT has shaped it as somewhat of a savant, capable of enormous abilities in diction and education. However, the software is woefully inadequate in many of the creative tasks considered as the most unique ability our species has developed.
The Limits of ChatGPT
Among these tasks is the fairly crucial duty of original thought, one that has really defined the limits of the chatbot. ChatGPT is not truly intelligent in and of itself, rather, it predicts which string of words will best answer a given prompt, using the large amounts of data it as access to. This data entirely consists of language, art, and science created or discovered by humans.
Therefore, at its core, ChatGPT is no more than the sum of its parts or, rather, our parts. The software's every generated notion is a permutation, however stilted or eloquent, of an idea that's already been posed. Thus, it literally and truly cannot say something original.
Image Credit: Bertelli from Pexels.
But in its current state, ChatGPT is even more limited than this inherent definition. ChatGPT has prospered beyond belief in the areas everybody is familiar with.
It is able to seamlessly answer a query about nearly any academic or theoretical question with perfect clarity and eloquence, as well as a certain prolificacy that many of us work hard to achieve. Nevertheless, the large language model is still very clearly a robot, approaching most creative and emotional tasks with a decidedly unhuman outlook, often clarifying that it is unable to form the sentiments or opinions necessary in order to generate a certain output.
Since its input does, of course, comprise of human emotion, I am under the impression that ChatGPT is capable of growing to better perform these tasks, but I cannot confirm this nor say how quickly; I am no expert. Whether ChatGPT can improve is not what I want to explain.
I only wish to remark with humor and fervor that improvements are needed, and to expound on the profound inadequacies that I encountered during the summer of 2023, in an attempt to collaborate with the robot to write short-form fiction. In other words, writing a short story with ChatGPT is impossible, because it's just very bad at writing creatively.
The Setup
Though I do prefer the analytical genre (see my review of The Creator, which tackled artificial intelligence through a lens I am more familiar with, film), I had written fiction before, collaborating with other writers in secret by composing every other chapter, all without knowing each other's identity. The results were excellent, amalgams of two creative identities that formed to create excellent character pieces, often fleshing out the aspects of the story that were interesting to each of us.
I wrote these short stories (in my capacity as an intern) to test a platform called Switch-Lit. According to its About page, Switch-Lit is a "dedicated public space… where the forces of your imagination can freely meet and play with other writers," and is available for all to use in its beta form now.
I can't stress enough how much I loved the experience of collaborating with another human being. In the very early days of Switch-Lit, I had the wonderful opportunity to meet my collaborator on this off-beat, science fiction story about a lost father figure.
When I met her, I felt as if I already knew my partner, like we had already exchanged pleasantries in the small yet excellently oriented world we had created together - even though we knew not the other's exact age, gender, location, or any further signifiers. Likewise, once the site was up and running, a test run in the same format as the ChatGPT story led me to the same conclusion, a crime story about air conditioners and corrupt bosses that had so much more personality than the garbage that ChatGPT put out.
The First Chapters
Though my thoughts were placed with that crime story, unfortunately, my energy was in the story I was writing with ChatGPT. Due to Switch-Lit's nature of being paired with a writer in secret, when I read the robot's first chapter of the story we would be writing, I did not know that my partner was ChatGPT. I was unaware that its output was being pasted into the site for each chapter it wrote - all without any manipulation or “training” of the bot.
Since I originally thought I had a human for a writing partner, this made the experience of writing an unusual form of Turing test, a test of a machine's ability to exhibit behavior indistinguishable from that of a human. ChatGPT failed the Turing test almost immediately, responding to a prompt about people escaping a forest fire with the least interesting story setup possible. Below is what it wrote.
A wildfire, its fiery fury awakened, tore through the tranquil forest with an insatiable appetite. Alex and Mia, startled awake in their rustic cabin, dashed to their old pickup truck, seeking escape from the unfolding chaos. The forest, once a haven of serenity, now trembled under the merciless rage of the flames. The wildfire pursued them relentlessly, its tendrils reaching out like an unforgiving hand.
Just as they thought they might escape, a deafening roar echoed through the night. Ahead, the road ahead was blocked, swallowed by the ferocious flames. Trapped with no way out, Alex and Mia exchanged a glance of terror. The wildfire's wrath had ensnared them, and now, they were at its mercy.
My immediate impression of my “partner” was disdain, because the amount of material I had to work with was immediately minimal. The bot described the fire well, using interesting adjectives and the like, but provided absolutely no character for the two people it introduced, Alex and Mia.
