This year, we added generative AI as a topic to address logo design, photo-graphics, and video synthesis. Here is an example of the basics for a video prompting plan. In this text, I explore the concept of authorship in the age of generative AI, drawing parallels with established theories from film and media studies, particularly auteur theory.

One of the interesting questions in film studies and media studies is authorship, or “auteur,” from its French roots. Who is the “author” of a film? Is it the screenwriter? The producer? The director? How much do the actors contribute to the creative process? A similar query asked where the meaning in the film experience is produced. Is it in the author? In the content or “text”? How about in the audience or the viewer? A related perspective asks how much authorship or meaning is limited or organized by the “genre,” such as a comedy, drama, or science fiction?

The proliferation of Large Language Models (LLMs) capable of generating sophisticated text and imagery has introduced a novel figure into the creative landscape: the AI prompter. This individual, through the crafting of textual instructions, elicits responses from AI systems, thereby participating in the creation of content that often blurs the lines of traditional authorship and creative control. The question is whether a human AI prompter can be considered the “author” of text or imagery produced by an LLM? This discussion has become very relevant in fields such as education, law, and publishing. It invokes debates around creativity, control, and the very definition of authorship, portending friction between emergent AI tools and social institutions.

Or is it more akin to a “curator,” as understood in museum studies —a figure who makes decisions about what to include, what to exclude, and how to narrate the story an exhibition tells? In the realm of AI-generated content, the “prompter” is the individual who crafts the audio, pictorial, or textual instructions (prompts) that guide the LLM to produce an output, be it an essay, a poem, or a digital image. The quality, specificity, and iterative refinement of these prompts significantly influence the final result. Alternative analogies that may capture different facets of the prompter’s interaction with LLMs include the “chef,” the “collaborator,” the “instrumentalist,” or “partner.” But these mostly fall outside the realm of film and media theory and will be mentioned only briefly.

This exploration is not merely an academic exercise; the increasing ubiquity of LLMs in diverse fields, from literary creation to legal analysis, necessitates new conceptual tools to understand this new form of human-machine interaction. The US Copyright Office ruled in May 2025 that the results of prompts can, for the most part, be copyrighted. The US Constitution recognized early the social value of such arrangements in Article I, Section 8, Clause 8 when it codified the following:

“To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.”

The “Copyright Clause” is meant to promote innovation and economic development and the developer’s of AI LLMs do not seem to have much incentive to oppose the Copyright Office’s “Copyright and Artificial Intelligence Part 3: Generative AI Training.” This is not to say that other content creators have agreed, including people whose likeness has appeared in AI produced content. The Copyright Office has stated its intention to “monitor developments in technology, case law, and markets, and to offer further assistance to Congress as it considers these issues.”

Understanding Auteur Theory and AI

Originating with early film critics in the 1950s, Auteur theory posited that the director is the primary creative force behind a film. She or he is the “author” whose personal vision, recurring themes, and distinct stylistic choices are imprinted on their body of work. An auteur, like Alfred Hitchcock, Akira Kurosawa, or Steven Speilberg is seen as exerting a high level of control over all aspects of production. So some film theorists argued they make their films recognizable and uniquely their own, regardless of the participation of other collaborators like screenwriters or actors.

American critic Andrew Sarris later popularized and systematized auteur theory, coining the term in his 1962 essay, “Notes on the Auteur Theory” in The American Cinema. Later, he proposed three concentric circles of criteria for evaluating directors: technical competence (the ability to make a well-crafted film), a distinguishable personality (a recurring stylistic signature evident across their films), and interior meaning (arising from the tension between the director’s personality and the material). It refers to the often unconscious or ambiguous, personal vision, themes, and preoccupations of the director that permeate their films, even when working with diverse material or within the constraints of a studio system. The theory celebrated directors like Alfred Hitchcock, who, despite working within the constraints of the Hollywood studio system, managed to produce distinctive and deeply personal works.[2]

French film critics associated with Cahiers du Cinéma and “la politique des auteurs” championed the film director as the primary artistic force whose personal influence, individual sensibility, and distinct artistic vision could be identified across their body of work. This approach shifted critical attention from studios or screenwriters to the director, viewing them as an artistic creator of the work. Auteur theory includes determining whether directors have a personal vision and style based on a consistent aesthetic and thematic worldview, demonstrated mastery of the medium, recurring themes and motifs, and a willingness to push the boundaries of the medium. Do they maintain significant control or influence over the final product?

A key aspect of the auteur was the identification of a particular film style encompassing: visual elements, narrative structures, and thematic preoccupations that could be consistently associated with a director, serving as their “authorial signature.” Auteur theory drew a critical distinction between “true auteurs,” who infused their work with a personal vision and artistic depth, and “metteurs en scène,” who were seen as competent but workmanlike directors merely executing the details of a script without a discernible personal stamp. For instance, Michael Curtiz who won Best Director for Casablanca (1942) was often placed in the latter category, while Nicholas Ray, known for the (1955) Rebel Without a Cause, starring James Dean was celebrated as an auteur.  

