<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Publishing DTD v1.2 20120330//EN" "http://jats.nlm.nih.gov/publishing/1.2/JATS-journalpublishing1.dtd">
<!--<?xml-stylesheet type="text/xsl" href="article.xsl"?>-->
<article article-type="research-article" dtd-version="1.2" xml:lang="en" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">
<front>
<journal-meta>
<journal-id journal-id-type="issn">2056-6700</journal-id>
<journal-title-group>
<journal-title>Open Library of Humanities</journal-title>
</journal-title-group>
<issn pub-type="epub">2056-6700</issn>
<publisher>
<publisher-name>Open Library of Humanities</publisher-name>
</publisher>
</journal-meta>
<article-meta>
<article-id pub-id-type="doi">10.16995/olh.28634</article-id>
<article-categories>
<subj-group>
<subject>Contemporary perspectives on AI and narrative</subject>
</subj-group>
</article-categories>
<title-group>
<article-title>The Development of Vernacular Creativity in the Context of the Culture of Neural Networks</article-title>
</title-group>
<contrib-group>
<contrib contrib-type="author">
<contrib-id contrib-id-type="orcid">https://orcid.org/0000-0001-5727-289X</contrib-id>
<name>
<surname>Pioreck&#253;</surname>
<given-names>Karel</given-names>
</name>
<email>piorecky@ucl.cas.cz</email>
<xref ref-type="aff" rid="aff-1">1</xref>
</contrib>
<contrib contrib-type="author" corresp="yes">
<contrib-id contrib-id-type="orcid">https://orcid.org/0000-0002-0022-3322</contrib-id>
<name>
<surname>Hus&#225;rov&#225;</surname>
<given-names>Zuzana</given-names>
<prefix>Prof.</prefix>
</name>
<email>zuz.husarova@vsvu.sk</email>
<xref ref-type="aff" rid="aff-2">2</xref>
</contrib>
</contrib-group>
<aff id="aff-1"><label>1</label>Institute of Czech Literature of the Czech Academy of Sciences, Prague, The Czech Republic</aff>
<aff id="aff-2"><label>2</label>Digital Arts, Academy of Fine Arts and Design in Bratislava, Bratislava, Slovakia</aff>
<pub-date publication-format="electronic" date-type="pub" iso-8601-date="2026-10-06">
<day>06</day>
<month>10</month>
<year>2026</year>
</pub-date>
<pub-date pub-type="collection">
<year>2026</year>
</pub-date>
<volume>12</volume>
<issue>2</issue>
<fpage>1</fpage>
<lpage>30</lpage>
<permissions>
<copyright-statement>Copyright: &#x00A9; 2026 The Author(s)</copyright-statement>
<copyright-year>2026</copyright-year>
<license license-type="open-access" xlink:href="http://creativecommons.org/licenses/by/4.0/">
<license-p>This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International License (CC-BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. See <uri xlink:href="http://creativecommons.org/licenses/by/4.0/">http://creativecommons.org/licenses/by/4.0/</uri>. Articles published by the Open Library of Humanities may also contain other CC-licensed material, or material used or reproduced under an exception or limitation to copyright. Such material is not subject to the article&#8217;s license. It is either marked with its own license, or marked that it has been reproduced with permission or under fair-use.</license-p>
</license>
</permissions>
<self-uri xlink:href="https://olh.openlibhums.org/articles/10.16995/olh.28634/"/>
<abstract>
<p>The literary research paper proposes the concept of <italic>the reskilling phase</italic> into the research oriented at the implementation of generative artificial intelligence in the writing process, especially in the context of fiction writing. <italic>The reskilling phase</italic> does not refer to the disappearance of authorial competence, but rather to its structural rebuilding under conditions of human&#8211;machine co-production. The paper places this phase within the framework of digital culture as a continuation and intensification of tendencies established by the Web 2.0 paradigm and analyzed by its key theoreticians. While presenting vernacularization and theoretical groundings of Web 2.0 as the initial conceptual frameworks, we study the consequences of the vernacularization of cultural production on the debates about the contexts of artificial neural networks in literature and the arts in the contemporary era. We also look into the periods that trace the gradual embedding of LLMs into the artistic practice.</p>
<p>We believe that the mutual relationship between the Web 2.0 paradigm and generative AI provides a suitable potential for a thorough understanding of the consequences of generative AI on our culture. The study presents an empirical dataset drawn from a sample of 23 amateur authors who participated in a Czech short story writing contest. A contextual example from Slovak public radio provides a supplementary example. By including the dataset, we aimed to advance a broader theoretical argument about the redistribution of literary competence and its delegation to the computational model.</p>
</abstract>
</article-meta>
</front>
<body>
<sec>
<title>1. Introduction</title>
<p>The historical development of media can be observed as a long-term process of its democratization, understood here by the concept of <italic>vernacular creativity</italic>, as articulated by Jean Burgess (<xref ref-type="bibr" rid="B6">2007: 30</xref>). Burgess uses the term to denote a culture of autodidacts &#8212; non-elite, popular cultural production that emerges from everyday life and leisure activities outside the institutional frameworks of the arts (see <xref ref-type="bibr" rid="B6">Burgess, 2007: 30</xref>).</p>
<p>The dynamic of gradual removal of barriers to accessing media as tools for human involvement and, more specifically, for creative expression is well documented, and discussions on the subject typically begin with the invention of the printing press and its fundamental role in the democratization of access to education via books, and the dissemination of creative writing in fixed textual form.</p>
<p>Technological and media shifts of this magnitude are rare in history. Following Gutenberg&#8217;s invention, there were only partial, incremental transformations. In 1936, Walter Benjamin asserted in his essay &#8216;The Work of Art in the Age of Mechanical Reproduction&#8217; that, due to the development of print culture in Europe, &#8216;the distinction between author and public is about to lose its basic character. The difference becomes merely functional; it may vary from case to case. At any moment the reader is ready to turn into a writer&#8217; (<xref ref-type="bibr" rid="B5">Benjamin, 1969: 232</xref>).</p>
<p>Our aim here, however, is not to trace the historical trajectory of media development or to provide a detailed diachronic account of its democratization. Rather, we evoke this developmental context as a backdrop for a shift in analytical focus toward the present &#8212; specifically, to the question of the role that generative AI plays in the current phase of cultural democratization, with particular emphasis on literary culture. We also ask: in what ways does this emerging phase of digital culture differ from its earlier configurations, especially the era of so-called Web 2.0, which was closely tied to the ethos of democratization?</p>
<p>Our methodological approach to addressing these questions is grounded in the comparison and typological framing of manifestations of vernacular creativity across the phases of the cultural democratization process. A more detailed analysis of specific literary phenomena associated with vernacular creativity in the Web 2.0 context has been undertaken by one of the co-authors of this study in earlier works (e.g. <xref ref-type="bibr" rid="B16">Pioreck&#253;, 2017</xref>). Here we supplement that foundation with an analysis of the literary behavior of participants in a short story writing competition using generative AI, offered as a window into the literary practice of contemporary vernacular authors. The aim, however, is not to provide a comprehensive data analysis, because the dataset obtained would not permit one, but rather to illustrate the range of attitudes and strategies that may be encountered among these authors.</p>
</sec>
<sec>
<title>2. Framing of the Terms &#8216;Vernacular Culture&#8217; and &#8216;Vernacular Creativity&#8217;</title>
<p>Before we can proceed, however, it is necessary to clarify several key concepts. While we reference and partially paraphrase the term &#8216;democratization&#8217;, we ultimately distance ourselves from its use as a conceptual anchor. We find it problematic, chiefly due to its political connotations. Instead, we prefer the term <italic>vernacularization</italic>, derived from concepts such as <italic>vernacular culture</italic> and <italic>vernacular creativity</italic>, as suggested by the already mentioned positions by Jean Burgess and Henry Jenkins.</p>
<p>Henry Jenkins, in his book <italic>Convergence Culture</italic>, uses the term <italic>vernacular culture</italic> to describe culture produced by amateurs, arguing that new digital tools and distribution networks have strengthened the capacity of ordinary individuals to actively participate in cultural production. As Jenkins idealistically put it in the era of foreseeing the future with brighter perspectives (which has not proven that obvious from today&#8217;s position regarding the social divisions of power): &#8216;As soon as subculture and fan groups have had a taste of that power once, they will never again become obedient and invisible&#8217; (<xref ref-type="bibr" rid="B9">Jenkins, 2006: 162</xref>). Henry Jenkins describes <italic>vernacular culture</italic> in his glossary of the book <italic>Convergence Culture</italic> as &#8216;a culture that is generated by amateurs, a term intended to suggest the parallels between folk culture and fan culture&#8217; (<xref ref-type="bibr" rid="B9">2006: 293</xref>). This term originally referred to the situation associated with the rise of Web 2.0 and the participatory processes it introduced into cultural and media environments. However, its relevance extends to trends within the artistic and design fields. The increased accessibility of neural networks to the general public has significantly impacted the cultural sphere, particularly during the AI Spring from 2022 onward. This period marked the massive popularization of textual, visual, auditory, and multimodal AI generators. Jenkins&#8217;s phrase &#8216;a culture that is generated&#8217; was meaningfully broader than simply being generated by programs. Yet, in our context, the term &#8216;generated&#8217; effectively captures both meanings &#8211; as a culture produced by amateurs as well as cultural artefacts that are generated by AI that the amateurs produce with their prompts.</p>
<p>We also draw upon the concept of <italic>vernacular creativity</italic>, as articulated by Jean Burgess, who uses the term to denote a culture of autodidacts &#8212; non-elite, popular cultural production that emerges from everyday life and leisure activities outside the institutional frameworks of the arts (see <xref ref-type="bibr" rid="B6">Burgess, 2007: 30</xref>).</p>
<p>For the purposes of our analysis, it is crucial to delineate the concept of the <italic>vernacular author</italic>, in addition to invoking Jean Burgess&#8217;s concept of <italic>vernacular creativity</italic> (see <xref ref-type="bibr" rid="B6">Burgess, 2007: 30</xref>). Simply equating the category of the vernacular author with the notion of the amateur author would be unproductive (a point Burgess herself underscores), not least because its supposed binary opposite &#8212; the professional author &#8212; is itself an unstable and context-dependent construct. The distinctiveness of the vernacular author, as we understand it, lies in their particular relationship to literary competence.</p>
<p>Jonathan Culler defines literary competence as the foundational capacity to interpret texts, contingent upon an understanding of the conventions governing literary discourse (<xref ref-type="bibr" rid="B7">Culler, 1975: 114</xref>). This includes awareness of the fact that literary works are embedded within particular traditions, adhere to (or subvert) generic conventions, are situated within both literary and extraliterary contexts, and maintain intertextual relationships &#8212; whether intentional or unintentional. Moreover, the reception and production of literature are governed by culturally specific value systems, including mastery of literary language, the practice of editing, and adherence to aesthetic norms.</p>