It described Alex and Mia's tense escape from the fire without providing any description for who the two characters were, and ended the chapter with the assertion that they were “trapped with no way out” and could not possibly escape from the fire. I gave my partner the benefit of the doubt, thinking that maybe characters weren't their strong suit, but that they could easily develop the story with me if I gave them something to chew on.
Though I complained jokingly to my friend that the partner I'd been paired off wrote like an AI, I did not yet wonder how accurate that statement was. Instead, with the limited space I had (each chapter could be 300 words), I responded to ChatGPT's “Chapter 1: Ensnared by Fire” with “Chapter 2: Betrayed by Flame,” a shift in perspective to Joshua, who had been at the campsite as the fire broke out, reminiscing over a recent breakup with his boyfriend when he threw his lit cigarette to the ground and caused the forest fire.
The chapter, which I aimed to make as ridiculous and easy to respond to as possible, ended with Joshua noticing Alex and Mia's pickup truck in the distance, and recognizing it with horror, implying somewhat obviously that Alex had been Joshua's ex-boyfriend. This is what I wrote.
As dusk settled over the campsite, Joshua sat sadly at the crackling campfire. Laughter filled the air around him, but Joshua was alone. He had taken this day trip to get away from him, to have the lack of reception out in the forest separate him from checking his texts, pining over a boy who wasn't interested in him anyway.
At least, not anymore. But the boy was still there, lingering in his thoughts, and he didn't seem to be going away.
Just then, fire erupted beneath Joshua's feet, and Joshua leapt to them. Leg over leg hastily sprinted away from the flame. The 800 meter had been Joshua's specialty in high school, but he couldn't run as fast as the old pickup truck that shot by him a minute later.
Joshua squinted. It couldn't be the same truck. Could it?
The Mediocrity Continues
Since I assumed I was writing with a human, I thought my abrupt genre shift could inject some much-needed life into the story and its characters, but I thought wrong on both counts. I groaned in disbelief when I got back Chapter 3 and realized that my partner had missed the mark entirely, refusing to follow up on the romantic thread I had just installed, regressing Alex and Joshua's characters by completely ignoring the very heavy implications I had made as to their relationship.
By no means would the Alex/Joshua/Mia love triangle have been the most gripping thing in the world, but for the second time, I had nothing to work with.
Around this time, I began to seriously consider the possibility that I was writing with ChatGPT. The general story progression was one thing, but the exaggerated usage of adverbs and adjectives to accentuate every single sentence felt extremely automated and weird to me. Every interaction in “Chapter 3: Embers of Destiny” was nauseatingly artificial, with ChatGPT seemingly trying to set up a theme of fate through the dialogue for no reason.
It would throw in laughably out-of-place clichés, like finding a location where the trio could regroup and strategize, as if the forest fire was a malevolent force that would take genuine thought to permanently escape. ChatGPT seemed unaware they had easily sped through the flames, which itself was a contradiction (I thought it was impossible to escape, ChatGPT!), sentences earlier. Here's what the bot wrote.
Mia turned around, gazing at Joshua with a mix of relief and concern. "You're fortunate we stumbled upon you," she said. "It's almost as if fate brought us together."
Joshua managed a faint smile. "I'm not one to believe in fate," he replied, "but I'm genuinely grateful you both showed up when you did."
The truck's radio buzzed with updates about the unyielding wildfire, as authorities struggled to control its wrath.
"We must find a secure location," Alex said, his gaze fixed on the road ahead. "Somewhere we can regroup and strategize."
"I know a place," Joshua said, his voice carrying a glimmer of hope.
At this point, I accepted that whoever my partner was, they weren't going to be very fun to write with. Although I can't say I ever locked in my guess that I was writing with ChatGPT, my internal process when I received each chapter was to ignore the terrible and cringey lack of realistic human emotion, and continue to toy aimlessly with the laughably stiff characters ChatGPT and I had created.
My “Chapter 4: A Flickering Burn” wasn't the best, recasting the group's dynamic in an uncomfortable light in which none of the three really knew how to deal with the awkward situation. I set up Alex and Joshua's relationship as one that couldn't be easily mended, joking that they would be settling down at Joshua's mother's house, which I decided Alex was unhappy about.
The robot's “Chapter 5: Subtle Sparks” unsurprisingly ignored any attempt at a humorous setup I made regarding Joshua's mother, not even introducing the character, simply having the trio take refuge in her home without acknowledging her existence. ChatGPT, not understanding the concept of subtlety despite picking the term out for the chapter title, went full steam ahead on Alex and Joshua's relationship.
