Chasing mistakes
Prompts and faults as parts of a creative process
-
Jonas Sjøvaag
CreaTeMe / University of Agder
A manifest, of sorts, made into an artwork using the same basic idea as described in this paper, but applied to visual output. It is made from two screenshots of mock code, written in a python-like style, in Sublime text, then assembled, reworked, and finished, in Photoshop.
It is based on a working method I employ regularly in my other profession, which is that of a graphical designer. More examples of this available on supremeconnection.no
Introduction
In the evolving landscape of contemporary art, algorithmic systems have opened up new dimensions for creative practice. Within the Scandinavian model of artistic research, where artistic processes themselves are seen as knowledge-producing, I explore not only the use of generative scripts as tools, but also as unconscious collaborators. This exposition focuses on a Python-based script that reinterprets written text, introducing distortions that reveal new creative possibilities.
The central research question guiding this inquiry is:
How can system-generated misinterpretations be harnessed as a method for artistic creation, challenging traditional notions of authorship, intention, and aesthetic coherence?
Rather than treating the script’s limitations as technical failures, I argue that they offer rich terrain for creative exploration. This approach builds on historical precedents such as the Dadaist cut-up technique and Alvin Lucier’s I Am Sitting in a Room (1969), which involved repeatedly playing and recording a voice in an acoustic space, allowing the room’s resonances to gradually transform the speech. In my work, this becomes a digital equivalent—where linguistic distortion replaces acoustic decay.
In dialogue with the past and the present
As Tom Milnes (2023) writes in his JAR exposition Ephemer(e)ality Capture: “Reflective, transparent, specular and patterned/repetitive objects were used to confuse the imaging algorithm … The research tests the limits of photogrammetry in an effort toward new image-making methods.” Similarly, my script deliberately confuses a linguistic system by running text through recursive speech-to-text (STT) and text-to-speech (TTS) cycles, producing glitches that resist clarity and challenge stable meaning. These errors are not interruptions but entry points—openings into alternative forms of interpretation and aesthetic experience.
Recent artistic research has foregrounded error and systemic misfunction not merely as aesthetic strategies, but as epistemological entry points. In the exposition mentioned above (Ephemer(e)ality Capture), algorithmic faults in photogrammetry become central to artistic inquiry, revealing how computational vision misinterprets material phenomena. Similarly, András Blazsek’s ∂ Topological Landscapes(2021) draws on failed scientific instrumentation and archival glitches to interrogate how technologies mediate—and often distort—truth claims. Both projects align with the artistic impulse to extract meaning from failure, and to leverage breakdown as a way of making the machinic visible.
My project resonates with this lineage. The misread texts generated by my script—repetitive, skewed, misaligned—operate in the same conceptual space as optical glitches or historical technical errors. They are not aesthetic flourishes but disruptions that invite, or even force, reflection. This project positions itself within this broader conversation by extending glitch-based methodologies into the realm of language and sonic composition, adding a textual and performative layer to the visual and archival modes explored by Milnes and Blazsek.
The use of error, glitch, and randomness in art has long been a strategy to challenge conventional aesthetic norms. Artists such as William Burroughs, Brian Gysin, and Nam June Paik foregrounded disruption as method. More recently, the work of practitioners engaging with computational systems has demonstrated how procedural unpredictability can produce aesthetically meaningful content.
Theoretically, this project also draws from John Searle's Chinese Room Argument(1980), which posits that syntactic operations do not constitute understanding. By applying this idea to script-generated outputs, I explore how meaning can emerge not from intentional semantics but from the artist's interpretive framing of structural error. It also shows, how old ideas emerge in a new form, through new possibilities made accessible to us by forces in our surroundings that most likely had other results in mind when creating their inventions and tools.
Method: The Python Script
Originally developed to generate ideas for lyrics, the script processes a base text through simple string manipulation and iterative STT/TTS cycles. It "reads" a text aloud, transcribes it back, and repeats this process multiple times, akin to Lucier’s recursive acoustic transformation.
Each cycle introduces subtle misreadings that diverge further from the original—transformations that are deliberately left uncorrected. Instead, they are embraced and used as raw material for further creative work. One day, the phrase "Blues for N0" emerged from this process—strange, uninvited, oddly evocative, and, importantly: likeable.
"In Rome": Iterative Transformation in Practice
The first outcome of this project, In Rome, is the one most directly inspired by Alvin Lucier's I Am Sitting in a Room (1969). Lucier’s piece is, as mentioned above, a recursive act of recording, and this conceptual strategy serves as a foundation for my own method, transferring the process from acoustic space into digital space, where the feedback loop is not spatial but linguistic and algorithmic. My approach borrows and recontextualizes this logic in a digital domain: using a custom Python script, or, more specifically, I loop a written text through a cycle of TTS and STT, allowing each iteration to introduce new distortions.
What begins as coherent language gradually collapses under its own weight—words slip, meanings shift, rhythms erode. The process reveals how quickly clarity can become abstraction, and how fragility resides at the core of every linguistic system. Each mutated version is both a trace of its origin and a new composition in its own right.
This erosion of structure and meaning serves as a method and a metaphor: it reflects how interpretation—both human and machinic—is never neutral. The work acts as a digital whisper game, where what is passed on is not preserved, but transformed. These mishearings aren’t corrected; they’re curated and accepted as part of the work. The project underscores the tension between control and surrender that characterizes much of artistic research involving technological systems.
Importantly, In Rome also produced a strangely interesting phrase—"Blues for N0"—which went on to inspire its own separate work (shown below). That moment of unexpected output marked a pivot from structured iteration to singular expression.
"Blues for N0": Embracing Misinterpretation
Blues for N0 is built around this single, unintended phrase—a textual slip generated during the In Rome process. There was something evocative and unplaceable about it, as if the script had stumbled into poetry. I decided to treat that phrase as a prompt in itself.