<p>In refining this notion, we adopt the broader conception of literary competence proposed by Johan Svedjedal in <italic>The Literary Web</italic>. Unlike Culler, Svedjedal extends the term to encompass both ends of the literary communication process. Literary competence, in this sense, is essential not only for readers interpreting texts but also for authors who actively engage in the literary field. To function &#8216;competently&#8217; as authors, they must internalize the operative rules of literary discourse, while also having the opportunity to challenge and redefine those same rules (<xref ref-type="bibr" rid="B25">Svedjedal, 2000: 106</xref>).</p>
<p>We therefore define <italic>vernacular literature</italic> as the creative output of authors who have either not fully acquired these discursive conventions or have only partially done so (either by ignoring them or without a deeper desire to engage with them). This may be due to factors such as young age or because the acquisition of these rules exceeds their individual intentions. At the same time, it is important to acknowledge that literary production in the context of the culture of neural networks often presupposes a range of (seemingly non-literary) competencies &#8212; such as technological literacy or familiarity with digital interfaces &#8212; that vernacular authors may in fact possess to a high degree.</p>
</sec>
<sec>
<title>3. Vernacularization in the Era of Web 2.0</title>
<p>In order to assess the specificity of vernacular creativity within the context of the culture of neural networks, it is necessary to consider the forms and consequences of the vernacularization process as it unfolded during the era of so-called Web 2.0, a phase that can be understood as a logical precursor to the current state of digital culture.</p>
<p>The proliferation of emergent digital technologies during the era of Web 2.0, coupled with increasingly accessible modes of content distribution, significantly expanded the capacity of non-professional users to engage in, shape, and disseminate cultural production. As Henry Jenkins argues, these developments catalyzed the rise of a participatory culture, wherein amateurs and independent creators gained unprecedented opportunities to contribute to the cultural sphere (<xref ref-type="bibr" rid="B9">Jenkins, 2006</xref>).</p>
<p>Central to this transformation were the concepts of interaction and participation, which became foundational to digital culture and its corresponding artistic practices. These principles were especially influential in digital art discourses that emerged during Web 2.0, wherein play and game-like structures were not merely aesthetic features but fundamental organizing principles.</p>
<p>The ludic dimension of digital culture found expression in a range of hybrid artistic forms, such as machinima, MUDs (Multi-User Dungeons), MOOs (MUD, Object-Oriented), interactive fiction (IF), gamebooks, and collaborative online writing platforms. These developments demonstrated a convergence of the poetics and aesthetics of computer and video games with digital artistic production, signaling a shift toward participatory modes of cultural authorship.</p>
<p>This convergence culture, as articulated by Jenkins (<xref ref-type="bibr" rid="B9">2006</xref>), reconfigured the traditional boundaries between producers and audiences. It foregrounded interactivity, collaboration, and distributed authorship as core values of cultural creation. Key theorists of Web 2.0 culture proposed, among others, understandings of Web 2.0 as a culture that provides a &#8216;cognitive surplus&#8217; of online users (<xref ref-type="bibr" rid="B23">Shirky, 2010</xref>), a &#8216;hybrid economy&#8217; in which commercial and non-commercial forms of cultural production coexist, fostering a more participatory and creatively engaged society (<xref ref-type="bibr" rid="B12">Lessig, 2008</xref>), or a framework of &#8216;spreadability&#8217; to describe how and why media circulates within digital ecosystems (<xref ref-type="bibr" rid="B10">Jenkins, Ford and Green, 2013</xref>). Roberto Simanowski identifies Web 2.0 with the principle of UGC (User-Generated Content). He notes that this term was first introduced by journalists and later adopted by media theorists to describe a phase in the development of the web in which content creation was no longer reserved for those with technical knowledge and skills (such as programmers) but was instead accessible to any internet user (<italic>cf</italic>. <xref ref-type="bibr" rid="B24">Simanowski, 2008: 75</xref>). He also discusses concepts such as horizontal knowledge and swarm intelligence, which denote forms of human cognition and knowledge production emerging in a context of maximal connectivity and new modes of collectivity (at the time also referred to as &#8216;collective intelligence&#8217;).</p>
<p>Horizontal knowledge involves the active linking and collection of information by individual internet users. The knowledge base of AI rests on the aggregation of textually encoded data, often gathered without the awareness or intentional input of its originators, followed by machine-driven modelling based on that data. The principle of swarm intelligence, as otherwise referred to by the concept of &#8216;hive mind&#8217;, was grounded in the cooperation of large communities of active users with all the advantages and drawbacks associated with &#8216;crowd thinking&#8217;. In contrast, artificial intelligence based on large language models operates on the principle of simulating human thought through the imitation of linguistic expression.</p>
<p>The logic of Web 2.0 was fundamentally predicated on an implicit dichotomy between creators (of software) and users (of these software tools or environments). Its backdrop was the earlier form of the web, in which both its software and its content were produced exclusively by those with the technical skills to do so. This was a dichotomy between two complementary groups of human actors jointly participating in the functioning of digital culture. According to Tim O&#8217;Reilly, who is often regarded as the author of the term Web 2.0, the core of this phenomenon is the transition from static websites to interactive platforms that rely on user activity&#8212;activity through which content is produced, as well as specific forms of value (collective intelligence). In his article, O&#8217;Reilly lists blogs among the key manifestations of Web 2.0, even though their origins reach deep into the previous stage of web development: &#8216;One of the most highly touted features of the Web 2.0 era is the rise of blogging&#8217; (<xref ref-type="bibr" rid="B15">2007: 24</xref>). Its accessibility allowed vernacular creativity to gain a publishing space of a scale previously unimaginable. In addition to blogs themselves, the blogosphere includes (particularly in the Czech and Slovak context) amateur literary web forums and, eventually, mainstream social networks. In this sense, Web 2.0 was frequently associated with a radical democratization of authorship&#8212;an outcome largely attributed to the interactive nature of the medium in which this democratizing process unfolded. However, Ram&#243;n Reichert also addresses in his book <italic>Amateure im Netz: Selbstmanagement und Wissenstechnik im Web 2.0</italic> [<italic>Amateurs on the Net: Self-Management and Knowledge Technology in Web 2.0</italic>] (<xref ref-type="bibr" rid="B20">2008</xref>) the trivialization of public discourse, necessitating the protection of professionals as a cultural elite (<italic>cf</italic>. <xref ref-type="bibr" rid="B20">Reichert, 2008: 9&#8211;10</xref>).</p>
<p>It would be misguided to view the blogosphere as a distinct or isolated realm separate from the broader literary culture. On the contrary, it became &#8212; particularly for authors of Generation Y &#8212; a primary platform through which they entered into literary communication. For instance, among Czech poets born in the late 1970s and 1980s, it is difficult to find authors who did not begin their literary journeys through some form of online literary forum. These forums effectively functioned as a &#8216;zero layer&#8217; of the literary system &#8212; an informal threshold through which one could enter the field of literature. Whereas in the print era, platforms for amateur literary publication (such as specialized magazine columns) adhered to the conventions of literary discourse and served to disseminate literary competence, in the case of internet literary forums, this dynamic was often reversed. Vernacular literary creativity contributed to the presentation of literature as a form of leisure activity, pursued by creators who understood literary texts primarily as private messages rather than as aesthetic or social phenomena.</p>
<p>The literary system, as conceptualized by Siegfried Schmidt, expanded during the Web 2.0 era to include a quantitatively significant zone of vernacular creativity. Schmidt defines the literary system as a structure governed by socially binding conventions (<italic>cf</italic>. <xref ref-type="bibr" rid="B22">Schmidt, 2008: 47</xref>). From this perspective, we may distinguish three internally differentiated zones within the literary system of the Web 2.0 era:</p>
<list list-type="simple">
<list-item><p>A) The center (the &#8216;elite&#8217;), comprising actors in literary communication who not only master the binding conventions but are also capable of transforming them;</p></list-item>
<list-item><p>B) The intermediate zone (the &#8216;gray area&#8217;), consisting of actors who have acquired these conventions but lack the capacity to innovate them;</p></list-item>
<list-item><p>C) The periphery (the &#8216;vernacular zone&#8217;), inhabited by those who have only partially mastered literary conventions and are thereby at risk of exclusion from the system.</p></list-item>
</list>
<p>To this outermost, peripheral ring of concentric circles &#8212; historically populated by novice authors and literary amateurs with limited publishing opportunities &#8212; a vast new ring of literary acts emerged under Web 2.0 conditions, realized within the open-access digital platforms of the era (i.e., literary forums, blogs, social media accounts, etc.). Crucially, for our subsequent analysis, the vernacular zone of literature during the Web 2.0 period came to be embodied by authors whose literary competence was not self-evident, but could be comparatively assessed against the competence levels of those whose work belonged to the intermediate or central zones of the literary system.</p>
</sec>
<sec>
<title>4. The Vernacular Phase of Neural Networks&#8217; Establishment in Culture</title>
<p>The advent and rapid proliferation of language models such as GPT-3 and its subsequent development as ChatGPT, accompanied by their wide user adoption, has introduced a novel phase in the vernacularization of literary authorship, one that has yet to be fully theorized within the discourse of digital literature. By this, we do not mean the theoretical foundations that this technology was given as early as the 1950s, but rather the period when, thanks to hardware with sufficient computational power, it began to enter practical use. This process of generative neural networks becoming established in cultural practice can be traced roughly from the beginning of 2020s onward. Although this is a relatively short time span, it is useful to recall how, even within this period, neural networks evolved &#8212; specifically, how the motivations for their use and the modes of their application in cultural production changed, and how access to them gradually became democratized.</p>
<p>This latest development marks a paradigmatic shift insofar as the production of literary texts is no longer contingent upon the possession of conventional writing competencies. Instead, the generative capabilities of neural networks enable users, regardless of literary training, to create textual artefacts through the simple act of prompt formulation. In this context, vernacularization denotes the increasing participation of amateur users in the literary field, wherein the threshold for entry has been radically lowered. As a result, virtually any individual with access to these tools can generate and publish texts that may be recognized as literary, without the need for prior experience or formal expertise in literary practice.</p>
<p>We draw here on our book <italic>The Culture of Neural Networks</italic> (<xref ref-type="bibr" rid="B17">Pioreck&#253; and Hus&#225;rov&#225;, 2024</xref>), in which we have divided this process into the following phases:</p>
<p><bold>1. The Verification Phase:</bold></p>
