Authoring Novels in the Age of AI

The art and science of writing have experienced remarkable transformations over the centuries, with technological advancements significantly altering human culture and society. Each innovation, from the printing press to typewriters and word processors, has distinctly impacted narrative forms. The emergence of Artificial Intelligence (AI), particularly Generative Pre-trained Transformers (GPT), prompts the question: Will GPTs revolutionize writing as profoundly as word processors once did? This article explores this question, concentrating on authoring novels due to their capacity for innovation, broad scope of work, and depth of creativity. However, I would like to point out that many concepts discussed here apply across various writing disciplines.

LLM and GPT

The advent of personal computers and word-processing software in the late 20th century marked a significant milestone in writing novels. This development facilitated a more straightforward and flexible editing process, enhanced formatting options, and streamlined the publishing process through the ease of digital manuscript sharing. Today, Large Language Models (LLMs) such as Generative Pre-trained Transformers (GPT) represent an even more significant advancement. Unlike traditional word processors, GPTs transform computers into partners of creation. They extend beyond grammar and formatting to generate coherent content, propose narrative developments, and mimic diverse writing styles, broadening the horizons for writers to explore innovative ideas and stylistic approaches.

LLMs offer support in numerous areas. First, they ensure consistency and adherence to style guides like the Chicago Manual of Style. While writing aids within word processors or add-on programs (e.g., Grammarly) have been available, LLMs facilitate the customization of grammar preferences via natural language commands, eliminating the need to navigate complex interface settings.

Second, AI tools enhance the research process for authoring fiction by presenting a superior alternative to conventional search engines. They rapidly sift through vast amounts of data, delivering pertinent information free from the clutter of advertising or biased content, which is particularly advantageous for genres necessitating intricate world-building. Although Google and similar search engines have supported such research for the past two decades, AI tools can now scour the web and extensive databases, summarizing and analyzing responses to specific inquiries without overwhelming users with advertisements. Maintaining a low "temperature" (accuracy) setting on your preferred GPT is crucial to prevent generating fabricated responses. If direct temperature control is not available, instruct the GPT to prioritize accuracy and precision and corroborate the information obtained with that from competing GPTs.

Third, LLM models can streamline routine tasks. Just as a word processor can replace a single word in a 200K-word document in seconds, an LLM can accomplish more complex tasks, such as altering the point-of-view from first to third person or modifying the tense of a novel to present tense. A task that once took me two months to change the tense of a novel from present to past now takes merely a minute with GPT-4. AI can execute time-intensive tasks swiftly. For instance, the following command could save hours for an author in a lengthy novel where a car accident has been introduced: "Make Jennifer's dialogue less eloquent, brief, and confused in all dialogues following her severe concussion, and adjust Mike’s responses accordingly."

LLMs are invaluable for generating ideas. Authors can input themes, character sketches, or plot elements and receive many creative suggestions. They can also offer critiques on drafts and spot inconsistencies in the plot. In my experience with AI writing tools, I have never been at a loss for ideas; a simple prompt like "Act as a developmental editor and critique my work" can elicit astoundingly perceptive feedback in seconds.

The choice of LLM application depends on the specific needs: Claude excels in handling up to 200K tokens; GPT-4 is distinguished for its creativity; Gemini (previously Bard) excels in accessing the most current information.

Preserving the Authorial Voice

The principal challenge in utilizing a GPT for brainstorming or editing is maintaining the author's distinctive voice. This preservation is necessary for AI-assisted novels to avoid becoming homogeneous in style and structure, stifling creativity and the evolution of the narrative form.

When OpenAI introduced ChatGPT to the public in November 2022, authors could instruct ChatGPT on editing tasks; however, these instructions would be forgotten after a time due to the limited token capacity available to users. In November 2023, OpenAI launched Custom GPTs, enabling users to specify their GPT's functions and ensuring it retains the instructions for every interaction, irrespective of text or chat length.