In this piece, the script’s interpretation of a short text is recorded only once—no recursive loop, no iterative processing. Instead, the result is paired with two video tracks: the first generated using Runway Gen-2 (prompted by the original text), and the second a response using footage I recorded myself. These two visual streams reflect different aesthetic logics—one autonomous and generative, the other human and interpretive.
In this way, Blues for N0 explores how meaning shifts not only through language but across media and authorship. It is also a system producing language without understanding it, and as such, Searle's argument3, mentioned before, is acted out here. Further, I frame the system’s misfire as a kind of accidental lyricism. The work doesn’t argue for intelligence within the system, but rather invites us to consider how meaning often emerges despite, or even because of, miscommunication. Here, the creative spark came from a misfired phrase, something I never would have written myself. The resulting audio was paired with generative video and later contrasted with hand-shot footage. This duality highlights tensions between programmed and human authorship, and between accidental genesis and intentional framing.
The piece will debut ath the Gothenburg Fringe in the autumn of 2025, but was originally made for the ARAK conference in Oslo in spring 2024.
By focusing on the script’s imperfections, both In Rome and Blues for N0 challenge the assumption that creative tools should strive for perfection. Instead, they foreground how real value often emerges from semantic slippage, unpredictability, and misinterpretation. This echoes critical reflections found in a JAR-publication (issue 0), where Otto von Busch, in the abstract of his exposition "Research Navigations" (2011) cautions that “the ‘reflective practice’ promoted by Donald Schön (1983) runs the risk of mere self‑gratification”. In my context this specifically implies that I see that faults will help avoiding recurring artistic situations. In the words of Tom Waits, in an interview quoted by Corinne Kessel (2009, 22): “You have to be careful when playing is no longer in the mind but in the fingers, going to happy places. Your hands are like dogs, going to the same places they've been…”. Further in that interview, he talks about playing instruments that are new to him to avoid this automated recurrence; chasing mistakes, or forcibly venturing into unknown, in much the same way I look for new meaning by purposely imposing a new semantic understanding by having an unconcious collaborator reciting my words back to me after interpreting them on its own.
As such, my work uses reflection—not to confirm my own intention—but to approach the edges of meaning, letting the system’s glitches generate interpretive space that wouldn’t otherwise appear. This framing moves it into a more nuanced artistic‑research conversation: reflection isn’t a theoretical comfort; it’s a strategy to engage more deeply in creative potential.
Composition, lyrics, vocals & synths by me. Guitars by Hans Martin Austestad. Recorded and mastered by Færder Audio. Video prompted by me, using Runway gen-2.
"Walk Away": visualizing lyrics through generative media
Video generated by Runway Gen-2, fed line by line, generating a few assets from each, then sorted and finished in Adobe premiere pro.
This is yet another example of the same recurrence-avoiding approach, this time taking a piece of text, structured and validated by my own approach and background, but using that text to genereate imagery through a similar, but this time visual, unconscious co-creator.
The Script as Collaborator
The artistic process was driven by improvisation, reactivity, and play. The script’s role was not deterministic but dialogical—an unpredictable partner whose outputs invited reinterpretation. It often felt like working with a musician who only responds in riddles. Rather than generating finalized content, it triggered associative thinking, surprise, and reflection.
However, the validity of this practice does not lie in the novelty or amusement often associated with early-stage AI or generative systems. What matters here is how the tool—in its failure to replicate human language seamlessly—reveals our own mechanisms of semantic interpretation. It asks us to question what meaning is, and how it is produced, especially when we are faced with language that has no obvious origin or logic. The script doesn’t just produce noise; it invites the artist, and any listener, to listen again, interpret again, and respond again.
Moreover, the tool’s limitations are inextricably tied to questions of access and inequality. The system used in this research was built with free, publicly available components—tools that are widely accessible but notably flawed compared to premium, paid alternatives. This disparity reveals a deeper layer in the work: those who can afford the most refined systems often bypass the very unpredictability that makes this kind of creative process possible. In contrast, the glitchy, unpolished results of lower-tier systems become fertile ground for exploration, and in many ways, are more revealing of the system’s internal logic.
As such, the script becomes a methodological device that reflects the values of artistic research: emergence, experimentation, and open-ended inquiry, while also foregrounding broader systemic concerns about technological access and the aesthetics of imperfection. As Zylinska (2020) notes, access to generative technologies is never neutral—it reflects broader dynamics of power, privilege, and visibility.
Ultimately, this research suggests that error, miscommunication, and systemic limitation are not merely artifacts to be corrected, but tools for uncovering latent creative potential.
References
- Burroughs, W.S. & Gysin, B. (1978). The Third Mind. Viking Press.
- Lucier, A. (1969). I Am Sitting in a Room.
- Searle, J. (1980). "Minds, Brains, and Programs." Behavioral and Brain Sciences, 3(3), 417–457.
- Milnes, T. (2023). Ephemer(e)ality Capture: Glitch Practices in Photogrammetry. Journal for Artistic Research (JAR), Issue 36. https://jar-online.net/en/issues/36/ephemer-e-ality-capture
- Zylinska, J. (2020). AI Art: Machine Visions and Warped Dreams. Open Humanities Press.
- Kessel, C. (2009). The Words and Music of Tom Waits. Praeger. https://dokumen.pub/the-words-and-music-of-tom-waits-9780313349065-0313349061.html
- von Busch, O. (2010). Research Navigations. Journal for Artistic Research (JAR), Issue 0. https://jar-online.net/en/exposition/abstract/research-navigations
- Blazsek, A. (2021). ∂ Topological Landscapes. Research Catalogue. https://www.researchcatalogue.net/view/817407/817408