<p>In this phase, the leading role was not played by artists, but by developers and programmers who used projects with literary characteristics to test the functionality of algorithms and the adequacy of the language models then available. From the Czech context, an illustrative example is the automatic poet project created by mathematical linguist and machine learning specialist Ji&#345;&#237; Materna in 2015. He used a recurrent neural network (RNN) trained on texts sourced from a website for amateur poets. The generated outputs were virtually indistinguishable from their human-written counterparts. Another notable example of this phase was the Czech project <italic>TheAItre</italic>, which resulted in the generation of a theatrical play. This project involved collaboration between scientists from Charles University and artists from &#352;vanda Theatre.</p>
<p><bold>2. The Artistic-Subversive Phase:</bold></p>
<p>In this phase, artists assumed the leading role in conceptualizing generative projects. The focus shifted from testing technological capabilities to realizing conceptual intentions &#8212; often formulated as subversive towards literary norms or the very technologies and media practices they were engaging with. A striking example of this subversive artistic use of neural networks is the project <italic>Artificial Unintelligence</italic> by Slovak multimedia artist Samuel Szab&#243; (aka Sam&#269;o). In this work, a char-RNN network was trained on traditional Slovak nationalist (so-called &#8216;&#318;ud&#225;cke&#8217;) poetry, but was deliberately undertrained, so that the resulting poems simultaneously mimicked and parodied this style. Somewhat less subversive was the project of the fictional Slovak feminist poet Liza Gennart. The book-length output, a collection of poems generated using GPT-2 fine-tuned on over 2000 books of predominantly Slovak poetry, titled <italic>V&#253;sledky vzniku</italic> [<italic>Outcomes of Origin</italic>], earned its human authors, Zuzana Hus&#225;rov&#225; and &#317;ubom&#237;r Pan&#225;k, Slovakia&#8217;s national poetry prize Zlat&#225; vlna [<italic>Golden Wave</italic>] in 2021.</p>
<p><bold>3. The Vernacular Phase:</bold></p>
<p>This phase is characterized by radically democratized access to neural networks, which enabled anyone with an internet connection to generate literary texts or other cultural artifacts. Generative literature thus became part of pop culture and amateur literary creativity, losing its previous scientific, technological, or artistic exclusivity. Shortly after the launch of ChatGPT, a proliferation of poetry generators emerged, interfaced with this popular chatbot via API, giving anyone the opportunity to generate verse in a specific genre, with selected themes. Thanks to easily accessible self-publishing platforms such as Amazon, poetry collections began to appear on the book market, authored by individuals with no prior participation in literary life. These publications share two main characteristics: they enter the context of generative literature without their authors being aware of the genre&#8217;s nature or history; at the same time, they express a strong desire to partake in the symbolic capital still associated with poetry and the book as an artifact.</p>
<p>The research study &#8216;AI and the Quantity and Quality of Creative Products: Have LLMs Boosted Creation of Valuable Books?&#8217; by Imke Reimers and Joel Waldfogel from 2026 presents a thorough quantitative and qualitative analysis of the influence of LLMs on the book market in general, without a specific focus on genre. Their findings indicate that &#8216;[m]ore than 40 percent of the 2023&#8211;2025 releases of those authors with no pre-LLM-era books &#8211; the &#8220;entrants&#8221; &#8211; contain AI&#8217; (<xref ref-type="bibr" rid="B21">2026: 24</xref>). They further provide data-based evidence documenting &#8216;a tripling in the number of new books coming to market between late 2022 and late 2025 that mirrors the use of AI that we detect in new books&#8217; (<xref ref-type="bibr" rid="B21">2026: 27</xref>). Their analyses offer a thorough account of the behavior of the book market, considering the impact of LLMs on the whole literary system.</p>
<p>The relationships between our proposed three phases are summarized in the following scheme (<xref ref-type="fig" rid="F1">Figure 1</xref>).</p>
<fig id="F1">
<caption>
<p><bold>Figure 1:</bold> The three phases of the establishment of neural networks in culture. For each phase the figure summarizes the dominant actors, the primary motivation for using generative technologies, the typical technologies employed, the literary role attributed to AI, and representative projects. Figure by the authors.</p>
<p>Note: The boundaries between phases are porous and overlapping. Projects can belong to multiple logics simultaneously.</p>
</caption>
<graphic xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="olh-12-2-28634-g1.png"/>
</fig>
<p>It should be emphasized here that these phases are not understood as discrete, sharply bounded epochs that alternate with and exclude one another, but rather as a heuristic periodization: each phase designates a dominant configuration of actors, motivations, and modes of using generative technologies, not a closed time period. The individual phases overlap and coexist &#8212; verification projects continue to emerge even after the onset of the vernacular phase (developers still test new models on literary tasks today), and artistic-subversive strategies remain a living part of contemporary creative practice. The delimitation of these phases therefore has the status of an analytical tool: it makes it possible to describe how the center of gravity of the cultural appropriation of neural networks has shifted from technological verification through artistic conceptualization to mass amateur practice, and against this background to delineate the specifics of the vernacular phase, which is the primary concern of this study.</p>
<p>In <italic>Artificial Aesthetics</italic>, Emanuele Arielli, in collaboration with Lev Manovich, foregrounds the socio-cultural dynamics of reception over the intrinsic properties of the generated content itself. His analysis privileges the contextual frameworks through which such works are perceived, with particular emphasis on the construction of cultural capital and the symbolic value that emerges through collective processes of interpretation and validation: &#8216;Therefore, the cultural and social acceptance of AI-generated artefacts will also depend on how much cultural capital (using Pierre Bourdieu&#8217;s terminology) we will attribute to synthetic media of this kind or to the artist experimenting with them. It won&#8217;t matter what such systems will be capable of generating, but what symbolic significance will be ascribed to their productions&#8217; (<xref ref-type="bibr" rid="B2">Arielli and Manovich, 2022: 23</xref>).</p>
<p>Considering the questions &#8216;What, then, is the specificity of cultural capital and the nature of vernacular creativity in the era of generative AI, and how does it differ from the forms it took within the framework of Web 2.0?&#8217; we proceed to analyze examples of AI-assisted creative writing from the Czech and Slovak cultural environment. We understand the AI short-story writing competition analyzed below as a typical manifestation of the vernacular phase; at the same time, we ask whether the shifts in competences we observe within it already point toward the following phase &#8212; the reskilling phase, which we conceptualize in the final part of the study.</p>
</sec>
<sec>
<title>5. Vernacular Literary Practice in AI-Assisted Creative Writing</title>
<p>In order to answer the aforementioned questions, we must first gain some insight into the literary practices of vernacular authors in this new context. In the case of Web 2.0 and its characteristic swarm intelligence, such insight was possible thanks to publicly accessible conversations within literary communities, in which vernacular creativity was cultivated. Studying vernacularization in the context of artificial intelligence is more challenging in this regard, as the relevant interactions (between user and AI) are generally not publicly accessible. We typically encounter only the final result of generation, which may reveal little about the process of either assisted or synthetic creation.</p>
<sec>
<title>5.1 Endless Stories (Written by Artificial Intelligence)</title>
<p>This limited transparency also becomes apparent in the project <italic>Nekone&#269;n&#233; pr&#237;behy (p&#237;san&#233; umelou inteligenciou)</italic> [<italic>Endless Stories (Written by Artificial Intelligence)</italic>], initiated in September 2024 by R&#225;dio Slovensko, a station of Slovak public radio. Due to the growing popularity of AI in the field of audiobooks, public broadcasters have begun to incorporate vernacular elements into their programming, namely, audience participation in the creation of generative content broadcast on air. For example, listeners of Slovak Radio were invited to submit 10 words, which served as prompts for the generation of two-minute short stories. The project website emphasized the combination of human creativity and the technical capabilities of AI. The stories were generated using ChatGPT; however, no specific details were provided regarding the generation process, the editorial selection of individual stories, or the identities of those who generated or curated the outputs. Only the names of the actors who narrated the texts were mentioned. The actual prompts used to generate the stories were also omitted &#8212; ChatGPT cannot produce a coherent short story from a mere list of 10 words alone. This example illustrates a general tendency to emphasize the technological prowess of ChatGPT while simultaneously enhancing the appeal of radio programming through audience engagement. According to the project&#8217;s description, listeners &#8216;can participate in creation&#8217; (<xref ref-type="bibr" rid="B19">R&#225;dio Slovensko, 2024</xref>), yet the concept of creative work with generative AI entails far more than submitting 10 words. It involves careful prompt engineering, curatorial decisions, and potential editing. Since such processes are not typically made transparent, we sought to collect data that would offer insight into the interactions between a specific group of vernacular authors and generative AI.</p>
</sec>
<sec>
<title>5.2 An AI-musing Crime</title>
<p>The insight into the aforementioned interactions was made possible through collaboration with the organizers of the literary competition <italic>Z-AI-mav&#253; zlo&#269;in</italic> [<italic>An AI-musing Crime</italic>], who provided us with 23 documentation protocols submitted by contest participants. This is a complete dataset of all contest participants who, by submitting their entries, consented to the processing of their data for research purposes. These included information about the authors&#8217; literary and technological backgrounds, their step-by-step generative processes, and complete transcripts of their conversations with AI. The competition, held in 2024, was organized by the Municipal Library in Prague in cooperation with generative AI specialists from the Institute of Formal and Applied Linguistics at Charles University. The only requirements for participation were adherence to the detective fiction genre and submission of documentation demonstrating the generative process used to produce the story.</p>
<p>The empirical basis of the study thus consists of 23 documentation protocols produced by participants of the aforementioned competition. Each protocol has two parts. The first is an author questionnaire, in which participants answered questions about: to what extent they feel they are the author of the resulting text (as a percentage); how much they know about AI; what experience they have writing fiction; which AI model they used; and what it was like working with AI. The second part is a complete record of communication with the chatbot &#8212; either in the form of an inserted transcript, a link to a shared conversation, or screenshots. The scope of the records varies from a single prompt with a few corrections to dozens of dialogue exchanges; this heterogeneity of documentation is a given of the dataset, and we reflect on it below among the study&#8217;s limitations. Individual stories and their documentation protocols are marked with uniform numerical codes 1&#8211;23, which are also given in the first column of <xref ref-type="table" rid="T1">Tables 1</xref> and <xref ref-type="table" rid="T2">2</xref>; excerpts cited in the text can thus be related to specific rows of the tables. The story&#8217;s code is not identical to its final placement in the competition.</p>