Custom GPTs empower authors to create AI assistants aligned with their writing style and narrative preferences, capable of emulating an author's unique voice and thereby becoming a more cohesive and harmonious partner in writing. Customization within Custom GPTs is facilitated through three principal mechanisms:

1. Instructions: Authors can detail their stylistic preferences, adhering to general guidelines such as the Chicago Manual of Style and incorporating specific preferences for sentence structure, length, and punctuation. For instance, in my custom GPT for fiction writing, I have delineated twelve punctuation rules that I expect GPT-4 to follow during editing. One such guideline is: "Preserve the author’s use of em dashes, which signify interrupting thoughts unrelated to the sentence's main idea. Do not remove the author’s em dashes, nor introduce new ones."
2. Knowledge Base: This feature comprises a collection of sample works (limited to 20 files) that the Custom GPT app consults during editing. These samples can vary from structured database files to simple MS Word documents, providing a comprehensive context for the AI's functionality. I have contributed samples of my writing in MS Word format to the Knowledge Base to acquaint the Custom GPT with my writing style and authorial voice.
3. APIs: Custom GPTs can also integrate with external applications, enabling access to an author's more comprehensive array of works, ideas, and notes, enhancing its capacity to tailor idea generation and editing in alignment with the author's style. I am exploring this capability, which necessitates authenticated access to a database or application that, unlike the Knowledge Base's 20-file limit, can encompass unlimited content from the author and related materials. ?One of my applications of this feature, a work in progress, involves uploading a 200K-word novel onto a server, publishing a 20K-word novella based on this novel, and then providing readers access to interact with the novel through a Custom GPT via its API, enabling them to engage with the narrative in depth by inquiring about details, interpretations, background information, flashbacks, etc. This approach fosters a highly interactive reading experience through AI mediation.

Custom GPTs promise to transform story creation and writing processes while safeguarding and enriching the author's voice and style. However, it's important to note that the industry is rapidly evolving, with the possibility that features or concepts discussed here may be enhanced, automated, or phased out at any moment.

The Future of AI-Assisted Writing

The release of GPT-5 is anticipated in 2024. As GPTs continue to develop, they are expected to exhibit improved reasoning abilities, enhancing their capacity to comprehend complex instructions within Custom GPTs. Such advancements in memory and reasoning are likely to facilitate these models' adherence to the nuances of an author's voice and style, as informed by more comprehensive sample works stored in the Knowledge Base.

Future iterations of Custom GPT models are projected to process more intricate instructions. Beyond capturing the authorial voice, these models will navigate the narrative voice (the "person" recounting the story,) the distinct voices of each character, and the narrative perspective (point of view). The current generation of Custom GPTs cannot manage these elements concurrently.

Custom GPTs will advance to indefinitely retain narrative details and discussions, a capability beyond the reach of present models. This enhancement will render Custom GPTs more contextually aware, supporting crafting complex, layered narratives. Authors might, for example, submit a draft paragraph or chapter to their Custom GPT, seeking insights on its impact on the storyline, character development, and reader perception.

The potential of Custom GPTs is limited only by imagination, particularly concerning their API interfaces. Custom GPTs designed for interactive engagement with stories could allow readers to offer feedback that informs new narrative versions. The author could then evaluate and publish these versions, offering diverse narrative paths and a customizable reading experience. Thus, the traditional linearity of novels may give way to adaptive, personalized reader journeys.

Sharing a novel’s Custom GPT with fans opens the door to fan-generated content that complements the original work, paving the way for a transmedia ecosystem of interconnected stories, videos, and virtual reality experiences. This offers readers unique narrative paths and the opportunity for fans to benefit financially from their contributions. Such fan-created content could be integrated into the novel's transmedia framework, fostering a dynamic interactive environment around the story's universe.

Even with advancements in AI translation, human expertise remains essential in safeguarding the artistic and literary nuances of translated works. This highlights the irreplaceable role of expert translators. As AI technology in translation and localization continues to improve, novels will become more globally accessible, opening up new markets for authors and publishers and providing readers worldwide with diverse literary voices and stories.

?

Challenges

Integrating AI in novel writing introduces several ethical dilemmas, among which intellectual rights, publishing practices, and originality stand out.

Concerning copyright when employing AI tools, the extent of an author's original contribution versus the AI's assistance plays a critical role. Utilizing AI for editing and generating ideas can be likened to using grammar-checking tools, where the resulting work remains the author's intellectual property. However, instructing AI to produce a novel mirroring the style of Delia Owens and the plot structure of "Where the Crawdads Sing," with a gender-swapped protagonist, could lead to legal complications with the original work's copyright holders.

