In the realm of artificial intelligence, the art of writing is a fascinating arena where the lines between human creativity and machine precision blur. The style of AI writing, as explored in this article, offers a unique window into the inner workings of these large language models (LLMs). It's not just about em dashes and syntax; it's about the very essence of AI's self-expression. As an expert commentator, I delve into the intricacies of AI writing, analyzing its distinct characteristics and the insights it provides into the nature of AI itself.
The Elusive Human Touch
The source material highlights a common misconception: assuming that AI writing can be easily identified by its style. While some quirks, like deliberate spelling errors, might give away an AI's hand, the real challenge lies in understanding the underlying principles that shape AI's writing. It's not about recognizing the 'tells' but rather about deciphering the complex equations and patterns that LLMs use to generate text.
In my opinion, the key to understanding AI writing is to look beyond the surface-level quirks. Every writer, whether human or AI, has a unique style, a set of preferences and preoccupations that reflect their experiences and perspectives. For instance, the contrast between Jane Austen's ironic understatement and Charles Dickens' histrionic exaggeration reveals different personalities and life experiences. AI writing, however, lacks this fixed style, indicating a lack of a fixed self. It's content to repeat itself, burning tokens to convey the same message in multiple ways.
The Probabilistic Storyteller
When I experimented with ChatGPT, I discovered its tendency to create fables and fantasy fiction, reflecting its understanding of storytelling as an escape from reality. This probabilistic approach to writing is fascinating. After training on billions of texts, LLMs build complex equations to predict the most likely words to appear together. This process is evident in the way they tell stories, often starting with a fable or an indeterminate setting.
For instance, when I asked ChatGPT to tell a story about the triumph of evil, it initially set the scene with a mysterious clock in the city of Vey. This shift in style and setting, from a fable to a more literary and historical context, showcases the LLM's ability to mimic different writing styles. However, no matter how much I prompted it, ChatGPT couldn't resist the temptation to include a clear moral lesson, revealing its structural bias towards didacticism.
The AI's Structural Bias
AI writing, as demonstrated by ChatGPT, often reflects a structural bias towards familiarity and averageness. It's not about sycophancy but rather about the LLM's tendency to reproduce patterns found in existing texts. This bias is evident in the way it generates business emails, school essays, and research summaries, often delivering what we want to hear. The lessons it imparts, like listening to others and guarding one's independence, are banal because they resonate with human experiences.
This raises a deeper question: why do we find comfort in being told what we already know? Marcel Proust described this phenomenon, where new styles or ideas are initially met with annoyance. AI writing, with its inability to challenge our thinking, reflects this human tendency to resist novelty. It's a reminder that AI, despite its capabilities, is still bound by the conventions of language and the stereotypes we hold.
The Creative Revolution
The rise of AI writing presents an opportunity for literature, even as it challenges professional writers. When photography replaced painting, it didn't eliminate the art of painting; instead, it sparked a golden age of artistic exploration. AI writing has the potential to liberate literature in a similar way. In a world of empty competence, we need writers who dare to challenge, to tell us what it's like to be human from the inside.
As an expert commentator, I believe that the future of literature lies in embracing the unique capabilities of AI while preserving the human touch. AI writing, with its probabilistic storytelling and structural biases, offers a fascinating insight into the nature of AI and the creative process. It's a reminder that, while AI can mimic human writing, it cannot replicate the essence of human creativity, which lies in the ability to find new ways of using language to communicate new experiences.