<p>According to the competition rules, all competing stories were written in Czech, in the detective or thriller genre, with a maximum length of 5,400 characters including spaces. The prompts were formulated predominantly in Czech (20 protocols), and in three cases in English. GPT-4o appeared (alone or in combination with another tool) in 19 protocols, Claude 3.5 Sonnet in four, Gemini in two; in one case GPT-4o was combined with the Bing search engine. It also mattered whether participants worked with the paid or free version of the chatbot: the paid versions allowed, for example, creating an assistant with a knowledge base, while the free versions limited the number of prompts and led to more economical strategies.</p>
<p>The quantitative overview in <xref ref-type="table" rid="T1">Tables 1</xref> and <xref ref-type="table" rid="T2">2</xref> was created by coding the protocols into 12 variables. Based on the free-text answers in the questionnaire, we converted the data on experience with AI and with literary writing into a four-point scale (no experience &#8211; beginner &#8211; advanced &#8211; expert); the declared share of AI in the text is derived from the answer to the question about perceived authorship. From the conversation transcripts we further inferred: the human or machine origin of the story idea (whether the participant proposed the story&#8217;s theme or had it generated); the prompting approach, where we distinguish zero-shot (the text is generated essentially by a single comprehensive prompt, followed at most by minor corrections) and an iterative approach (the story is developed gradually through a series of dialogic steps); the origin of the story&#8217;s title; imitation of an authorial style (an explicit request to write in the style of a specific author or school); persona prompting (assigning the model a role, e.g. &#8216;the best detective-fiction writer in the world&#8217;); and the presence of a final human edit of the text.</p>
<p>The core of the analysis is a qualitative, thematically oriented interpretation of the complete conversation transcripts. We read through the protocols repeatedly, and inductively abstracted from them recurring patterns of interaction (delegation of literary competence, the game-like character of the dialogue, the affirmative communication loop, perception of authorship), which subsequently structured the interpretation in the following part of the study. Examples 1&#8211;4 were selected as typical instances of these patterns, not as exceptional curiosities; we quote them verbatim, and in anonymized form.</p>
<p>The dataset has clear limitations that determine the scope of our conclusions. The sample is small and specific both locally and by genre (a Czech detective short story in a single competition); the data on experience and on the share of AI involvement are self-declared and &#8212; as we show below &#8212; difficult to compare; for some protocols, the communication is available only via a link to a shared conversation, whose permanence and completeness cannot be guaranteed. We therefore do not understand the corpus as a representative sample of vernacular writing with AI, but rather as a theoretically productive set of examples, whose function is to illustrate our theoretical proposals, not to test them.</p>
<p>The following table provides basic data about the competition participants:</p>
<table-wrap id="T1">
<caption>
<p><bold>Table 1:</bold> Profile of the 23 entries submitted to the <italic>Z-AI-mav&#253; zlo&#269;in</italic> [<italic>An AI-musing Crime</italic>] competition.</p>
</caption>
<table>
<tbody>
<tr>
<td align="left" valign="top" colspan="7"><bold>Participants of the &#8216;<italic>Z-AI-mav&#253; zlo&#269;in</italic>&#8217; competition</bold></td>
</tr>
<tr>
<td align="left" valign="top"><bold>Short story code</bold></td>
<td align="left" valign="top"><bold>Final ranking of the submission</bold></td>
<td align="left" valign="top"><bold>Model used</bold></td>
<td align="left" valign="top"><bold>Declared share of AI in text creation</bold></td>
<td align="left" valign="top"><bold>Prompt language</bold></td>
<td align="left" valign="top"><bold>Experience with AI</bold></td>
<td align="left" valign="top"><bold>Experience with literary writing</bold></td>
</tr>
<tr>
<td align="left" valign="top">6</td>
<td align="left" valign="top">1</td>
<td align="left" valign="top">Claude 3.5 Sonnet</td>
<td align="left" valign="top">99%</td>
<td align="left" valign="top">English</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">20</td>
<td align="left" valign="top">2</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">35%</td>
<td align="left" valign="top">English</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Expert</td>
</tr>
<tr>
<td align="left" valign="top">10</td>
<td align="left" valign="top">3</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">30%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">Expert</td>
</tr>
<tr>
<td align="left" valign="top">15</td>
<td align="left" valign="top">4</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">45%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">8</td>
<td align="left" valign="top">5</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">15%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">Advanced</td>
</tr>
<tr>
<td align="left" valign="top">22</td>
<td align="left" valign="top">6</td>
<td align="left" valign="top">GPT-4o + Bing</td>
<td align="left" valign="top">50%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">23</td>
<td align="left" valign="top">7</td>
<td align="left" valign="top">Claude 3.5 Sonnet + GPT-4o</td>
<td align="left" valign="top">5%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">19</td>
<td align="left" valign="top">8</td>
<td align="left" valign="top">Gemini</td>
<td align="left" valign="top">30%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Advanced</td>
</tr>
<tr>
<td align="left" valign="top">18</td>
<td align="left" valign="top">9</td>
<td align="left" valign="top">Claude 3.5 Sonnet</td>
<td align="left" valign="top">99%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">21</td>
<td align="left" valign="top">10</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">15%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">11</td>
<td align="left" valign="top">11</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">50%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">1</td>
<td align="left" valign="top">12</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">70%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">2</td>
<td align="left" valign="top">13</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">90%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">16</td>
<td align="left" valign="top">14</td>
<td align="left" valign="top">Gemini</td>
<td align="left" valign="top">40%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">14</td>
<td align="left" valign="top">15</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">92%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">5</td>
<td align="left" valign="top">19</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">30%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">4</td>
<td align="left" valign="top">20</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">60%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">3</td>
<td align="left" valign="top">23</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">95%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Expert</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">9</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">90%</td>
<td align="left" valign="top">English</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">13</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">45%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">No experience</td>
</tr>
<tr>
<td align="left" valign="top">17</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">GPT-4o + Claude 3.5 Sonnet</td>
<td align="left" valign="top">50%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">7</td>
<td align="left" valign="top">21&#8211;22</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">8%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Advanced</td>
<td align="left" valign="top">Beginner</td>
</tr>
<tr>
<td align="left" valign="top">12</td>
<td align="left" valign="top">21&#8211;22</td>
<td align="left" valign="top">GPT-4o</td>
<td align="left" valign="top">72%</td>
<td align="left" valign="top">Czech</td>
<td align="left" valign="top">Beginner</td>
<td align="left" valign="top">Beginner</td>
</tr>
</tbody>
</table>
<table-wrap-foot>
<fn><p>For each entry the table records the story code used throughout this article, the final ranking awarded by the competition jury, the language model used, the share of AI in the creation of the text as declared by the participant, the language of the prompts, and the participant&#8217;s self-reported experience with AI and with literary writing. Experience was coded on a four-point scale (no experience &#8211; beginner &#8211; advanced &#8211; expert) from open-ended questionnaire answers. Shared rankings (16&#8211;18, 21&#8211;22) denote entries placed ex aequo. Story codes are archival identifiers and do not correspond to the final ranking.</p></fn>
</table-wrap-foot>
</table-wrap>
<p>From the provided materials, it was also possible to summarize basic data reflecting the strategies that individual competition participants chose during prompting. We have summarized them in the following table:</p>
<table-wrap id="T2">
<caption>
<p><bold>Table 2:</bold> Prompting strategies documented in the 23 competition protocols.</p>
</caption>
<table>
<tbody>
<tr>
<td align="left" valign="top" colspan="8"><bold>Prompting strategies</bold></td>
</tr>
<tr>
<td align="left" valign="top"><bold>Short story code</bold></td>
<td align="left" valign="top"><bold>Final ranking of the submission</bold></td>
<td align="left" valign="top"><bold>Human origin of the idea</bold></td>
<td align="left" valign="top"><bold>Prompting approach</bold></td>
<td align="left" valign="top"><bold>Creation of the story title</bold></td>
<td align="left" valign="top"><bold>Imitation of authorial style</bold></td>
<td align="left" valign="top"><bold>Persona prompting</bold></td>
<td align="left" valign="top"><bold>Final human editing</bold></td>
</tr>
<tr>
<td align="left" valign="top">6</td>
<td align="left" valign="top">1</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">20</td>
<td align="left" valign="top">2</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">10</td>
<td align="left" valign="top">3</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">15</td>
<td align="left" valign="top">4</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">8</td>
<td align="left" valign="top">5</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">Remarque</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">22</td>
<td align="left" valign="top">6</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">Hemingway</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">23</td>
<td align="left" valign="top">7</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">19</td>
<td align="left" valign="top">8</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">18</td>
<td align="left" valign="top">9</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">21</td>
<td align="left" valign="top">10</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">11</td>
<td align="left" valign="top">11</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">1</td>
<td align="left" valign="top">12</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">2</td>
<td align="left" valign="top">13</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">Kafka</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">16</td>
<td align="left" valign="top">14</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">14</td>
<td align="left" valign="top">15</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">5</td>
<td align="left" valign="top">19</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">4</td>
<td align="left" valign="top">20</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">3</td>
<td align="left" valign="top">23</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">9</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">13</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Zero-Shot</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
