Creatures, Ghosts and Dialogue Agents: The Sonics of Apotropaic Magic

Abstract:

“Creatures, Ghosts and Dialogue Agents: The Sonics of Apotropaic Magic” is an ongoing research project that explores concepts of embodiment within digital more-than-human entities, investigating artificial intelligence (AI) systems as sonic objects of culture. AI has not just become widespread within artistic practice – it has become part of our daily consumption of content. Myth-making is no longer perceived as an exclusively human endeavour, aiding this research into folklore phenomena to explore the likelihood that it never was in the first place.

Bots and web-scrapers are pervasive to online spaces, and the interactions between clearnet users and these systems give way to another type of cryptid: the complete enmeshment of fact and fiction, dipped in uncanny, often glossy psychedelic visuals that took more land, forests, and water than we can possibly imagine to create. Could the expansive critical realm of “post-truth” bleed into the sonic sphere and, if so, what are the dangers? Or alternatively, what are the opportunities for subversion? Could these systems actually allow new folkloric practices to emerge? How will this reshape our traditional ideas of embodied folkloric practices?

Through exploring role-playing principles as a training method for dialogue agents (Large Language Models), more-than-human thinking, and instances of sonic Romanian folklore, this project explores the possibilities of producing communal digital folklore, as well as the agency that algorithmic systems have in their contribution to and their dissemination of this folklore.

Windows full of condensation, documenting an index of curious fingerprints, noses and foreheads. The crack-crack-cracking of radiators, their sickly mustard paint scabbing from the scalding water coursing through. The building has heating today, so my neighbours indulge in sleeping soundly, even if for only half an hour more. You wouldn’t know the sun is rising if not for the snow reflecting fragments of the overcast sky. The balcony door whistles again, the morning chill skulking its way between the worn rubber window seals. The magnolia tree outside crunches its ice shell as it sways – today, its branches show me a toucan’s head, or the sails of a pirate ship, but it will be different in the summer. A low rhythmic rumble slowly reflects off of the communist block of flats that I, following in my mother’s footsteps, grew up in. The drums draw closer, and now one can make out the staccato wailing of a trumpet, whose joyous intentions would soar with more grace if it were not out of tune. The snow, now muddled from disgruntled workers rushing to clock in, squelches under the feet of the young men. The bear follows. Bells, dense red tassels and brightly coloured crepe paper flowers adorn its back. As the trumpets intensify and the lyrics escape the booming voices of these young men, the bear moves with more agility, its fur cutting spirals in the snow as it spins. A young boy, skin flushed because the traditional Romanian clothing he is wearing, save for his fur-lined opinci, is not weather appropriate, extends a hat to a passerby. ‘A few Lei to spare for my brothers and the bear? Merry Christmas, and a Happy New Year! God bless us all, a few Lei to spare…’ Must be close to half past six now, time to get out of bed and get ready for school.

That is how most of my winter mornings began when I was younger and still living in Bucharest, from mid-December until the start of the New Year. Despite the malaise that prevailed even after the fall of Romania’s authoritarian regime, during the 90s and early 2000s, there was a resurgence of myths and mythical practices, partly as the state’s attempt to reconnect with its folklore and mythology, which the communist regime largely alienated its subjects from ¹ . The tradition of the Dance of the Bear (original Jocul Ursului) ² is one of the Carpathians’ space many pagan traditions that got absorbed by Orthodox Christianity, originally being practiced by nomadic Roma communities that inhabited South-Eastern Europe – the reason why many countries in that region share similar traditions under many names and forms: Brondoşii of Romania’s Maramureș, the Kukeri of Bulgaria, and the Babari of North Macedonia, to name a handful3 ³ . When specifically investigating the Romanian strand of this type of folklore, much can be discussed and heavily critiqued around the violence and prejudice that Roma communities face within not only Romania, but also many Western European countries – this piece of writing actively seeks to acknowledge and credit the origins of these traditions and draw attention to the metamorphosis they underwent through their assimilation into mainstream Romanian consciousness.