The “method actor” analogy reinforced the prompter is a director who guides the suggests the LLM that can be treated as an actor. The LLM “act outs” a specific role or persona to solve problems or generate content. By “casting” the LLM in a role (e.g., “You are a historian…”) and providing a “script” (the detailed prompt), the prompter can elicit more structured, context-aware, and human-like responses. This framework emphasizes the prompter’s role in setting the scene, defining the character, and guiding the performance.

A skilled prompter often has a specific vision for the desired output. They use carefully chosen words, structures, and iterative refinements (prompt engineering) to steer the AI towards this vision. This can involve defining style, tone, subject matter, and even attempting to evoke specific emotions or ideas, much like a director outlines a scene.

Experienced prompters can develop recognizable patterns or approaches to prompting that yield particular types_of results from specific AI models. They learn the nuances of the AI and how to elicit desired aesthetics or textual qualities. Prompting is rarely a one-shot command. It often involves a back-and-forth conversation, a process of trial, error, and refinement, akin to a director working through multiple takes or editing choices. Even if the AI generates multiple options, the prompter often makes the final selection, curating the output that best aligns with their initial intent. This act of selection can be seen as a creative choice.

However, the US Copyright Office notes that AI models operate with a degree of unpredictability but concluded that content produced with AI can be copyrighted.

Key concerns of an auteur theory include determining who is the primary creative force, whether they have a personal vision and style based on a consistent aesthetic and thematic worldview, demonstrated mastery of the medium, recurring themes and motifs, and a willingness to push the boundaries of the medium. Finally, do they maintain significant control or influence over the final product?

The US Copyright Office’s stance that AI-generated works can be copyrighted with substantial human input further complicates this relationship. If the generated work is not copyrightable in the prompter’s name alone, or if the AI’s contribution is deemed substantial enough to negate sole human authorship, the prompter’s status as an “auteur” in a legally recognized sense is undermined.

However, current legal and creative consensus, notably highlighted by bodies like the US Copyright Office, generally holds that AI-generated works are not copyrightable unless there is substantial human creative input beyond mere prompting. The reasoning is that even detailed prompts can lead to unpredictable outputs from the AI, meaning the prompter may not have the same level of direct, granular control over the final work as a traditional artist. The AI model itself, with its complex algorithms and vast training data, plays a significant, if not primary, role in the generation process. The LLM’s output is inherently shaped by the massive datasets it was trained on, a factor far beyond the prompter’s control and introducing a vast external influence on the “style” and content.

By “carefully crafting prompts,” the user provides the model with context, instructions, and examples that help it understand the prompter’s intent and respond meaningfully. This includes setting clear goals, defining the desired length and format, specifying the target audience, and using action verbs to specify the desired action. Such detailed instruction suggests a high level of intentionality on the part of the prompter, aiming to steer the AI towards a preconceived vision.  

Arguments supporting a significant authorial role for the prompter often emerge from discussions about writers’ creative practices with AI. Studies indicate that writers utilize AI to overcome creative blocks, generate starting points, and then actively shape the AI’s output into something they consider useful, thereby maintaining a sense of ownership and control over the creative process. This active shaping and refinement, driven by the writer’s “authenticity” and desire for “ownership,” can be seen as analogous to an auteur’s imposition of their vision onto the raw materials of filmmaking.  

The Auteur Analogy Falters

Despite these points of correspondence, applying the auteur analogy to the AI prompter faces significant challenges. LLMs’ construction complicates the notion of a singular, controlling vision. These models are trained on vast datasets of existing human-created text and images and function by “mimicking human writing.” This training design makes it difficult to disentangle the prompter’s “pure” vision from the inherent capabilities, biases, and stylistic tendencies embedded within the LLM’s architecture and training data. AI may not be a neutral tool but an active, albeit non-conscious, participant in the generation process.  

This trajectory leads to the “black box” problem. The prompter rarely has full transparency or control over the internal workings of the LLM. While a film director ideally orchestrates various human and technical elements (cast, crew, script, camera), the prompter interacts with a system whose decision-making processes are often opaque. The output can sometimes be unpredictable, with LLMs even known to “hallucinate” or generate unexpected results, challenging the idea of complete authorial control.  

Intellectual property law presents another major hurdle. Current legal frameworks, particularly in jurisdictions like the United States, generally require human authorship for copyright protection. AI, as it stands, blurs the lines between authorship, ownership, and originality. If the generated work is not copyrightable in the prompter’s name alone, or if the AI’s contribution is deemed substantial enough to negate sole human authorship, the prompter’s status as an “auteur” in a legally recognized sense may be undermined.