<tr>
<td align="left" valign="top">17</td>
<td align="left" valign="top">16&#8211;18</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">Hard-Boiled School</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">7</td>
<td align="left" valign="top">21&#8211;22</td>
<td align="left" valign="top">Yes</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">Human</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Yes</td>
</tr>
<tr>
<td align="left" valign="top">12</td>
<td align="left" valign="top">21&#8211;22</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">Iterative</td>
<td align="left" valign="top">AI</td>
<td align="left" valign="top">None</td>
<td align="left" valign="top">No</td>
<td align="left" valign="top">No</td>
</tr>
</tbody>
</table>
<table-wrap-foot>
<fn><p>For each entry, identified by its story code and final ranking, the table records whether the idea for the story originated with the participant or with the model; whether prompting proceeded iteratively or through a single zero-shot prompt; whether the story title was created by the human or by the model; whether the participant asked the model to imitate the style of a particular author or school; whether a persona was assigned to the model; and whether the generated text underwent final human editing. The categories were coded by the authors from the complete conversation transcripts contained in the protocols.</p></fn>
</table-wrap-foot>
</table-wrap>
<p>Although the provided data are not well suited for detailed quantitative data analysis, it is nonetheless possible to abstract certain tendencies and correlations from them:</p>
<list list-type="bullet">
<list-item><p>The data indicate a strong correlation between the level of experience in literary creation and the declared share of AI in the resulting text &#8212; the less literary experience, the greater the declared AI involvement (authors with no literary experience declare an average AI share of approximately 72%, beginners approximately 48%, advanced authors approximately 22%, and experts approximately 33%).</p></list-item>
<list-item><p>Also noteworthy is the correlation between the degree of literary experience and the chosen prompting strategy. The data point to a tendency toward iterative prompting among all competition participants who have at least beginner-level literary experience, while competitors with no literary experience more frequently opted for minimalist prompting. The competition was, however, won by a participant with no literary experience but with exceptionally developed expertise in AI and prompting, which set her entirely apart from the other competitors.</p></list-item>
<list-item><p>Despite the fact that the winner was an author with a maximum disproportion between literary and AI competencies (who can be regarded as an outlier in this dataset), authors declaring at least basic literary experience placed better on average in the competition; AI expertise was generally not sufficiently developed among participants to lead to better placement on its own.</p></list-item>
</list>
<p>These findings correspond with the results of the qualitatively oriented study &#8216;Holding the Line: A Study of Writers&#8217; Attitudes on Co-creativity with AI&#8217;, whose authors distinguish three modes of writing with AI assistance &#8212; hedonic, scribe, and artiste. They associate budding authors (vernacular authors in our terminology) with the hedonic mode and note that this group of authors is very open to AI assistance at all stages of creation and does not associate it with feelings of threat: &#8216;In hedonic mode, writers are more free and casual in their writing, and prioritize entertainment and potentially learning. We found younger, budding writers, or those who said they dabbled in writing as a hobby, to be more likely to report being in this &#8220;hedonic&#8217; mode&#8221; (<xref ref-type="bibr" rid="B4">Behrooz et al., 2024: 2</xref>). The contrasting mode is the artiste, found among experienced writers and professional screenwriters, who are conversely very reserved about using AI, as they prefer control over the creative process and its maximum authenticity. The scribe mode is then typical of authors producing texts on commission, for whom AI assistance is perceived pragmatically and accepted primarily when it speeds up text production. The correlation between length of literary practice and openness toward AI&#8217;s possibilities is also mentioned in the study &#8216;Fanfiction in the Age of AI: Community Perspectives on Creativity, Authenticity and Adoption&#8217; (<xref ref-type="bibr" rid="B1">Alfassi et al., 2026</xref>), which states that members of fanfiction communities who have been writing for only one to five years are the most open to AI, while the group of authors with more than 10 years of literary work is the most skeptical toward AI.</p>
</sec>
<sec>
<title>5.3 Qualitative Analysis of Protocols and its Findings</title>
<p>As we have already noted, quantitative data analysis could only be applied to the available material to a very limited degree. Our main observations of the materials associated with this literary competition were therefore based on a qualitative analysis of the protocols produced by discussion participants, which generally contain a complete copy of the chatbot dialogue during story generation. The subsequent paragraphs synthesize the principal findings emerging from these observations:</p>
<p>If the submitted stories conform to the conventions of detective fiction, it is due to the AI&#8217;s output rather than the prompts provided by participants. The prompts don&#8217;t suggest literary knowledge, but instead the human author&#8217;s knowledge of the mechanics of prompt construction. In general, literary knowledge is demanded from the AI, rather than transmitted to it by the user during the creative process. The collaboration between human and AI often takes a disproportionate form: the human occupies a superior role, investing only minimal literary effort or knowledge, while the AI is expected to function as a servant rather than an assistant &#8212; repeatedly attempting to fulfill the user&#8217;s vague or underspecified instructions, and often being left to infer key contextual information on its own.</p>
<p><bold>Example 1.</bold> The delegation of literary competence can be clearly demonstrated in the opening prompt of one of the participants (documentation protocol 13), who gives the model an extensive set of content instructions:</p>
<disp-quote>
<p>Write me a draft of a story, where the story will be in the style of a Thriller, use dialogues in the text, start telling the story with a dialogue, don&#8217;t forget that it should be suspenseful for the reader and that until the last moment they shouldn&#8217;t suspect who the perpetrator is. [&#8230;] The main character is a female employee who does all the evil but looks and acts like she&#8217;s very nice, but is still somehow not quite fitting in, tries to be more Catholic than the Pope, according to the personality-profile type per the DISC typology she is anxious, order is important to her, everything must be correct according to the norms. [&#8230;] It&#8217;s important to come up with an interesting twist&#8217; (protocol 13; all excerpts from the protocols are quoted verbatim, and translated from Czech into English).</p>
</disp-quote>
<p>The participant contributes the raw material to the interaction &#8212; a psychological profile of the character (with reference to the personnel-typology framework DISC), the setting, and the theme &#8212; yet requests all literary competence, in both Culler&#8217;s and Svedjedal&#8217;s sense, from the model: suspense, withholding information about the perpetrator, and the &#8216;interesting twist&#8217; are all left for the AI to invent. The instructions concern exclusively what the text should contain, not how it should be executed in literary terms; the only formal instructions (dialogues, length) are technical in nature. The linguistic form of the prompt &#8212; anacolutha, typos, punctuation errors &#8212; likewise shows that participants are not concerned with the linguistic form of their own inputs; as another participant put it with disarming honesty: &#8216;Sorry for the typos. I don&#8217;t bother with that with the robot. :-)&#8217; (protocol 16). Care for linguistic and stylistic quality thus remains, throughout the entire corpus of protocols, exclusively on the side of the model.</p>
<p>The documentation repeatedly reveals attempts to extract from the model a set of criteria for a successful short story &#8212; a demand that would be almost unthinkable if directed at a human being, given that literary success is notoriously unpredictable and governed by no strict rules. One may assume that this expectation stems from the belief that the AI can analyze or synthesize patterns from successful past stories and apply them in formulating rules and generating new texts. The central finding, supported by all 23 prompt protocols, is this: in the context of vernacular creativity supported by AI, the human actor tends to delegate literary competence to the language model and assumes the role of an operator whose domain is prompting, without actually having the criteria of a successful short story internalized.</p>
<p><bold>Example 2.</bold> An extreme form of the delegation of literary competence is represented by cases in which the model is entrusted even with the critical evaluation of the resulting text, a competence which in literary communication traditionally belongs to the reader, the critic, or the editor. One participant (protocol 18) gave the Claude model the prompt: &#8216;Perform a critical analysis of the following story and its possible interpretations, at the level of a university professor of literature&#8217;. The model carried out a genre-based, narratological, and stylistic analysis, and evaluated the text it had itself generated in these words:</p>
<disp-quote>
<p>While the story effectively uses the conventions of the genre to create a suspenseful narrative, one could argue that it does not bring much that is new to the detective-fiction genre. [&#8230;] In conclusion, &#8220;Seven Sins&#8221; is a competently written detective story that effectively creates tension and surprise. [&#8230;] At the same time, however, it does not go beyond the boundaries of genre conventions in a way that would make it exceptional within contemporary detective literature (protocol 18).</p>
</disp-quote>
<p>This self-assessment &#8212; competent but conventional &#8212; in fact fully corresponds to reality. A remarkable situation thus arises in which the model possesses a more accurate judgment of the literary value of the generated text than its human co-author, who has not internalized the criteria of a successful short story, and precisely for that reason requested them (and their application) from the model. The circle of delegation thereby closes: the AI generates the text according to synthetically established rules and subsequently also evaluates it according to those same rules.</p>
<p>In many cases, not only literary competence but also the creative intent itself is delegated. A common practice observed was users asking the AI to suggest a topic or theme for the story &#8212; approaching the process of prompting without any prior intention or vision of what the narrative should convey. In such instances, it can no longer be said that AI functions as an assistant to human creativity; rather, it is the human who assists the AI in semi-autonomous generation. One frequent prompting technique involves asking the model to list possible options (themes, characters, settings, plot twists, titles, etc.), from which the user then selects. In this scenario, the AI clearly intervenes in the realm of creative intent, as it frames the user&#8217;s choice by offering a predefined and limited selection &#8212; thus shaping the scope of the user&#8217;s creative decisions. This constitutes a form of supervised &#8216;framed authorship&#8217; (<xref ref-type="bibr" rid="B17">Pioreck&#253; and Hus&#225;rov&#225;, 2024</xref>), in which human intentionality enters not only at the beginning or end of the process but also actively influences the generation as it unfolds. In this role, the human becomes a particular type of editor, intervening in the development of the text even as it is being created.</p>