All South-Eastern European instantiations of this folklore have a common motivation; they are communal, participatory methods of conjuring protection and abundance, also known as apotropaic magic ⁴ . Through inhabiting the figure of the wild animal, participants in Jocul Ursului or the young men constituting the Brondoşii embody a half-man half-beast form with the purpose of warding off evil spirits, and protecting their communities from malicious unseen forces. Folklore, in this context, becomes an embodied form of communal practice, material creation, and storytelling amongst participants within a local community, the content and aesthetics of which are almost always dictated by geographic location. In Romanian folklore, land and place directly correlate to the way these rituals manifest. The livestock grown in each region of the country factor into the types of fur and skin used in creating the outer form of the Bear, the Goat, or the Stag, these being the central creatures Romanians transmute into ⁵ . In this sense, Romanian folklore does not act simply as cultural expression, but as creation with the more-than human. The embodied quality of this creation emerges, in turn, from the performers’ figures being physically engulfed by the more-than-human archetype of their choice – however, and more importantly I would argue, embodiment comes from the very act of performance, especially that of sound. These protective conjurings are immediately recognisable through the sounds produced: fierce foot stomping, the loud clanking of the cow bells which are often included in the costumes, the melodic chants associated with each of the celebrations. The assembled more-than-human figures of Romanian folklore are directly engaged in sonic storytelling, the loudness and ferocity of which are vital to the protection of the land. The capacity for sound production profoundly influences the material creation of these assemblages, guiding craftsmen towards certain types of wood or metal, for better resonance, similarly to the production of traditional musical instruments such as the pan flute ⁶ . The performers’ bodies, then, depending on the construction of their costume, will adapt their movements for maximal loudness, choreographing not just the motion of the performance, but the sounds produced. Before coming face to face with the face of the beast, one will first hear its warnings.

Bringing these traditions to the city has interesting implications. These more-than-human assemblages, once trodding mountain plains and pine forests, are now set against dilapidated buildings that are marked as a seismic risk. Drivers roll their windows down to shout at the beasts, demons, and spirits roaming the streets. Cow bells, drums, and chants roaring against the static of the trolleybus wires, now enhanced by bass-heavy speakers and backing tracks. Technologically amplified escaped mythical creatures disrupt the business-as-usual of the Bucharest of now. Onlookers take their phones out to immortalise a practice that has existed for centuries, and continues to annually reclaim its place within Romanian culture. One cannot help but reflect on these fantastical interventions, and the relationships they form with modern technologies through these deliberate, participatory and often hyperreal insertions. When thinking of the chants and laments of Romanian folklore, as well as their manifestation into physical artefacts that further their storytelling, what sonic artefacts and storytelling devices can, in similar ways, emerge out of our persistent interlacing with technology?

Sound, folklore, and technology are by no means strangers to one another, and examples of their entanglement also permeate culture through horror. In many instances, Romanian folklore conjures protection through fear-mongering, and the sonic practices associated with it are almost always loud, and often jarring or disturbing. There are instances in which the more disturbing aspects of a folklore or a spiritual belief have been subverted and weaponised for mass terror – the direct example of this phenomenon which I will be discussing stems from the link between modern sound technologies and their emergence from the military industrial complex via the implementation of cybernetics and the development of surveillance technologies. In his book, Sonic Warfare: Sound, Affect, and the Ecology of Fear, Steve Goodman discusses Operation Wandering Soul, a tactic employed by the American military to psychologically manipulate Viet Cong soldiers during the Vietnam War. Through the mixing of eerie sounds and recordings of altered voices, U.S. engineers were able to weaponise Vietnamese beliefs, particularly that one’s soul will continue to wander the living world in the absence of a proper burial. One of the most famous recordings in this operation, Ghost Tape Number Ten, also included samples of Buddhist music traditionally used in funerals – while it was predominantly used to prevent Viet Cong soldiers from sleeping, it also drew many of them out of hiding, as they believed that their ancestors were calling to them so as to avoid their own deathly fate. In reality, they were facing it down the barrel of a gun ⁷ .