The debate over prompter-as-auteur is thus deeply intertwined with these evolving legal definitions. The lawsuits involving creators like Scarlett Johansson and organizations like Getty Images and The New York Times against AI companies for using copyrighted material in training data further complicate the picture, as the very foundation upon which the LLM generates content is itself a site of contested authorship.  

Moreover, many instances of AI prompting might more accurately align with the role of the metteur en scène rather than the true auteur. A prompter might be highly skilled in eliciting specific outputs from an LLM, demonstrating technical competence. However, they may be seen as proficient technicians rather than visionary artists without a consistent, distinguishable personal style, thematic depth, or “interior meaning” traceable across a body of their AI-generated works. The inherent weaknesses often found in AI-generated writing—such as blandness, repetitiveness, or a lack of overarching logical structure — can also limit the perceived artistic merit of the output, thereby challenging the prompter’s claim to full auteurship if they are simply instructing a tool with such limitations.  

Furthermore, a significant aspect of the prompter’s role involves navigating the LLM’s potential for bias, inaccuracy, and “hallucinations.” This requires a curatorial-like responsibility to ensure that the information or content presented is sound, ethically considered, and appropriately contextualized. A museum curator has an ethical duty to research, authenticate, and provide accurate context for artifacts. Similarly, an AI prompter, especially in professional or public-facing applications, must critically vet and potentially correct or contextualize AI output to prevent the dissemination of falsehoods or biased information. This positions the prompter as a gatekeeper, quality controller, and interpreter of the AI’s output—all key curatorial functions.  

Despite these compelling parallels, the curator analogy also has its limitations when applied to AI prompting. Traditionally, curators work with existing, often tangible, artifacts or discrete pieces of information. LLMs, however, generate new content, albeit derived from their training data. The question then arises, is the prompter curating “generated” data, or are they more accurately co-creating it? This ambiguity blurs the line between curation and creation.

The “collection” from which an AI prompter “selects” is also fundamentally different from a museum’s holdings. An LLM’s latent space represents a near-infinite realm of potential outputs, not a predefined set of objects. This abundance makes the act of “selection” by prompting a more generative and less bounded process than choosing from a finite collection. The prompter is not merely selecting from a catalog but actively shaping what can be chosen or brought forth by the structure of their prompts. This suggests a more active, co-creative form of curation than traditional models imply, where the collection is dynamic and responsive to the prompter’s interaction.  

Acknowledging that the AI prompter’s role is not monolithic is crucial; it varies significantly based on the level of skill, labor, and “intellectual expression” invested, ranging from a simple user to a highly skilled co-creator. At one end of the spectrum, a user might issue simple, direct instructions to an LLM for a straightforward task, acting more as a client or basic operator. On the other hand, a highly skilled individual might engage in complex, iterative dialogues with AI, meticulously refining prompts and outputs in a process that resembles deep co-creation. The level of skill and labor or “intellectual expression” invested by the prompter can differ dramatically, and this variance directly impacts how their role is perceived and classified. A casual user asking an LLM to “write a poem about a cat” is performing a different function than an artist spending weeks crafting and refining prompts to achieve a specific aesthetic for a series of generated images or a legal expert carefully structuring queries to extract nuanced information for a case.  

Structural and Post-Structural Theoretical Challenges to Auteur Theory

Structuralism and its successor, post-structuralism, launched a powerful critique of the traditional notion of “authorship,” particularly challenging the idea of the author as the sole, authoritative source of a text’s meaning. Two of the most influential figures in this critique are Roland Barthes and Michel Foucault. Roland Barthes, in “The Death of the Author” (1967), argued for the “death of the author” as a key concept relevant to literary and textual analysis. His key arguments included that focusing on the author’s biography, intentions, or psychological state to interpret a text is a misguided and limiting practice. He believed that the author’s life and experiences are irrelevant to the meaning of the work once it is produced. He famously claimed that a text is not a linear expression of an author’s singular thought but rather “a multi-dimensional space in which a variety of writings, none of them original, blend and clash.” A text is a “tissue of quotations drawn from innumerable centers of culture,” meaning it comprises pre-existing linguistic conventions, cultural references, and discourses.

For Barthes, the meaning of a text is not fixed by the author but is produced in the act of reading. The reader is the one who brings together the various strands of meaning in the text. “The birth of the reader must be at the cost of the death of the Author.” This move liberates the text from a single, imposed meaning and opens it up to multiple interpretations. Barthes viewed the author not as a “creator” in the traditional sense, but as a “scriptor” – someone who merely sets down words, drawing from an already existing linguistic and cultural archive.

Michel Foucault’s “What is an Author?” (1969) shared Barthes’ skepticism about traditional authorship but approached the issue from a different angle, focusing on the historical and institutional construction of the “author-function.” His main point was that the “author” is not a natural or timeless entity but a specific function that emerged within certain historical and discursive practices. The author-function is a way of classifying, circulating, authenticating, and assigning meaning to texts within a given society.