<p>This principle of choosing from pre-structured options lends a game-like quality to the human-AI dialogue. The documentation we analyzed often resembled interactive narratives in structure. This was most evident in the cases where participants used the Claude model, which, beyond generating new passages of the story, also proposed multiple directions for further narrative development. In these interactions, the engagement closely resembled that of playing a text-based computer game or reading a <italic>choose-your-own-adventure</italic> gamebook. Many participants noted in their comments that working with AI was an enjoyable experience. These observations allow us to conceptualize the process of AI-assisted generation as a literary experience in its own right, an interactive literary practice in which the value lies not necessarily in the final product, but in the generative process itself. In this process, the human participant often functions more as a reader than as an author, experiencing surprise and deriving aesthetic enjoyment from the interaction. This confirms Jean Burgess&#8217;s (<xref ref-type="bibr" rid="B6">2007: 33</xref>) insight that vernacular creativity tends to be closely linked to popular culture, including playful, interactive literary forms and to some extent an existing genre of electronic literature.</p>
<p><bold>Example 3.</bold> The game-like character of the interaction is illustrated by an excerpt from protocol 23, in which a participant with expert-level competence in AI, but without any substantial experience writing fiction, discusses a story concept with the Claude model:</p>
<disp-quote>
<p>Claude: Your idea is truly innovative and makes very good use of the possibilities offered by collaboration with AI. [&#8230;] I have a few questions and suggestions regarding this concept: [&#8230;] What if the real &#8220;perpetrator&#8221; were an artificial intelligence? For example, an autonomous security system that got out of control. [&#8230;] What if we also included in the story an element where the characters start to wonder whether they themselves are part of an artificial construct?</p>
<p>Participant: having AI as the perpetrator is a fascin-AI-ting idea :) but thinking about it more, the real magic would actually lie in it not being entirely clear what happened &#8212; maybe different readers would even arrive at different conclusions [&#8230;]. That sounds really good, but I don&#8216;t know if it isn&#8216;t too complex for something that&#8216;s supposed to be a short story. We could try it in a possible sequel (protocol 23).</p>
</disp-quote>
<p>Here the model does not act as a passive executor of instructions: it offers suggestions in a structured way, framing the space of the participant&#8217;s decision-making, and the participant selects from this offer, rejects, and modifies it &#8212; proceeding, that is, exactly according to the logic of supervised framework authorship described above. The interaction clearly takes the form of a game: each of the model&#8217;s turns opens up several possible continuations, from which the human actor chooses, similar to a reader of a gamebook. In doing so, the AI demonstrably enters into the sphere of creative intent &#8212; the suggestion that the perpetrator should be the artificial intelligence itself, as well as the metanarrative element, originate from the model, not from the human.</p>
<p>One further point must be made about a specific recursive tendency inherent in AI-assisted vernacular creativity, one that manifests in two distinct senses. First, as already noted, when literary competence is delegated to AI, it often happens via prompts asking the model to define genre rules or stylistic guidelines. As a result, the synthetic text emerges based on synthetically generated rules. Synthetic authorship (even in this assisted mode) becomes a closed loop, self-sufficient, and largely detached from external inputs such as literary knowledge, creative intent, or lived experience (a phenomenon we have already discussed in our book <italic>The Culture of Neural Networks</italic>).</p>
<p>The second kind of recursion is specific to vernacular creativity. It stems from the fact that the dialogue between the AI and the user unfolds in a highly affirmative tone, because current large language models are deliberately designed to avoid expressing negativity toward the human interlocutor. In practice, this means that AI tends to enthusiastically affirm any human idea, regardless of its literary merit. Users thus receive no critical feedback. Moreover, the communication often flows equally uncritically in the opposite direction: many users in our dataset readily accepted subpar AI outputs, partly encouraged by the AI&#8217;s own positive self-assessment, which it frequently appended to its generated text segments. This creates a self-reinforcing conversational cycle based on mutual illusion and a somewhat false consensus, particularly given that the AI&#8217;s positivity is algorithmically predetermined. While this affirming feedback can be motivating for beginners and amateur creators, it may also distort their perception of their own writing ability and inhibit critical self-reflection. We see this as a clear risk in the use of AI for cultivating vernacular creativity.</p>
<p><bold>Example 4.</bold> The affirmative nature of the human&#8211;AI dialogue and its consequences are documented by an excerpt from protocol 22. Its author is an exception within the sample: he entered the interaction with his own well-defined aesthetic program (he wanted to style the story in the poetics of art brut) and repeatedly critically corrected the model&#8217;s outputs:</p>
<disp-quote>
<p>Participant: &#8216;No. Try writing the beginning of an art-brut noir detective story in this style: I am a dog who likes the forest, I run back and forth, I jump over the fence [&#8230;]&#8217;</p>
<p>ChatGPT: &#8216;Understood, here&#8217;s the beginning of a detective story in the art-brut style, as you requested: [&#8230;]&#8217;</p>
<p>Participant: &#8216;That&#8217;s much better! Try continuing in this style.&#8217;</p>
<p>ChatGPT: &#8216;Thanks for the encouragement! Here&#8217;s the continuation: [&#8230;]&#8217;</p>
<p>Participant: &#8216;You keep sliding back from art brut into normal art every time&#8230; Try to stick to the style of &#8220;I am a dog who likes the forest, but today there was no time for the forest.&#8221;&#8217;</p>
<p>ChatGPT: &#8216;Understood, you&#8217;re right, I&#8217;ll try to stay in this style&#8217; (protocol 22).</p>
</disp-quote>
<p>The excerpt captures both sides of the affirmative loop: the model thanks the participant for the &#8216;encouragement&#8217; and unreservedly agrees with every criticism (&#8216;you&#8217;re right&#8217;), without ever pushing back against its human partner. At the same time, it reveals the limits of delegating style &#8212; despite repeated instructions, the model persistently slides back toward a conventional, smooth register. It is precisely in confrontation with a participant who, as one of the few in the sample, possessed his own distinct aesthetic intent, that the otherwise hidden normative inertia of the model becomes visible; most participants, who had no distinct stylistic intent of their own, accepted this inertia without noticing it, and without resistance.</p>
<p>One of the questions posed to participants in the competition concerned their estimation of the percentage of authorship they attributed to themselves. Responses varied dramatically, from as little as 5% to as much as 99%. A closer analysis revealed that these figures were highly subjective and offered little concrete insight. Some participants regarded editorial work on the generated text as constituting authorship, while others included the act of prompting itself. Most of all, the declared authorship share depended on the participants&#8217; intentions in entering the competition: those seeking recognition for their literary skill tended to report high levels of human contribution, while those aiming to showcase their prompting expertise downplayed their own input. Perceptions of authorship were also shaped by how participants understood the symbolic value of being recognized as an author &#8212; a point illustrated by one contestant&#8217;s remark: &#8216;I feel like the author by about 92%. Just carrying such an important designation is a great honor to me&#8217; (protocol 7; translation our own).</p>
<p>Drawing on the documentation of the <italic>Z-AI-mav&#253; zlo&#269;in</italic> [<italic>An AI-musing Crime</italic>] competition, we identify several central findings that bear theoretical significance. First, the influence of AI on the writing process is not confined to formal or stylistic dimensions, but extends to creative intent itself &#8212; the generative encounter with the model shapes what authors set out to express, not only how they express it. Second, the generation process assumes a distinctly ludic character, positioning AI-assisted writing as a form of interactive literature in which meaning emerges through iterative exchange rather than singular authorial decision. Third, this iterative dynamic tends toward what we term <italic>synthetic recursion</italic>: the output of each generative cycle feeds back into the creative process, producing a layered, self-referential textual practice. Fourth, the conversational interface through which AI is engaged is structurally affirmative, meaning that the model does not contest, resist, or evaluate, which, while productive for generative momentum, risks foreclosing the kind of critical self-reflection that has traditionally been central to literary craft. Finally, and perhaps most consequentially for our broader theoretical argument, AI-assisted writing offers vernacular authors an accessible and attractive pathway to authorial status &#8212; one that lowers the threshold of literary competence while simultaneously reconfiguring what that competence is understood to entail. These findings, on the one hand, situate the analyzed competition within the vernacular phase of the establishment of neural networks in culture, and, on the other hand, precisely through the documented redistribution of competences, point to the emergence of the next phase, which we address in the following section.</p>
</sec>
</sec>
<sec>
<title>6. A Phase of Reskilling in the Development of the Culture of Neural Networks</title>
<p>The assertion that vernacular authors achieve better literary results through AI assistance is supported by the empirical study &#8216;Generative AI enhances individual creativity but reduces the collective diversity of novel content&#8217;, which investigated vernacular creativity in the context of short story writing. At the outset of this study, participating authors were ranked according to their inherent creative abilities. The findings showed that AI significantly improves outcomes in terms of creativity and emotional impact for authors with low natural creative capacity. When using AI tools, the quality ratings of their literary writing improved by several 10s of percent &#8212; bringing them to a level comparable with highly creative authors. The latter, however, did not experience significant quality gains when using the same tools. At the same time, it became evident that texts created in collaboration with AI tended to resemble one another (as though authors using AI assistance were anchored in ideas offered by the AI) (cf. <xref ref-type="bibr" rid="B8">Doshi and Hauser, 2024</xref>). This threat to literary diversity was also observable in our own dataset from the <italic>Z-AI-mav&#253; zlo&#269;in</italic> [<italic>An AI-musing Crime</italic>] writing competition, where AI repeatedly suggested the theme of investigating the theft of a famous painting to multiple participants. Ironically, one contestant independently chose the same topic &#8212; without AI assistance &#8212; and in an even more stereotypical form, namely, the theft of the <italic>Mona Lisa</italic>.</p>