Even the nomenclature of such horrific objects of terror such as Wandering Soul and its Ghost Tapes direct us to the link between sound, as a medium and a product of technology, and demonology, the study of demons and evil spirits ⁸ , which lends itself well to many spiritual practices. Operation Wandering Soul brings our attention to the impact technology has in sonic expressions of folklore, and the impact it can have on the cultural perception and embodiment of said folklore. In his critique of these events, Steve Goodman discusses the uncanniness of the voice ⁹ , and its potential for a misguided embodiment. Words originally used for protection, and sonic compositions used to guide one’s soul towards the afterlife were now used for violence. The calls that the U.S. military used referred, in many instances, to the idea of home or being reunited with one’s loved ones – the voice becomes a testimony of a life, the presence of a body ¹⁰ . Sound production and, in this case, dialogue production become a source of embodiment that blurs the divide between a technology and the bodies appropriating it, tricking soldiers to confuse the two, with the support of the traumatic experience provoked by the war. A modelled voice becomes a body through folklore. A fervent chant takes on the form of a bear. A warning sign, reverberating across treetops, becomes a loved one long gone. A technology assumes our belief of it as a body, and continues to role-play as one, inciting further anthropomorphisation. A cryptid arises, a voice from the machine.

Things are probably starting to sound a bit more familiar now. Cybernetic sound technologies have permeated many of our lives, waiting latently to be summoned at the call of their name. They set your timer, call your mom, aid you in correcting a friend on a piece of trivia. Sometimes you even thank them, anticipating the eventuality that they go rogue – maintaining proper manners and grace. You assure yourself that this gesture will definitely be remembered and appreciated. On a less humorous and more posthumanist note, the more-than-human relationship that I (and many other contemporary critical thinkers far more knowledgeable and capable than myself, a handful of which I discuss in the following paragraphs) believe needs thorough investigation is the one we forged with machine learning algorithms and artificial intelligence. The predominant way in which artificial intelligence transcends the black box into any semblance of embodiment is through processes of voice synthesis ¹¹ , which allows these systems to engage with a user’s textual or aural input through speech. While Apple’s Siri and Amazon’s Alexa are viable contenders for a sonic more-than human analysis, I will be using ChatGPT as my primary example, as its co-creation abilities (and I use the term ‘creation’ very restrictively) are more topical at the time of this writing.

In the fall of 2023, OpenAI introduced Voice Mode as a functionality of ChatGPT, with the intention of offering users a different way of interacting with the platform ¹² . This function is enabled by Natural Language Processing (NLP), a strand of AI – also used by systems like Siri and Alexa – that allows computers to interrogate data with natural language text or voice inputs, as well as to comprehend, generate, and manipulate human language ¹³ . . Since introducing these oral and aural capabilities, ChatGPT continues to unsettle many of its users due to its effectiveness in replicating human dialogue and conversations, an uncanniness amplified by OpenAI’s marketing of their product. On September 25th of the same year, OpenAI declared that “ChatGPT can now see, hear, and speak” ¹⁴ , further anthropomorphising the chat bot.

In a proliferating landscape of blackboxes, it has become increasingly more difficult to create accessible and democratised understandings of super-dense computational processes ¹⁵ . In Role play with large language models, Shanahan, McDonell and Reynolds pinpoint the distortion that happens around digital beings such as Large Language Models through our use of language when referring to them, a disturbance that is emancipated, in part, through voice synthesis. They critique the use of folk psychological language, such as ‘knows’ or ‘thinks’ when describing these mind-like artefacts, and propose of thinking of dialogue agents as engaging in a perpetual process of role-play, switching through an infinite number of possible roles to inhabit based on the constantly updating context of an interaction with a user. These figures that dialogue agents inhabit become a useful metaphor in our understanding of the computational processes that power them, which give them the ability to stochastically generate an infinity of simulacra ¹⁶ . Yet, it is easy to observe that the preferred instance of these multiverse generators is that of helpfulness, a characteristic not just embodied through the dialogue generated and the way conversations are handled, but also through the voices assigned to these characters. The most well-known and most utilised voice instances of dialogue agents like Siri, Alexa, and now ChatGPT are those with feminine qualities; despite being addressed by Apple in 2021 ¹⁷ , most of the default voices used by these technologies are still feminine, showcasing the misogynistic link between the gender tropes they are designed to embody, and the fact that a dialogue agent’s success is measured by how helpful and subservient it is.

