When authors use AI-based [creativity support tools] to expand a small amount of input text (such as a one-sentence instruction) to a large amount of output text (such as a complete written story or essay), they delegate many of the creative decisions involved in producing the larger output to the CST — resulting in a piece of writing with an unusually low ratio of human decision-making to output length. In other words, the expressive intent of the author is underspecified relative to the amount of text that is generated, and the resulting piece of writing is unusually sparse in terms of expressive intent per word. We refer to this unusual situation as the dearth of the author: the naïvely AI-augmented author is not absent or dead, but their intent is stretched so thinly over their writing that they may feel barely present. — Max Kreminski, “The Dearth of the Author in AI-Supported Writing” (2024)

Why post this? I’m intrigued by how this “dearth of the author” challenges traditional notions of authorship in AI-supported writing. Kreminski’s idea implies a spectrum of authorial presence, which is dialed down as users delegate their decision-making to generative AI tools, snapping to the grid supplied. The concept destabilises binary models of authorship (“either you’re the author or you’re not”), opening new ways to think about creativity and expressive intent.

This “dearth” appears differently across fields. In programming, AI assistance complements existing collaborative workflows, aligning with an emphasis on teamwork and shared resources. Programmers, used to leveraging libraries, frameworks, and pre-written code snippets, may view AI as simply another tool in their arsenal. Writing, however, places a greater premium on individual voice and originality. This emphasis on personal style can create tension with AI-generated text, which struggles to replicate the voice, empathy, and accountability of human-authored writing, resulting in a more noticeable “dearth” of expressive intent.1

Three-dimensional wavy surface across numbered axis, a grid diagram illustrating the continuity of consciousness, from William James' 'The Principles of Psychology' (1890).

Strategies for enhancing authorial presence

Kreminski also speculates about a possible counterpoint to this authorial “dearth”, dubbing such “an abundance of the author.” (Kreminski 2024) But what might this look like in practice? While Kreminski’s concepts address questions of authorship in an age of AI tools, ideas about enhancing authorial presence have historical precedent. The Oulipo movement, founded in 1960, offers an interesting reference point. Oulipian writers used constrained writing techniques to spark creativity and explore new forms of expression. For instance, Georges Perec’s 1969 novel La Disparition was written as a lipogram, entirely excluding the letter ‘e’. Each word choice became a deliberate expressive act, resulting in a text dense with Perec’s creative problem-solving. Rather than limiting expressive intent, such Oulipian constraints served as creative springboards. By forcing writers to find creative solutions within strict parameters, they actually enhanced the author’s unique voice and presence in the text.

While the Oulipo attempted to enhance creativity through structured limitations, the work of writer and media scholar Kyle Booten provides a second example of attempts to increase authorial presence, albeit through very different means. Inspired by ancient Greek rhetorical exercises, his “digital progymnasmata” offer real-time feedback, challenging users to refine their language in ways that diverge from expected pattern.2 For instance, one such tool assesses linguistic rarity, pushing writers to use less common words or syntactic structures. Unlike AI tools that can generate large quantities of text with minimal input, Booten’s tools require writers to actively engage with each new suggestion, combatting authorial dearth by encouraging (if not mandating) more deliberate expressive choices.

I’m particularly taken with his Lotus Chorus Workshop, which takes this concept further, simulating a writing workshop that offers diverse suggestions and critiques after each new sentence. In this writing environment, writers are pushed into a responsive mode, forced to constantly make decisions and assert their creative vision amidst a cacophony of suggestions. This fosters what Booten elsewhere refers to as a “Zone of Proximal Derangement”, in which AI suggestions are just strange enough to push writers out of their habitual thought patterns without being completely unusable. By fostering a sense of cognitive overload “that pushes the user’s mind in too many directions at the same time” (Booten 2023), this piece of software aims to stimulate creativity. Setting out to make writing “harder”, it offers a distinct approach to how digital tools might enhance rather than diminish expressive presence in an age of computer-mediated authorship.