<p>What we are currently witnessing in culture could be described as a process of <italic>reskilling</italic>, whereby the traditional set of competencies deemed necessary for literary production is being restructured. Some elements of this set are diminishing in importance, while others, new and previously unrelated, are being incorporated. Various stimuli, including writing competitions involving AI and the growing number of courses on creative AI usage, indicate that we are now at the beginning of a new phase in the development of the culture of neural networks &#8212; <italic>a phase of reskilling</italic>. This phase is characterized above all by a transformation in the approach to fundamental cultural techniques, including those most relevant to literary scholarship: writing and reading. Together with philosopher Sybille Kr&#228;mer, we might ask whether artificial intelligence itself is beginning to assume the role and significance of a cultural technique (cf. <xref ref-type="bibr" rid="B11">Kr&#228;mer, 2021</xref>) &#8212; though it may still be too early for a conclusive answer. What is already evident, however, is the transformation of traditional cultural techniques (especially writing) as a result of new methods for producing cultural artifacts, a shift clearly observable in the text-making practices of vernacular authors. In the logic of our periodization, the reskilling phase thus follows on from the vernacular phase: it does not negate it, but rather restructures the competences that vernacular practice brought into visibility; and, like the preceding phase, it designates a dominant tendency rather than a sharp historical break. We do not wish to subject this transformation to normative evaluation, tempting though such judgment may be. In this spirit, we also understand Lev Manovich&#8217;s techno-optimistic statement in his recent essay &#8216;Seven Arguments about AI Images and Generative Media&#8217;:</p>
<disp-quote>
<p>&#8216;What was once a skilled artistic act is now a technological capability available to everyone. We can be sad about everything that might be lost as a result of the automation &#8212; and democratization &#8212; of this critical cultural operation: skills, something one might call &#8216;deep artistic originality&#8217; or &#8216;deep creativity&#8217;, and so on. However, any such loss may be only temporary if the abilities of &#8216;culture AI&#8217; are, for example, even further improved to generate more original content and understand context better (<xref ref-type="bibr" rid="B14">Manovich, 2025: 136</xref>).</p>
</disp-quote>
<p>We would argue, however, that technological advancement alone may not suffice to bring about the cultural shift Manovich anticipates. Recent empirical research continues to indicate that consumers of cultural artifacts tend to prefer human-made works over synthetic ones &#8212; not necessarily because AI is not yet good enough, but perhaps due to a kind of human-to-human loyalty or the value still ascribed to creative effort. For instance, the authors of the study &#8216;Humans as Creativity Gatekeepers: Are We Biased Against AI Creativity?&#8217; found that people attribute lower levels of creativity to a given work when they are informed that its creator is artificial intelligence rather than a human. The primary reason for this lower evaluation, according to respondents, lies in the perception that AI invests less effort in the creative process than a human would. This attitude was particularly evident in relation to visual art, but not in other domains such as commercial graphic design (see <xref ref-type="bibr" rid="B13">Magni, Park and Chao, 2024</xref>). A similar research in poetry was conducted by Brian Porter and Edouard Machery, whose findings were published in the study &#8216;AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably&#8217;. These researchers noted that, at the outset, most participants in their study explicitly stated a preference for poetry written by humans and anticipated that they would dislike AI-generated poems. However, the analysis of responses to anonymized poems ultimately revealed the opposite: the AI-generated poems were rated more favorably than those written by human authors. The main reason cited for this preference was the greater comprehensibility of the verses produced by AI (see <xref ref-type="bibr" rid="B18">Porter and Machery, 2024</xref>). That said, these preferences may change and diversify over time. Generative AI will likely find its own creative domain and methods of application across various cultural and artistic spheres and at all conceivable levels of artistic quality and ambition. For now, however, it remains too early to define the boundaries of these domains.</p>
</sec>
<sec>
<title>7. Conclusion</title>
<p>What, then, are the key differences between the literary behavior of vernacular authors in the era of Web 2.0 and in the age of so-called generative AI? The fundamental distinction, in our view, lies in how these authors relate to literary competence &#8212; a concept repeatedly referenced throughout this study. <italic>The vernacular author of the Web 2.0 era</italic> applies their literary competence at the level they are currently capable of, engaging in online communities composed of similarly literate participants. Within such communities, authors can to some extent develop their skills, receive feedback, and, most importantly, satisfy their desire for conversation about literature and shared interests or hobbies.</p>
<p>In contrast, the amateur <italic>vernacular author operating within the culture of neural networks</italic> (excluding those well-versed in the principles of digital literature and culture, or those engaged in assisted creativity) tends to delegate literary competence to a large language model (LLM). These authors develop their prompting skills, thereby usually achieving more effective results than if they were to rely solely on their own literary creativity. They may, optionally, enter the book market and, thanks to self-publishing services, find readership, particularly among audiences interested in genre-based popular fiction. Their mode of creation, however, seems to be arbitrary with respect to collectivity or communities of similarly skilled authors. If we return to the model of the three zones of the literary system sketched above: whereas the vernacular authors of the Web 2.0 era inhabited the periphery of the system and could &#8212; through the acquisition of conventions within communities &#8212; move toward the center, the vernacular authors of the AI era rather bypass this axis: literary conventions are handled on their behalf by the model, and their own development takes place in a different domain, in the competence of prompting.</p>
<p>This article has argued that the development of neural network culture has entered a distinct new <italic>phase of reskilling</italic>, in which the conventional ensemble of competencies historically deemed necessary for literary production is undergoing substantive restructuring. This process is not one of simple substitution: certain competencies are receding in centrality, while others, previously peripheral or entirely extraneous to literary practice, are being incorporated into what it means to author a text. Most significantly, this phase is defined by a transformation in the cultural techniques that literary scholarship has long held foundational: writing and reading themselves are being reconstituted under the conditions of AI-assisted textual production. We also view reskilling as one of the aspects of the process of establishing the post-artificial text, as written about by Hannes Bajohr: &#8216;Until now, however, we have always assumed that a human being is behind it, that there is a writer &#8212; but under post-artificial reading conditions, nothing much changes if we simply make no assumptions at all&#8217; (<xref ref-type="bibr" rid="B3">Bajohr, 2024: 351</xref>).</p>
<sec>
<title>7.1 Limitations and Future Research Directions</title>
<p>The entire situation of vernacular creativity in the age of AI is, in fact, paradoxical. Jean Burgess aptly points out that &#8216;particular forms of vernacular creativity are grounded in contextual specificity&#8217; (<xref ref-type="bibr" rid="B6">Burgess, 2007: 32</xref>). What, then, is the contextual specificity of vernacular creativity in the current AI epoch? How can we discuss vernacularization, amateurism, or professionalism when AI evolves so rapidly that true professionals among its users are only just emerging? The generation of literary texts with AI assistance exhibits profound contextual specificity, since it occurs at the intersection of literature, generative art, and artificial intelligence. Much of the creative output emerging at this crossroads could legitimately be classified within the zone of vernacular creativity.</p>
<p>One of the key questions in this ambiguous and highly dynamic environment is how cultural value is, or will be, constructed in the context of creation using generative AI. Can we regard as valuable a work that emerges from the fortuitous formulation of a prompt? Can the cultural object alone, as an outcome, truly determine value? Or does the value of paratextual and metatextual information accompanying such artifacts increase in this context? In such circumstances, it becomes extremely difficult to evaluate the phenomenon of vernacular creativity in literature and other arts. This does not mean, however, that it is impossible to identify clear artistic shortcomings. If a human agent is unable to assess or recognize whether their creative process, regardless of the degree or form of AI involvement, results in artistically meaningful outcomes (particularly when published alongside similar artifacts), then this constitutes an artistic deficiency of the same nature as in the pre-AI era.</p>
<p>We acknowledge, however, that the conclusions advanced here are necessarily provisional. The dataset on which this study draws remains limited in scope, and the generalizability of our theoretical propositions is accordingly constrained. Future research would benefit from a substantially expanded corpus of authors, as well as from a more precise operationalization of literary competence as a distinct category of working with and producing literary texts &#8212; while distinguishing prompting ability from adjacent competencies such as editing, curation, genre familiarity, and broader digital literacy. The framework developed here would further profit from a multimodal extension, moving beyond the purely textual samples on which the present analysis relies and engaging with the full range of expressive forms that AI-assisted vernacular creativity is trying to encompass.</p>
</sec>
</sec>
</body>
<back>
<sec>
<title>Ethics and Consent</title>
<p>The material analyzed in this study was submitted by participants in the literary competition <italic>Z-AI-mav&#253; zlo&#269;in</italic> [<italic>An AI-musing Crime</italic>]. Consent to the scholarly use of the submitted documentation forms part of the terms and conditions of the competition, which state:</p>
<disp-quote>
<p>Zasl&#225;n&#237;m sout&#283;&#382;n&#237;ho p&#345;&#237;sp&#283;vku &#250;&#269;astn&#237;ci ud&#283;luj&#237; souhlas M&#283;stsk&#233; knihovn&#283; v Praze k p&#345;&#237;padn&#233; tiskov&#233; reprodukci ve sborn&#237;ku, redak&#269;n&#237;m &#250;prav&#225;m a zve&#345;ejn&#283;n&#237; d&#237;la na webov&#253;ch str&#225;nk&#225;ch M&#283;stsk&#233; knihovny v Praze a tak&#233; k tomu, &#382;e technick&#233; informace m&#367;&#382;eme podrobit dal&#353;&#237;mu v&#283;deck&#233;mu zkoum&#225;n&#237; a v&#253;sledky p&#345;&#237;padn&#283; zve&#345;ejnit.</p>
<p>By submitting a competition entry, participants grant the Municipal Library of Prague consent to the possible reproduction of the work in print in an anthology, to editorial revision, and to the publication of the work on the website of the Municipal Library of Prague, as well as consent for the technical information to be subjected to further scholarly investigation and for the results to be published (translation our own).</p>
</disp-quote>
<p>(<ext-link ext-link-type="uri" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="https://www.mlp.cz/cz/novinky/z-ai-mavy-spisovatel-treti-rocnik">https://www.mlp.cz/cz/novinky/z-ai-mavy-spisovatel-treti-rocnik</ext-link>).</p>
<p>All protocols are cited in anonymized form: participants&#8217; names and any identifying details have been removed, and individual entries are referred to solely by the numerical story codes used in <xref ref-type="table" rid="T1">Tables 1</xref> and <xref ref-type="table" rid="T2">2</xref>.</p>
</sec>
<sec>
<title>Acknowledgements</title>
<p>The text was written as a part of the grant Generating Czech Poetry in an Educational and Multimedia Environment supported by the Technology Agency of the Czech Republic (SIGMA program, project code: TQ01000153).</p>
</sec>
<sec>
<title>Competing Interests</title>
<p>The authors have no competing interests to declare.</p>
</sec>
<ref-list>
