Preserving Writer Values in AI Writing Assistance Tools

Recent large language models (LLMs) have demonstrated significant advances in the area of creative writing applications, ranging from poems [2] and video scripts [15] to screenplays [10]. Beyond traditional performance metrics like accuracy [11], coherency, and relevance [9], these AI-powered writing tools have further demonstrated their potential in the context of creative writing to proofread texts [1], describe scenes and characters, and inspire creative writers [5].

Despite the benefits of AI writing tools, their possible conflict with values core to writers may limit their appeal. Are writers transitioning from being the primary creators to mere curators in co-writing processes? If used carelessly, these tools may override creative writing's inherent values as a deeply personal, human endeavor. This research investigates writers’ core values, examining how those beliefs guide their application of AI writing aids. Interviews with 8 creative writers reveal a strong emphasis on ownership from a lens of authentic expression, as well as writer considerations on what makes the process enjoyable.

We find that writers use AI to unblock themselves, preferring aid where they struggle in the writing process, while limiting its role to keep the enjoyment of the process. As we move forward, it is crucial to engage in ongoing dialogue with the writing community to ensure that AI tools enrich the writing landscape, fostering creativity while preserving the cherished values of authenticity and personal engagement in the creative process.

Support for writing through computational means has evolved significantly, beginning with the advent of initial spell-checkers [13] and advancing to contemporary software that guides story creation [3]. These tools serve various purposes, from simplifying the task of texting  [14] and assisting in selecting words with precise connotations [4], to enhancing emotional writing [12], and facilitating writing in professional settings [7].

While prior research has engaged real-world writers to evaluate the capabilities of AI writing tools, it often conducted their evaluation in a controlled lab setting: writers are invited to communicate their needs with researchers and then try using the implemented prototypes for some predefined tasks. These evaluations have often prioritized system usability and user engagement. As a result, little research has examined how writers might use these AI-powered writing tools in their daily writing experience. However, the daily writing experience is a deeply personal and human endeavor to these creative writers. Without a close examination of their daily writing experience, it is unlikely to observe how these writing tools might compromise the intrinsic values of writing. Prior interview studies have explored writerly values on receiving support in the writing process [6]. Others have explored longer term insights from professional writers using an AI tool, examining how these tools integrate into and affect the daily creative processes [8].

We conducted 8 semi-structured interviews with writers who regularly use AI (6 fiction writers, 2 blog writers) detailed in Appendix 1. Our interview questions are structured to understand writers’ interactions with AI tools across different stages of the writing process for all manners of tasks (grammar checks, writing prose, brainstorming ideas, and developing characters). Interviews lasted 90 minutes, with 30–40 minutes spent observing live-writing sessions where participants continued or began new pieces and discussed prior AI-assisted work (Appendix A.2). This was presented as time for participants to work on what they would have even had they not participated in the interviews. This enables us to capture the nuanced interactions in the translating and reviewing stages, as well as to gain insights into the role of AI in the planning phase of writing. We are especially interested in writers’ motivations around using, not using, or correcting AI output. For the majority of our interviews, participants stated that they would not have done things differently without observation, with one participant answering that they would have written more slowly if not observed. Each participant was paid $40 in the form of gift cards. This study was reviewed by our IRB and deemed exempt.

We analyzed the interviews with the first author leading the open coding process. We find two values that writers cherish: 1) authentic expression, or the feeling that the writing is true to their vision and original expression, with the AI helping them achieve it; and 2) the love of the writing process, or specific aspects of writing that they would like to preserve, even while using AI. We also observe a tension between how some writers rationally view their use of AI and how they feel.

Our preliminary findings highlight the importance of authentic expression in writers’ interactions with AI in the context of how each individual writer expresses these values, whether that is through the ideas or prose itself. We find that one of the main benefits to using AI in the writing process is getting unstuck, which can inform design decisions for future creative writing applications. These tools should aim to support aspects of the process that writers dislike while preserving the aspects that they enjoy. One direction of exploration could be to focus more on a notion of closeness to the vision of the writer as a metric. We see that writers are still working out their relationship to LLMs and how to negotiate their values. In addition, the current stigma surrounding using AI for writing complicates how writers would like to share their work. Another avenue of exploration is tracking the provenance of a written piece, so that authors can track where suggestions came from at different levels of sentences, paragraphs, beats, and overarching ideas. Not only does this enable them to reflect on their usage of these tools, it may provide a way to be transparent about how AI was used that also shows their own effort and creative vision in the process, especially with the insight that AI use that doesn't contribute directly to the text output can still be helpful to the writing process. In continuing our research, we wish to ask what makes a good AI writing assistant that puts writers’ values first? How do we think about where, when, and how much the AI should be involved compared to the writer?