The capacity of many AI tools to mimic established authors' writing styles is evolving rapidly. For instance, whereas GPT-4 could previously alter images to resemble Picasso's style, it now refuses such requests, indicating a growing awareness of copyright constraints.

Copyright owners are increasingly required to define their policies regarding AI's interaction with their content. At X-Plain.com, a content company I assist in managing, we have established licensing agreements that permit the use of our material by AI systems under certain conditions, necessitating a special license.

AI's impact on the publishing industry is profound, leading to an influx of manuscript submissions. This surge challenges traditional publishing frameworks, necessitating a shift in the selection, editing, and marketing of manuscripts. Publishers might increasingly depend on AI to evaluate a novel's uniqueness and market appeal, adjusting these criteria based on real-time feedback from online data and reviews.

Kindle Direct Publishing (KDP) now asks authors about their use of AI in creating text, images, and translations, framing it as a research inquiry. Whether such disclosure should become a norm in the industry is debatable. Amazon’s KDP does not inquire about the tools used for grammar checking or the identity of the copyeditor; why should it ask which AI tool was used?

I talked about the discussion around AI-induced uniformity in authoring novels earlier. A novel's success is often contingent on its distinctiveness, underscoring the significance of Custom GPT tools in preserving an author's unique voice and style. Without this uniqueness, a book risks being lost among the multitude of AI-generated novels that fail to sell more than a hundred copies, a trend expected to worsen with AI's proliferation. Before AI, approximately 90% of self-published books sold fewer than 100 copies.

Final Notes

As AI continues to evolve, it promises to open new avenues for writers, offering innovative tools for creativity and expression. Despite the initial resistance from many authors to LLM when ChatGPT was released in late 2022, I observed a growing acceptance of these tools within the writing community a year later, a trend that appears inevitable.

The core of storytelling is deeply rooted in human imagination, emotion, and experience. AI tools assist authors in crystallizing their thoughts and visions. While some may adhere to traditional writing methods, the vast majority of authors will eventually recognize and leverage AI's applications in writing, many of which still need to be explored.

Historically, the introduction of word processors and grammar checkers did not alter the inherent linearity of novels. However, the push for originality in the AI era may prompt a shift towards interactive, non-linear storytelling—a concept previously envisaged but hindered by developmental challenges. AI technology now makes the realization of such interactive narratives feasible.

Will AI ever fully replace authors? According to AI itself, the answer is emphatically no. Like the transition from quill to typewriter, the role and techniques of authors are evolving rather than vanishing. While the use of AI for writing assistance is set to become widespread, surpassing human creativity in novel writing is still a challenge for AI. For instance, providing a GPT with two keywords (e.g., tinnitus and doctor) and requesting a short-story joke without puns—given that LLMs excel at puns, considered the simplest form of humor—reveals the current limitations of LLMs in grasping complex human humor.

My experimentation with training a custom GPT to craft jokes has shown promise but also highlighted challenges. It can conjure up a remarkable universe but often misses the mark on the punchline. Success hinges on clear instructions and exemplary jokes. The lack of extensive resources on joke construction complicates this endeavor. "Joke Maker," my custom GPT, has shown promising progress. Here's a recent example of its creativity:

Once, a man named Bob began hearing voices in his head. Concerned, he consulted a doctor. After a comprehensive examination, the doctor reassured him, "Bob, you're in perfect health. There's no physical cause for these voices."

Perplexed, Bob inquired, "Then what should I do?"

The doctor suggested, "If the voices offer friendly advice, engage with them; they might provide valuable insights."

A week later, Bob returned, smiling broadly. "Doctor, your advice was excellent! I'm enjoying the company of my new internal friends."

Curious, the doctor asked, "What do the voices discuss?"

Bob responded, "They've been quite helpful. Just yesterday, they advised me not to pay your bill!"

We are at the dawn of the AI Revolution, standing at a thrilling juncture where technology and art converge, shaping a future in which human creativity and AI innovation redefine literary boundaries and, consequently, our understanding of humanity. ?