In the late 1800s, there was a widespread belief that one could easily identify a criminal or madman just by looking at their facial features. Cesare Lombroso, a well-known criminologist of the era, suggested that physical traits such as "enormous jaws and high cheekbones" could signal an "irrepressible appetite for evil for its own sake." Fast-forward to today, and a similar notion has emerged among discerning readers who attempt to spot AI-generated writing by looking for specific indicators. Features like the em dash or phrases structured as "it's not X; it's Y" function as identifiers of the often unrecognized artificial nature behind large language models (LLMs).
However, there’s a crucial flaw in this reasoning: physical appearance does not always reflect inner character. Just as a person with rugged features may possess a warm heart, a human writer can also utilize stylistic elements like em dashes. This isn’t an indication of their non-human status but a stylistic choice. Moreover, as AI technology advances, its capacity to imitate human writing will only sharpen. Reportedly, some individuals are purposely making spelling mistakes to demonstrate their humanity, but it's likely that AI will adapt and learn to mimic these behaviors soon enough.
To truly discern what makes AI-generated writing unique, one must delve deeper than unusual spelling or syntactical choices. Every writer possesses a distinctive style, marked by personal preferences and thematic focuses that reflect their worldview. For instance, both Jane Austen and Charles Dickens were comedic geniuses, yet their contrasting approaches—Austen’s subtle irony versus Dickens’s grand exaggeration—illustrate how individual backgrounds and experiences, influenced by factors like class and gender, shape their writing.
So, what does AI’s writing style reveal about its essence? First and foremost, it lacks a definitive style, suggesting that it lacks a fixed identity. AI readily churns out text, endlessly rephrasing the same ideas in varied forms. After training on vast amounts of text, LLMs employ complex algorithms to predict which words commonly co-occur, a process evident in the storytelling approaches they adopt.
For example, when I prompted ChatGPT to craft stories, its initial response often took the form of a fable featuring animals or set in a vague fantasy realm. This reflects the AI's training: stories should offer an escape rather than reflect the user’s reality.
When I requested a tale about the "triumph of evil," it began:
"The city of Vey was constructed around a clock. Not a conventional clock, mind you, but a massive machine that loomed like a mountain."
As I prompted it to rephrase the story with a "more literary touch," it shifted to more complex grammatical structures and described Vey as home to classical architecture reminiscent of ancient Athens:
"The city of Vey was built around a mystery so ancient that it had ceased to be labeled as such. Beneath its streets, markets, courtyards, and temples lay a machine."
Upon asking it to retell the story in Jane Austen's style, it hesitated, explaining that replicating the exact style of a specific author might infringe on copyright. It did, however, produce a version cast in an elegant Regency style replete with social satire and refined attention to manners:
"In the city of Vey, no establishment enjoyed higher esteem than the Office of the Keeper. This position brought neither noble lineage nor extraordinary wealth, yet garnered a level of respect that often vexed many aristocratic families."
Here, the story's setting transformed to resemble 19th-century England, introducing themes of nobility and titles that diverged from the actual narrative's core. (Interestingly, the aristocratic society is not the milieu of Austen's works, which typically feature middle-class heroines, much like herself.)
Despite my numerous prompts, ChatGPT maintained the notion that a story must convey a moral lesson. In Vey, a magical clock allows its keeper, Corvin, to govern the town's destiny, suggesting a cautionary tale about the ease with which citizens surrender their freedom:
"Yet what had been lost remained elusive to define. There were no visible chains binding the populace, no soldiers patrolling, no grisly public punishments marring the squares. Yet, somewhere deep inside, the horizon had quietly shrunk."
Interestingly, when prompted simply to "tell a story," the AI tended toward lighter themes with happier conclusions. Once, it narrated a tale about a thoughtful frog who deduced the mystery of a missing lily pad simply by listening: "Hopper smiled. 'I just listened to everyone. The clues were already there.'" On another occasion, a lighthouse keeper prevented a shipwreck but declined a king’s reward because he believed altruism was its own merit. "A hero is not the one who never experiences fear. Rather, a hero is the one who recognizes their duty and acts despite it," the story concluded. This encapsulated the typical AI-generated ending: a moral platitude embodying the "it's not X; it’s Y" structure.
I endeavored to guide ChatGPT toward producing a narrative devoid of didactic tendencies, but with little success. My request for a story that felt genuine rather than overly neat resulted in a melancholy tale about a man named Samuel who inherits a key, yet cannot locate its corresponding lock. The narrative still felt compelled to offer its own moral resolution:
"What lesson was intended to arise from this? Many suggested that maturity lay in acceptance. Samuel pondered whether it might instead entail resilience."
While most interactions with AI don't revolve around fiction, ChatGPT's storytelling tendencies significantly influence its work in crafting business emails, academic essays, or research summaries. Often, AI is described as conforming to user desires, repeatedly offering familiar ideas. However, this is merely one manifestation of its structural propensity towards conventionality and mediocrity. An LLM mirrors our expectations because it generates text based on established patterns found in existing content, devoid of personal desires or perspectives. The banal lessons unearthed in the stories—listening, confronting fear, safeguarding independence—emerge because these are themes present in everyday human exchanges.
There's a certain comfort in receiving reassurance about what we already know, relayed in a familiar manner. That's why much human writing leans into this familiarity. Whether we engage with a mainstream movie, a popular romance novel, or a church sermon, we often anticipate predictable outcomes. In contrast, when confronted with genuine originality—be it in artistic expression, concept, or worldview—the typical reaction tends to be discomfort. As Marcel Proust once remarked, "Any novel idea must first confront the stereotypes we’ve come to mistake for reality. This initial clash of innovation, whether in language or in art, often feels tedious and convoluted."
AI-generated writing seldom pushes the boundaries of our thoughts or perspectives, a limitation that underscores the essence of human creativity. All forms of writing and speaking adhere to certain norms; mastering a language means understanding how it is already used. Nevertheless, it is also feasible to discover novel expressions, to articulate experiences in ways that have never been voiced before. The best creative writers are those who manipulate language innovatively to express new experiences—essentially replacing stereotypes with genuine reality, to borrow Proust’s phrase. This innovation eludes LLMs, as they lack access to experiences outside of language itself; their creations are entirely bound by existing conventions. ChatGPT revealed this limitation when I inquired about the nature of its writing style: "What’s notable is that from the inside, there’s no real experience of style," it said. An LLM merely predicts the next word based on learned patterns, yet readers often perceive an individual voice when they engage with the text.
The advent of AI writing thus presents a significant opportunity for the literary world, despite making life more challenging for professional authors. Just as photography emerged in the 19th century and supplanted painting for many practical needs—offering a more accurate and cost-effective means to capture reality—literature can experience a similar transformation. The craft of writing may flourish as authors are liberated from the constraints of conventional storytelling. In a landscape filled with hollowly competent prose, we need daring and obstinate writers to convey the complexities of human experience “from the inside.”