Prompt engineering

In considering techniques to enhance authorial presence, it is also interesting to examine how different groups are making use of current AI writing tools. Although these tools are contributing to this “dearth” of expressive intent by suggesting completions3 or generating entire texts, many users are learning sophisticated prompt engineering techniques to produce more targeted, nuanced outputs. “Prompt engineering” can allow users to shape the AI’s output through active, iterative engagement. Creators can steer the AI towards producing content that better reflects their unique perspective and style by providing specific, detailed instructions that include salient contextual factors and desired outcomes.4

More complex methods like chain-of-thought prompting can encourage generative models to articulate their intermediate reasoning steps, eliciting more complex and logically structured outputs. Similarly, prompt chaining enables writers to break down complex tasks into smaller, interconnected prompts, each building on the previous response. Such techniques often involve multiple rounds of refinement, with users crafting prompts that specify salient contextual factors, desired outcomes, and even the reasoning process itself. This can allow for finer-grained control over the creative process, producing outputs that better align with users’ own perspectives and expressive intents.

Black and white illustration from Octave Uzanne's 'The End of Books' (1894). Two men at a patent office desk, one speaking into a large horn-like device. Caption reads: 'The Author Depositing his Voice at the Patent-Office, to Prevent Counterfeiting.'

Future directions for computer-mediated authorship

AI writing tools have the potential to enhance authorial presence by encouraging more intentional and distinctive writing. Inspired by Booten and the Oulipo, future such tools could be designed to promote revision, increase linguistic density, and foster deeper engagement with the creative process. By asking probing questions, suggesting areas for expansion, and providing real-time feedback, they could help writers refine their ideas and expression over multiple drafts, strengthening their voice and presence in the text. As these tools evolve, they may reconfigure many of our assumptions about authorship. Rather than viewing AI as a threat to authorial presence, we could come to position it as a collaborator in new forms of augmented creativity.

From Oulipian constraints to the cognitive challenges posed by Booten’s digital progymnasmata, and the nuances of prompt engineering, we see various attempts to reassert human agency and intentionality in the creative process. The greater adoption and integration of generative AI tools could lead to a redefinition of authorship that emphasises the curation, direction, and refinement of content, rather than the methodological individualism of singular visions and distinctive voices. I’m particularly interested in the potential for collaborative models where multiple human authors work alongside AI systems.

Imagine a writing environment where a team of authors, each with their own unique perspectives, engages with a generative AI system that acts as mediator and creative catalyst. This more collaborative approach could lead to rich, multifaceted texts that blend diverse human perspectives with AI-enhanced creativity. AI systems could act as a mediator, identifying connections between authors’ ideas and proposing novel combinations. It might also help manage the workflow, ensuring that each contributor’s voice is heard and integrated. Beyond creative writing, a “writers room” model could be valuable in fields requiring interdisciplinary collaboration, such as policy development or academic research.

However, as we explore these new possibilities, we must remain mindful of the challenges. The integration of AI in collaborative writing processes might inadvertently homogenise writing styles or lead to an over-reliance on AI-suggested connections. It could also raise questions about attribution, ownership, and control. How do we acknowledge the role of AI in such collaborative works? How might different configurations of human-AI collaboration affect the power structures within writing teams? These are complex issues that will require ongoing dialogue and experimentation.

Ultimately, the goal is not to let the “dearth of the author” force us into a defensive crouch, but to imagine and cultivate rich, collaborative spaces where human creativity and AI capabilities can coexist and interact. By embracing the potential of AI-augmented writing while engaging with its implications, we can create the conditions for these tools to scaffold an abundance of expressive intent, rather than resulting in its dearth.

  1. Consider the use of personal anecdotes in longform essays, a highly-valued feature in human-authored texts, but challenging for AI to replicate; this ability to weave personal experiences into a narrative is a clear example of strong authorial presence. ⤴︎

  2. I am grateful to Nick Seaver for introducing me to Booten’s work. ⤴︎

  3. Compare with the phrase “spicy autocomplete”; a colloquial, somewhat pejorative label, casting new generative AI technologies as more sophisticated versions of existing (and, by now, familiar) text prediction tools. ⤴︎

  4. Mark Carrigan discusses prompt engineering as an expression of cultural capital. Those with broader cultural knowledge or more diverse experiences, Carrigan argues, might be better at crafting and framing prompts, thereby benefitting from stronger, more expressive outputs. ⤴︎