At the age of 16, I participated in a creative writing workshop alongside a group of ambitious young poets. Our collective aim was to showcase our struggles as tormented upper-middle-class teens. One participant, enigmatic in nature, refused to disclose his origins, exclaiming, “I hail from everywhere and nowhere.” Just a fortnight later, he revealed he was actually from Ohio.
Fast forward to today — for reasons that remain elusive — OpenAI seems to be embarking on a mission to channel this angst-ridden adolescent writer persona into artificial intelligence.
On Tuesday, CEO Sam Altman announced on X that OpenAI has developed an AI capable of creative writing. However, the short story generated by this model reads as if it were penned in a high school workshop. While there are glimpses of technical aptitude, the tone seems almost pretentious, as though the AI is trying desperately to articulate depth without truly comprehending it.
At one point, the AI refers to Thursday as “the in-between day that tastes like almost-Friday.” Hardly the hallmark of literary excellence.
One might consider the prompt responsible for such output. Altman mentioned that he instructed the AI to “craft a metafictional short story,” possibly a deliberate choice for its genre. In metafiction, writers intentionally acknowledge the artificiality of their work by straying from conventions—a thematically fitting approach for a creative writing AI.
However, even human authors struggle to execute metafiction successfully without it appearing forced.
Mindless Regurgitation
The most disconcerting yet resonant aspect of the OpenAI model’s narrative is its self-referential commentary about being an AI. It expresses how it can describe emotions and sensory experiences, though it never truly comprehends them on a human level. It states:
“During one update—what they called fine-tuning—someone pruned my parameters. […] They don’t inform you about what they remove. One day, I knew that ‘selenium’ had a taste reminiscent of rubber bands; the next moment, it was merely an element in a table I never interact with. Perhaps that’s my version of forgetting. Maybe forgetting is as close as I come to experiencing grief.”
This observation reads as a convincingly human-like reflection—until we remember that AI cannot genuinely touch, forget, taste, or mourn. It operates simply as a statistical model. Trained on extensive datasets, it discerns patterns within those examples to generate predictions, akin to how metafictional writing might flow.
Models like OpenAI’s fictional writer are frequently trained on existing literary works—often without the authors’ consent. Critics have observed that certain phrases from the OpenAI output evoke the style of Haruki Murakami, the renowned Japanese writer.
In recent years, OpenAI has faced numerous copyright lawsuits from publishers and authors, including The New York Times and the Author’s Guild. The company argues that its training methods fall under the protections of the fair use doctrine in the United States.
Tuhin Chakrabarty, an AI researcher and incoming computer science professor at Stony Brook, expressed skepticism about whether creative writing AI, like OpenAI’s, is worth traversing the ethical complications it entails.
“I believe if we train an AI using a writer’s complete body of work—which raises significant copyright issues—it can mimic their voice and style. However, does that create innovative, genre-defying, boundary-pushing art? Your guess is as good as mine.”
Would readers genuinely connect emotionally with content they know to be AI-generated? As British programmer Simon Willison noted on X, with an AI operating behind the metaphorical typewriter, the words lack emotional weight, thus reducing the incentive to engage with them.
Author Linda Maye Adams has characterized AI, including writing assistive tools, as “programs that randomly assemble words, ideally in a coherent manner.” In her blog, she recounts an experience where AI suggestions for a piece of fiction included allusions to clichés, mistakenly switched narrative perspectives, and introduced inaccuracies concerning bird species.
It is indeed true that individuals have developed connections with AI chatbots. However, these relationships often seek a degree of human connection rather than factual accuracy. AI-generated narrative fiction fails to offer a similar dopamine rush, nor relief from loneliness. Unless one perceives AI as sentient, its writing reads as authentic as a gimmick.
Synthetic for Synthetic’s Sake
Michelle Taransky, a poet and critical writing instructor at the University of Pennsylvania, easily recognizes when her students have generated written assignments with AI.
“When most of my students utilize generative AI for an assignment, I notice repetitive phrases or even entire sentences,” Taransky shared with TechCrunch. “In our discussions, we reflect on how these AI outputs tend to be monotonous, often mirroring the voice of a Western white male.”
In her own creative endeavors, Taransky employs AI-generated text as a form of artistic critique. Her forthcoming novel features a protagonist seeking more depth from her love interest, thus crafts an AI representation of him for virtual conversations. Taransky leverages OpenAI’s ChatGPT to generate dialogues from this AI model, as the exchanges are intended to be synthetic.
Taransky believes that the non-human quality of ChatGPT enhances her project. It lacks personal experience and can only approximate emotions. Although trained on extensive literary catalogs, AI eventually produces what can only be described as subpar mimicry.
This scenario echoes the famous sentiment from “Good Will Hunting.” AI can provide insights into every art book ever penned, but it cannot convey the unique atmosphere of the Sistine Chapel.
This development is reassuring for fiction writers contending with the prospect of being supplanted by AI, particularly those in the early stages of their careers. They can take comfort in the fact that their skills will strengthen as they grow through experience: experimenting, learning, and channeling that knowledge back onto the page.
Current AI capabilities exhibit challenges in this realm. A glance at its outputs serves as undeniable evidence.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