The author is not a person, but a discursive principle. The author-function is not equivalent to the individual writer. It’s a set of rules and constraints that govern how we understand and use an author’s name. For instance, the author’s name serves as a mark of ownership (property), a way to hold someone responsible for transgressive statements, and a means to unify a body of diverse works.

Foucault linked the emergence of the author-function to systems of control and power, particularly the rise of copyright law and the need to regulate speech. Attributing a text to an author became a way to police discourse and assign responsibility, especially for subversive or dangerous ideas. Foucault emphasized that the author-function has not always existed or applied equally to all types of texts. For example, scientific texts often gain authority from their content rather than solely from their author, whereas literary texts are more heavily tied to the author’s name.

In summary, Barthes and Foucault challenged the romanticized, humanist view of the author as a solitary genius from whom all meaning emanates. Instead, they argued that meaning is either generated by the reader (Barthes) or shaped by complex social, historical, and institutional forces (Foucault), effectively “decentering” the author from their traditional position of authority. This critique had a profound impact across the humanities, including film studies, by shifting focus from the individual creator to the structures of language, discourse, and reception.

By applying Barthes and Foucault, we can argue that the LLM’s output is not a direct, unmediated expression of the prompter’s “genius” or sole intent. It is a complex interplay of the prompt, the LLM’s massive training data (a vast, unauthored “library” of human discourse), and its internal algorithms.

The “meaning” of AI-generated content is fluid and negotiated between the prompt, the LLM’s “knowledge,” and the audience’s interpretation. The emphasis on the prompter as “author” is often driven by practical, legal, and social needs to assign responsibility and fit new technology into old frameworks, rather than a true reflection of the creative process. The prompter is more accurately a skilled orchestrator of existing information and patterns or a “curator” and “editor” of potential outputs rather than a solitary inventor. This critical lens helps to move beyond a simplistic “human-as-master-of-machine” narrative, acknowledging the distribution of creativity and meaning-making in the age of AI.

Summary and Final Thoughts

Furthermore, as a visual studies and AI professor, I now incorporate generative AI, prompting questions about who the “author” is in AI-generated content. Traditionally, film studies grappled with authorship (the “auteur”), considering if the director, screenwriter, or others were the primary creative force. Similarly, the meaning of a film could be attributed to the author, the content, or the audience.

The emergence of the AI prompter—the human who crafts instructions for Large Language Models (LLMs)—creates a new debate. Is the prompter the “author” of the AI’s output? The text considers applying auteur theory to the prompter, noting similarities like the prompter’s vision, iterative refinement (prompt engineering), and selection process. It even uses the analogy of a “method actor” for the LLM, guided by the “director” (prompter).

However, the post argues that the auteur analogy could ultimately falter. LLMs, trained on vast human-created datasets, operate as “black boxes” with unpredictable outputs, making it difficult to attribute a singular, controlling vision to the prompter. Legal frameworks still largely require human authorship for copyright. The prompter might often function more like a “metteur en scène” (a competent technician) rather than a visionary auteur.

Instead, the post suggests the prompter’s role might be more akin to a curator, responsible for vetting, correcting, and contextualizing AI output due to the LLM’s potential for bias and “hallucinations.” However, this analogy also has limitations, as prompters generate new content rather than just selecting existing artifacts, making their role a blend of curation and co-creation.

Citation APA (7th Edition)

Pennings, A.J. (2025, Nov 1) The AI Prompter/Conversationalist as Auteur? Examining LLM Authorship Through the Lens of Film Theory. apennings.com https://apennings.com/books-and-films/the-ai-prompter-conversationalist-as-auteur-examining-llm-authorship-through-the-lens-of-film-theory/

Notes

[1] The notion of a prompter is being challenged because it is able to store conversations and responses.
[2] Okwuowulu, Charles. (May 2016). Auteur Theory and Mise-en-scence construction: A Study of Selected Nollywood Directors.
[3] Copyright and Artificial Intelligence
Part 3: Generative AI Training. Pre-publication version
A REPORT of the Register of Copyrights, May 2025

© ALL RIGHTS RESERVED

Not to be considered financial advice.

---

Anthony J. Pennings, PhD is a Professor at the Department of Technology and Society, State University of New York, Korea and a Research Professor for Stony Brook University. He teaches Visual Rhetoric, AI, and broadband policy. From 2002-2012 he taught digital economics and digital media management at New York University. He also taught in the Digital Media MBA at St. Edwards University in Austin, Texas, where he lives when not in Korea.

Related

Category: artificial intelligence, books and films, intellectual property, social media, visual literacy and rhetoric
Tags: Auteur > Auteur Theory > author-function > Barthes > Cahiers du Cinéma > LLMs

Comments

Comments are closed.