Reimagining Storytelling: Intertextuality and Machine Learning as Tools for Narrative Creation
Joana Prochaska
By participating in the workshop BETWEEN THE LINES by visual artist LOREM, Joana Prochaska explored new ways of storytelling through the use of machine learning. Building on our recent lab, Latent Assemblies, which took place in December 2024, Joana delves deeper into LOREM’s workshop, uncovering the artist’s approach to reimagining storytelling. In this second part, she highlights LOREM’s use of machine learning as a tool to reveal hidden connections in storytelling and unlock new possibilities for narrative creation.
Storytelling has always been a fundamental way to understand and share human experience. Today, with advancements in Machine Learning (ML), new tools are emerging to transform the way narratives are created and interpreted. LOREM’s two-day workshop explored these possibilities, focusing on how ML can be used as a means to uncover the interconnectedness of texts and generate new forms of creative expression. By combining critical perspectives with hands-on techniques, participants examined how embedding and interpolation methods can reinterpret archives and reshape storytelling into something collective and relational.
This paper reflects on the workshop’s key themes, including intertextuality, the ethical considerations of ML, and its potential for creative innovation. Through this lens, ML is not seen as a replacement for human creativity but as a dynamic tool to reveal new layers of meaning within existing texts and to bridge gaps in cultural representation.
Intertextuality: A Foundation for Reimagining Narratives
Intertextuality, a concept championed by literary theorists like Ferdinand de Saussure and Jacques Derrida, highlights the relational nature of all texts. Derrida’s assertion that “there is no outside-text” underscores how texts exist within a web of interconnected meanings, constantly referencing and reshaping each other. During the workshop LOREM demonstrated how ML models like GPT-4 operate similarly. These models use latent spaces — abstract, lower-dimensional representations of high-dimensional data — where the meaning of any data point depends on its context and connections to others.
During the workshop, participants explored how ML’s latent spaces mimic this intertextuality, offering a way to combine different texts and create hybrid narratives. Embedding techniques, which translate words into mathematical vectors, allow for nuanced relationships between texts to emerge. By leveraging these techniques, we can engage with archives in novel ways, blending and reimagining content to produce something entirely new.
Challenging the Co-Creation Paradigm
Within his practice, Lorem emphasized the importance to critically examine how we conceptualize AI. Rather than perceiving AI as a sentient being he stated that it is more applicable to see it as a magnifying instrument. While the first metaphor treats AI as an autonomous entity, capable of independent thought and creativity, the second sees ML as a means to magnify and analyze data, revealing patterns and insights that might otherwise remain hidden.
Participants were encouraged to view ML as a method for exploring relationships between humans and their textual archives. By shifting the focus away from “co-creation” and towards thoughtful application, ML becomes a tool to deepen understanding and expand the boundaries of storytelling.
Ethical Considerations of Large Language Models (LLMs) and Dataset Augmentation
One of the workshop’s discussions centered on the ethical dimensions of ML. The datasets used to train LLMs, often vast and opaque, are riddled with biases and errors. For example, the LAION dataset, frequently cited in generative AI projects, disproportionately represents Western, Educated, Industrialized, Rich, and Democratic (WEIRD) demographics. Such biases skew the resulting models, limiting their ability to reflect diverse cultural perspectives.
LOREM emphasized the importance of addressing these gaps by augmenting datasets through embedding techniques. Embedding allows for the inclusion of underrepresented voices and perspectives, effectively reshaping the model’s knowledge base. This process situates ML as a tool for addressing inequities in data representation, enabling creators to work with richer, more inclusive narratives. By reframing ML in this way, storytelling becomes an opportunity to bridge cultural divides rather than reinforce them.
Embedding Techniques: Unlocking New Narrative Possibilities
A significant portion of the workshop was dedicated to hands-on experimentation with embedding techniques. Participants learned how embeddings, which capture the contextual relationships between words, could be used to generate new narratives. Unlike fine-tuning, which modifies the core model, embeddings overlay additional layers of meaning, situating the data within specific cultural or thematic contexts.
An intriguing application of this was the interpolation of texts. By blending one text with another — such as mixing 75% of one source with 25% of another — participants created hybrid narratives that challenged traditional ideas of authorship. These techniques allowed for outputs such as screenplays, poetry, and even entirely new literary genres, showing how ML can act as a bridge between various sources of inspiration.
Rethinking Authorship and Creativity
One of my takeaways from the workshop was the challenge it posed to traditional notions of authorship. Participants were encouraged to let go of control over the creative process, embracing the idea that the distinction between human and machine authorship is less important than the resulting narrative. This shift reflects a move towards collective and relational creativity, where the focus is on the interplay of ideas rather than individual ownership.
As LOREM noted, this perspective invites us to think about storytelling in a less individualistic way. ML becomes a means of exploring shared realities, blending personal and collective histories to generate narratives that transcend conventional boundaries.
Conclusion
The workshop highlighted the potential of Machine Learning to transform storytelling. By framing ML through the lens of intertextuality, LOREM invited participants to reimagine archives as living, relational spaces where new narratives can emerge. Through embedding techniques and thoughtful engagement with the ethical challenges of AI, the workshop demonstrated how ML can be used to amplify underrepresented voices, challenge traditional authorship, and foster innovative forms of expression.
Rather than positioning ML as a replacement for human creativity, the workshop emphasized its role as a tool to uncover hidden connections and generate fresh insights. By embracing this perspective, we can unlock new possibilities for narrative creation, using technology to enhance and expand our understanding of storytelling in the digital age.
Joana Prochaska is an emerging writer and researcher with a profound interest in environmental humanities, cultural analysis and artistic research. With an interdisciplinary background in Politics, Psychology, Law and Economics and Comparative Cultural Analysis, Joana writes about the intricate relationship of humans, environment and more-than-human species in the digital age.