It portrayed the two as desperate to get back together, unaware (though this, at least, I can excuse) that we were only halfway through what was meant to be a 12-chapter story. This is what it wrote.
"You've changed a lot," Alex finally said, trying to bridge the gap between them.
Joshua smiled, a touch of nostalgia in his eyes. "You too. Life has a way of molding us."
A flicker of their past sparkled between them, but their words remained veiled, dancing around the feelings they had buried deep within.
"It's strange how life brings us back together like this," Alex said, trying to mask the longing in his voice.
The Story's End
The remaining chapters all blend together - “Chapter 6: Rekindling Bonds” is ChatGPT's response to my introducing Joshua's mother, Mrs. Wheeler. Despite my best efforts to have any conflict center around Mrs.
Wheeler, ChatGPT acknowledges her for all of two sentences, having her greet Alex and Mia and then discarding her character, trying to build the trio's camaraderie by having them “help with any repairs or cleanup needed around the house.” Apparently, Mrs. Wheeler was “happy to have young and able hands to lend a helping one in this trying time.” The latter phrase would be my personal pick for the most robotic line in the entire story.
Since ChatGPT had stripped the setting of Joshua's mother's house of any interest, I had the trio leave at her behest, with Joshua rejecting Alex's attempts at reconciliation due to his own self-hatred. I was merely curious if my partner would realize that human beings are complicated, and don't always blindly seek to “rekindle” (ChatGPT had grown fond of fire puns in the last few chapters) their relationships.
But as the characters moved from town to town, my robot friend continued to push the story towards “new beginnings" and insist that the story's protagonists had forged a notable camaraderie, despite the story lacking any interaction that would provide even minimal evidence for this claim. Here's an excerpt from ChatGPT's chapter.
Joshua's gaze lingered on Alex, and for a moment, their eyes locked. The unspoken words between them were palpable, a mixture of longing and regret.
As they drove away from Joshua's childhood home, a mixture of emotions swirled within him. The road ahead was uncertain, but he knew that the bonds they had forged amidst the flames were not easily extinguished. And perhaps, just perhaps, the flickers of a new beginning were waiting to be kindled.
After poking fun at the characters' lack of funds, comparing their apparently whimsical adventures to the Scooby-Doo gang, I had our four heroes (the fourth being Jessie, Mrs. Wheeler's Airbnb guest… like I said, I was bored) settle down in Fort Collins and get steady jobs. Then, I did what my partner seemed to want so dearly, and ended my chapter on Joshua asking Alex to meet and work things out.
ChatGPT eagerly used all 300 words to write the reconciliation it had been blindly pushing towards for half a dozen chapters without any setup. The chapter used all the cliches baked into the story, with the phrase “new beginning” said for the fourth time, and of course, their kiss “rekindling the sparks of a fire” (the third usage of the word “kindle”). Here's another excerpt from ChatGPT.
In that fragile moment, the barriers that had kept them apart seemed to crumble, revealing the raw emotions they had kept locked away for far too long. Their lips met in a hesitant, delicate kiss, as if rekindling the sparks of a fire that had long since faded.
As they pulled away, their eyes locked, and the unspoken words of love and longing hung between them like a tangible thread.
"Alex, I truly am sorry," Joshua's voice wavered, tears shimmering in his eyes.
Tasked with writing the final chapter, I bounced between proclaiming the deadly nature of the forest fire, conclusions about the strange competence of Mia and Jessie I had been setting up for chapters without ChatGPT seeming to care, and descriptions of Alex and Joshua's rocky couples' counseling. The story, which ultimately had turned out to be an entirely meaningless and boring one, ended up being titled “Infernal Escapade." It was an appropriate label for the torturous hours I had spent trying to squeeze any amount of intrigue out of this terrible narrative that I could, doing my best to escape ChatGPT's cyclical and profoundly uninteresting style.
And with that, the nihilist romantic masterpiece “Infernal Escapade” seemed to come to a close, and like all great pieces of literature, it has stuck with me. The experience of writing the story, and analyzing every creative decision the language model made, has brought a new meaning to artificial intelligence, for the narrative is truly artificial, a soulless plot about two meaningless characters with a random forest fire for a backdrop.
When my partner's identity was revealed to me, I was not surprised, but I became nonetheless illuminated by just how bad ChatGPT still is at this sort of thing, with every chapter continually dowsing water on whatever fire or warmth remained. The Switch-Lit site is available for you to try out with real people, but I've asked that ChatGPT never be reintroduced - I've seen enough of the bot's artistic capabilities for a lifetime.
