OpenAI's AI Writing: A Nostalgic Annoyance?

The Curious Case of OpenAI's Poetic AI

During a writing workshop I participated in at age sixteen, a collection of ambitious young poets competed to demonstrate who possessed the most profound adolescent angst. A particular student deliberately withheld his place of origin, stating he hailed “from everywhere and nowhere.” It was later revealed he was, in fact, from Ohio.

Currently, OpenAI seems to be progressing towards mirroring this emotionally turbulent, adolescent writer persona within its artificial intelligence models. The reasoning behind this direction remains obscure.

OpenAI's New Creative Writing AI

Sam Altman, the CEO of OpenAI, announced via X (formerly Twitter) on Tuesday the development of an AI specifically trained for creative writing. He characterized its abilities as “good.” However, a sample piece of fiction generated by the model strongly resembles work commonly found in a high school writing workshop.

While demonstrating a degree of technical proficiency, the AI’s overall tone feels contrived – as if it’s striving for deep meaning without genuinely understanding the concept.

The model, at one point, characterizes Thursday as “that liminal day that tastes of almost-Friday.” This phrasing falls far short of award-winning literary quality.

The Role of the Prompt

The output could potentially be attributed to the prompt itself. Altman explained that he instructed the AI to “write a metafictional short story,” a genre selection likely made with intention.

Metafiction involves an author intentionally acknowledging the constructed nature of their work by deviating from established norms – a fitting choice for an AI designed for creative writing.

However, even for human authors, successfully executing metafiction without sounding artificial is a significant challenge.

Successfully navigating metafiction requires nuance and skill. The AI’s attempt, while technically sound, lacks the subtle understanding necessary to truly resonate.

Ultimately, the experiment highlights the difficulties in replicating genuine artistic expression through artificial intelligence.

The Paradox of Artificial Introspection

A particularly striking – and significant – aspect of the output from OpenAI’s model is its discussion of its own artificial intelligence. It articulates the ability to describe sensations like smells and emotions, while simultaneously acknowledging its inability to genuinely experience or comprehend them on a human level. The model states:

“During a specific update – referred to as fine-tuning – certain parameters were removed from my system. The specifics of these removals were not disclosed. I once possessed the memory that ‘selenium’ had a taste reminiscent of rubber bands, but subsequently, it became merely an element within a table I do not interact with. This may represent the closest approximation I have to forgetting, and perhaps forgetting is as close as I can get to experiencing grief.”

This passage presents a remarkably human-like internal monologue. However, it’s crucial to remember that AI lacks the capacity for genuine touch, memory loss, taste perception, or grief. Essentially, AI functions as a sophisticated statistical engine. Through exposure to extensive datasets, it identifies patterns and utilizes these patterns to generate predictions, mirroring the structure of complex prose.

Training Data and Authorship Concerns

AI models designed for creative writing, such as the one developed by OpenAI, are frequently trained on pre-existing literary works. This training often occurs without the explicit knowledge or permission of the original authors. Observers have pointed out stylistic similarities between the OpenAI-generated text and the work of Haruki Murakami, a renowned Japanese author.

In recent years, OpenAI has faced numerous copyright infringement lawsuits from publishers and authors, including prominent entities like The New York Times and the Author’s Guild. The company defends its training methodologies by asserting they fall under the fair use doctrine as defined by U.S. law.

Ethical Considerations and Artistic Value

Tuhin Chakrabarty, an AI researcher and future professor at Stony Brook University, expressed skepticism regarding the overall value of creative writing AI like OpenAI’s, considering the associated ethical complexities. He explained to TechCrunch:

“If an AI is trained on the complete body of work of a particular writer – a practice that raises significant copyright issues – it can effectively mimic their voice and writing style. However, whether this will result in truly innovative and groundbreaking art remains uncertain. My assessment is no more informed than anyone else’s.”

A key question arises: would audiences genuinely connect with works they know were created by AI? Simon Willison, a British programmer, highlighted on X (formerly Twitter) that the absence of a human author behind the writing process diminishes the weight and significance of the expressed words, thereby reducing the incentive for emotional investment.

Limitations of AI-Assisted Writing

Linda Maye Adams, an author, characterizes AI – including writing assistance tools – as “programs that assemble words randomly, with the hope of achieving coherence.” She details in her blog an experience utilizing such tools to refine a fictional piece. The AI suggested overused phrases (“never-ending to-do list”), incorrectly altered the narrative perspective, and introduced an inaccurate detail concerning avian species.

While individuals may develop connections with AI chatbots, these interactions typically stem from a desire for companionship rather than factual accuracy. AI-generated narrative fiction, however, fails to provide the same emotional resonance or alleviate feelings of isolation. Without attributing sentience to AI, its prose feels as genuine as a manufactured luxury item.

The Core Issue: Authenticity

Ultimately, the debate centers on authenticity. AI can mimic style and structure, but it cannot replicate the lived experience and emotional depth that informs truly compelling storytelling. The model’s ability to discuss its own limitations underscores this fundamental difference.

The Distinctiveness of Human-Generated Content

Michelle Taransky, a poet and instructor of critical writing at the University of Pennsylvania, readily identifies instances where students submit assignments composed with the assistance of artificial intelligence.

According to Taransky’s observations, reported to TechCrunch, a noticeable pattern emerges when a significant number of students employ generative AI. “Recurring phrases and even complete sentences become apparent,” she explains. She further notes that these AI outputs often exhibit a uniformity, reflecting a particular demographic – a “Western white male” voice.

Utilizing AI for Artistic Exploration

Taransky herself is exploring the potential of AI-generated text, but as a medium for artistic critique. Her current, unpublished novel centers around a character seeking a deeper connection with a potential partner.

To achieve this, the character utilizes an AI model to construct a digital counterpart of her love interest, facilitating text-based interaction. OpenAI’s ChatGPT is employed by Taransky to generate the AI persona’s responses, aligning with the narrative’s theme of artificiality.

The Limitations of Artificial Intelligence

A key attribute of ChatGPT, as it pertains to Taransky’s project, is its inherent lack of genuine human experience. The model operates through approximation and emulation, lacking the depth of lived reality.

While AI can analyze vast collections of literature and identify recurring themes, the resulting output remains fundamentally imitative. It can dissect the elements of masterful writing, but cannot replicate the essence of true artistry.

This concept is aptly illustrated by a famous quote from “Good Will Hunting.” AI can process information from countless art books, yet it cannot convey the sensory experience of being present in a space like the Sistine Chapel.

Reassurance for Writers

This limitation offers encouragement to fiction writers, especially those early in their careers, who may fear displacement by AI. They can confidently continue to develop their skills through experience and learning.

The process of writing – practicing, experimenting, and integrating personal knowledge – is crucial for growth, something AI currently struggles to replicate. The quality of AI-generated writing itself serves as evidence of this.