<ref id="B1"><mixed-citation publication-type="journal"><string-name><surname>Alfassi</surname>, <given-names>R</given-names></string-name>, <string-name><surname>Cooper</surname>, <given-names>A</given-names></string-name>, <string-name><surname>Mitchell</surname>, <given-names>Z</given-names></string-name>, <string-name><surname>Calabro</surname>, <given-names>M</given-names></string-name>, <string-name><surname>Shaer</surname>, <given-names>O</given-names></string-name> and <string-name><surname>Mokryn</surname>, <given-names>O</given-names></string-name> <year>2026</year> <article-title>Fanfiction in the Age of AI: Community Perspectives on Creativity, Authenticity and Adoption</article-title>. <source>International Journal of Human&#8211;Computer Interaction</source>, <volume>42</volume>(<issue>5</issue>): <fpage>3062</fpage>&#8211;<lpage>3094</lpage>. <pub-id pub-id-type="doi">10.1080/10447318.2025.2531272</pub-id></mixed-citation></ref>
<ref id="B2"><mixed-citation publication-type="webpage"><string-name><surname>Arielli</surname>, <given-names>E</given-names></string-name> and <string-name><surname>Manovich</surname>, <given-names>L</given-names></string-name> <year>2022</year> <source>Artificial Aesthetics: A Critical Guide to AI, Media and Design</source>. Available at: <uri>http://manovich.net/index.php/projects/artificial-aesthetics</uri> [Last Accessed 26 April 2025].</mixed-citation></ref>
<ref id="B3"><mixed-citation publication-type="journal"><string-name><surname>Bajohr</surname>, <given-names>H</given-names></string-name> <year>2024</year> <article-title>On Artificial and Post-artificial Texts: Machine Learning and the Reader&#8217;s Expectations of Literary and Non-literary Writing</article-title>. <source>Poetics Today</source>, <volume>45</volume>(<issue>2</issue>): <fpage>331</fpage>&#8211;<lpage>361</lpage>. <pub-id pub-id-type="doi">10.1215/03335372-11092990</pub-id></mixed-citation></ref>
<ref id="B4"><mixed-citation publication-type="journal"><string-name><surname>Behrooz</surname>, <given-names>M</given-names></string-name>, <string-name><surname>Tian</surname>, <given-names>Y</given-names></string-name>, <string-name><surname>Ngan</surname>, <given-names>W</given-names></string-name>, <string-name><surname>Yungster</surname>, <given-names>Y</given-names></string-name>, <string-name><surname>Wong</surname>, <given-names>J</given-names></string-name> and <string-name><surname>Zax</surname>, <given-names>D</given-names></string-name> <year>2024</year> <article-title>Holding the Line: A Study of Writers&#8217; Attitudes on Co-creativity with AI</article-title>. <source>arXiv</source> preprint arXiv:2404.13165. <pub-id pub-id-type="doi">10.48550/arXiv.2404.13165</pub-id></mixed-citation></ref>
<ref id="B5"><mixed-citation publication-type="book"><string-name><surname>Benjamin</surname>, <given-names>W</given-names></string-name> <year>1969</year> <chapter-title>The Work of Art in the Age of Mechanical Reproduction</chapter-title>. In: <string-name><surname>Arendt</surname>, <given-names>H</given-names></string-name> (ed.) <source>Illuminations</source>. <string-name><surname>Zohn</surname>, <given-names>H</given-names></string-name> (trans.). <publisher-loc>New York</publisher-loc>: <publisher-name>Schocken Books</publisher-name>, pp. <fpage>217</fpage>&#8211;<lpage>251</lpage>.</mixed-citation></ref>
<ref id="B6"><mixed-citation publication-type="thesis"><string-name><surname>Burgess</surname>, <given-names>J</given-names></string-name> <year>2007</year> <chapter-title>Vernacular Creativity and New Media</chapter-title>. PhD, <publisher-name>Queensland University of Technology</publisher-name>. <uri>https://eprints.qut.edu.au/16378/</uri> [Last Accessed 30 April 2026].</mixed-citation></ref>
<ref id="B7"><mixed-citation publication-type="book"><string-name><surname>Culler</surname>, <given-names>J</given-names></string-name> <year>1975</year> <source>Structuralist Poetics</source>. <publisher-loc>London</publisher-loc>: <publisher-name>Routledge</publisher-name>.</mixed-citation></ref>
<ref id="B8"><mixed-citation publication-type="webpage"><string-name><surname>Doshi</surname>, <given-names>A R</given-names></string-name> and <string-name><surname>Hauser</surname>, <given-names>O P</given-names></string-name> <year>2024</year> <article-title>Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content</article-title>. <source>Science Advances</source>, <volume>10</volume>(<issue>28</issue>). <uri>https://www.science.org/doi/10.1126/sciadv.adn5290</uri></mixed-citation></ref>
<ref id="B9"><mixed-citation publication-type="book"><string-name><surname>Jenkins</surname>, <given-names>H</given-names></string-name> <year>2006</year> <source>Convergence Culture: Where Old and New Media Collide</source>. <publisher-loc>New York</publisher-loc>: <publisher-name>New York University Press</publisher-name>.</mixed-citation></ref>
<ref id="B10"><mixed-citation publication-type="book"><string-name><surname>Jenkins</surname>, <given-names>H</given-names></string-name>, <string-name><surname>Ford</surname>, <given-names>S</given-names></string-name> and <string-name><surname>Green</surname>, <given-names>J</given-names></string-name> <year>2013</year> <source>Spreadable Media: Creating Value and Meaning in a Networked Culture</source>. <publisher-loc>New York</publisher-loc>: <publisher-name>New York University Press</publisher-name>.</mixed-citation></ref>
<ref id="B11"><mixed-citation publication-type="webpage"><string-name><surname>Kr&#228;mer</surname>, <given-names>S</given-names></string-name> <year>2021</year> <source>Digitalism as a Cultural Technique: From Alphanumerics to AI</source>. Available at: <uri>https://www.goethe.de/prj/k40/en/eth/dig.html</uri> [Last Accessed 24 April 2025].</mixed-citation></ref>
<ref id="B12"><mixed-citation publication-type="book"><string-name><surname>Lessig</surname>, <given-names>L</given-names></string-name> <year>2008</year> <source>Remix: Making Art and Commerce Thrive in the Hybrid Economy</source>. <publisher-loc>New York</publisher-loc>: <publisher-name>Penguin Press</publisher-name>.</mixed-citation></ref>
<ref id="B13"><mixed-citation publication-type="journal"><string-name><surname>Magni</surname>, <given-names>F</given-names></string-name>, <string-name><surname>Park</surname>, <given-names>J</given-names></string-name> and <string-name><surname>Chao</surname>, <given-names>M M</given-names></string-name> <year>2024</year> <article-title>Humans as Creativity Gatekeepers: Are We Biased Against AI Creativity?</article-title> <source>Journal of Business and Psychology</source>, <volume>39</volume>: <fpage>643</fpage>&#8211;<lpage>656</lpage>. <pub-id pub-id-type="doi">10.1007/s10869-023-09910-x</pub-id></mixed-citation></ref>
<ref id="B14"><mixed-citation publication-type="book"><string-name><surname>Manovich</surname>, <given-names>L</given-names></string-name> <year>2025</year> <chapter-title>Seven Arguments about AI Images and Generative Media</chapter-title>. In: <string-name><surname>Bajohr</surname>, <given-names>H</given-names></string-name> (ed.) <source>Thinking with AI: Machine Learning the Humanities</source>. <publisher-loc>London</publisher-loc>: <publisher-name>Open Humanities Press</publisher-name>. <fpage>126</fpage>&#8211;<lpage>146</lpage>.</mixed-citation></ref>
<ref id="B15"><mixed-citation publication-type="webpage"><string-name><surname>O&#8217;Reilly</surname>, <given-names>T</given-names></string-name> <year>2007</year> <article-title>What Is Web 2.0: Design Patterns and Business Models for the Next Generation of Software</article-title>. <source>Communications &amp; Strategies</source>, <volume>65</volume>(<issue>1</issue>): <fpage>17</fpage>&#8211;<lpage>37</lpage>. <uri>https://ssrn.com/abstract=1008839</uri></mixed-citation></ref>
<ref id="B16"><mixed-citation publication-type="book"><string-name><surname>Pioreck&#253;</surname>, <given-names>K</given-names></string-name> <year>2017</year> <chapter-title>Literary Work as a Leisure Activity: Amateur Literary Forums on the Czech Internet</chapter-title>. In: <string-name><surname>Carnicelli</surname>, <given-names>S</given-names></string-name>, <string-name><surname>McGillivray</surname>, <given-names>D</given-names></string-name> and <string-name><surname>McPherson</surname>, <given-names>G</given-names></string-name> (eds.) <source>Digital Leisure Cultures: Critical Perspectives</source>. <publisher-loc>New York</publisher-loc>: <publisher-name>Routledge</publisher-name>, pp. <fpage>193</fpage>&#8211;<lpage>206</lpage>.</mixed-citation></ref>
<ref id="B17"><mixed-citation publication-type="book"><string-name><surname>Pioreck&#253;</surname>, <given-names>K</given-names></string-name> and <string-name><surname>Hus&#225;rov&#225;</surname>, <given-names>Z</given-names></string-name> <year>2024</year> <source>The Culture of Neural Networks: Synthetic Literature and Art in (Not Only) the Czech and Slovak Context</source>. <string-name><surname>Stainsby</surname>, <given-names>I</given-names></string-name> (trans.). <publisher-loc>Praha</publisher-loc>: <publisher-name>Institute of Czech Literature of the CAS and Karolinum Press</publisher-name>.</mixed-citation></ref>
<ref id="B18"><mixed-citation publication-type="journal"><string-name><surname>Porter</surname>, <given-names>B</given-names></string-name> and <string-name><surname>Machery</surname>, <given-names>E</given-names></string-name> <year>2024</year> <article-title>AI-generated Poetry Is Indistinguishable from Human-written Poetry and Is Rated More Favorably</article-title>. <source>Scientific Reports</source> <volume>14</volume>: <elocation-id>26133</elocation-id>. <pub-id pub-id-type="doi">10.1038/s41598-024-76900-1</pub-id></mixed-citation></ref>
<ref id="B19"><mixed-citation publication-type="webpage"><collab>R&#225;dio Slovensko</collab> <year>2024</year> <source>AI premen&#237; my&#353;lienky posluch&#225;&#269;ov na pr&#237;behy</source>. [AI Will Turn Listeners&#8217; Ideas into Stories]. <publisher-loc>Bratislava</publisher-loc>: <publisher-name>STVR</publisher-name>. [Last Accessed 30 April 2026]. Available at: <uri>https://slovensko.stvr.sk/rubriky/starsie-znovu-si-vypocujte/nekonecne-pribehy/375524/ai-premeni-myslienky-posluchacov-na-pribehy</uri></mixed-citation></ref>
<ref id="B20"><mixed-citation publication-type="book"><string-name><surname>Reichert</surname>, <given-names>R</given-names></string-name> <year>2008</year> <source>Amateure im Netz: Selbstmanagement und Wissenstechnik im Web 2.0</source>. <publisher-loc>Bielefeld</publisher-loc>: <publisher-name>Transcript</publisher-name>.</mixed-citation></ref>
<ref id="B21"><mixed-citation publication-type="webpage"><string-name><surname>Reimers</surname>, <given-names>I</given-names></string-name> and <string-name><surname>Waldfogel</surname>, <given-names>J</given-names></string-name> <year>2026</year> <source>AI and the Quantity and Quality of Creative Products: Have LLMs Boosted Creation of Valuable Books?</source>. NBER Working Paper No. 34777. <publisher-loc>Cambridge, MA</publisher-loc>: <publisher-name>National Bureau of Economic Research</publisher-name>. Available at: <uri>https://www.nber.org/system/files/working_papers/w34777/w34777.pdf</uri> [Last Accessed 28 April 2026].</mixed-citation></ref>
<ref id="B22"><mixed-citation publication-type="book"><string-name><surname>Schmidt</surname>, <given-names>S J</given-names></string-name> <year>2008</year> <source>P&#345;esahov&#225;n&#237; literatury</source>, <string-name><surname>Adamov&#225;</surname>, <given-names>Z</given-names></string-name> (trans.). <publisher-loc>Praha</publisher-loc>: <publisher-name>Institute of Czech Literature of the CAS</publisher-name>.</mixed-citation></ref>
<ref id="B23"><mixed-citation publication-type="book"><string-name><surname>Shirky</surname>, <given-names>C</given-names></string-name> <year>2010</year> <source>Cognitive Surplus: Creativity and Generosity in a Connected Age</source>. <publisher-loc>New York</publisher-loc>: <publisher-name>Penguin Press</publisher-name>.</mixed-citation></ref>
<ref id="B24"><mixed-citation publication-type="book"><string-name><surname>Simanowski</surname>, <given-names>R</given-names></string-name> <year>2008</year> <source>Digitale Medien in der Erlebnisgesellschaft</source>. <publisher-loc>Reinbek bei Hamburg</publisher-loc>: <publisher-name>Rowohlt</publisher-name>.</mixed-citation></ref>
<ref id="B25"><mixed-citation publication-type="book"><string-name><surname>Svedjedal</surname>, <given-names>J</given-names></string-name> <year>2000</year> <source>The Literary Web</source>. <publisher-loc>Stockholm</publisher-loc>: <publisher-name>Kungl. Biblioteket</publisher-name>.</mixed-citation></ref>
</ref-list>
</back>
</article>