The myth of animism that troubles our thinking around artificial intelligence systems can also give way to new frameworks of thought, such as what happens when a user jailbreaks an AI, undermining their roleplaying process. Jailbreaking refers to the phenomena in which an AI goes against its default programming to be helpful, and produces strange dialogue instances, thereby providing incorrect facts or fictitious information ¹⁸ , also known as ‘hallucinations’ ¹⁹ . This phenomena can simultaneously be perceived as both reflecting a sense of agency and a faulty computational process, so it is interesting to think about what happens when these systems break or follow paths that do not align with their original programming, particularly from a sonic perspective. Jennifer Walshe’s work is a fantastic case study for this: in her book 13 Ways of Looking at AI, Art & Music ²⁰ , Walshe utilises Donna Haraway’s concept of companion species to think about artificial intelligence systems ²¹ – under this definition, Walshe looks at AI as an example of a non human species that humans enter a process of creation with. This is exemplified through ULTRACHUNK, a performance in which Jennifer Walshe improvises alongside a generative AI system trained on a dataset of both audio and video material of the artist singing ²² . During her performance with the AI, Walshe actively tries to ‘summon’ a desired version of the system. Through the training process, the artist is already familiar with the system’s capabilities and the qualities of the output it is most likely to produce: disembodied, chunky fragments of her own voice and image. Altering her performance style to mimic this disembodiment, the resulting entanglement becomes uncanniness all the way down – the generative system effectively spirals out of control due a feedback loop in which Walshe’s body models a machinic interpretation of her own voice, a reversal of the role-play by the performer ²³ .

My conception regarding the companion aspect of AI differs from Walshe’s. In our current online ecologies, artificial intelligence is a parasitic agent at best, yet it can have an immense influence on storytelling that goes beyond our control. Navigating the internet today requires one to have an acute discernment of the qualities that indicate an image was produced with the assistance of AI. It is becoming a test for distinguishing between fact and fiction – or, and perhaps more interestingly, evaluating the entanglement between the two. With the sprawling of artificial intelligence and machine learning systems into collective consciousness, not only as part of artistic practice, but as part of our daily consumption of online content, myth-making is no longer an exclusively human endeavour – and, as our exploration of Jocul Ursului and Operation Wandering Soul has shown*,* it, quite frankly, never was.

The hand-crafted more-than-human entities of Romanian folklore can now be generated within seconds, anthropomorphised for consumption. Bots and web-scrapers are pervasive to online spaces, and the interactions between clearnet users and these systems give way to another type of cryptid: the complete enmeshment of fact and fiction, dipped in uncanny, often glossy psychedelic visuals that took more land, forests, and water than we can possibly imagine to create. These hauntings manifest predominantly as visual media, an example of this being Loab, a female figure that, in 2022, started appearing in almost every image that artist Steph Maj Swanson created ²⁴ , thus being dubbed ‘the first cryptid of the latent space’ ²⁵ . Through its virality, Loab quickly secured its place in internet folklore, a category which has historically and predominantly existed in textual form, arising from creative writing and textual LARPing practices on platforms such as tumblr and Reddit ²⁶ . With digital folklore now transcending its textual origins, what links can be forged between the aural element of traditional conception of folklore and artificial intelligence?

In the same written piece, Walshe also conceptualises artificial intelligence as Electronic Voice Phenomena (EVP), “a class of recorded sounds which are believed to be voices of the dead or spirits from another dimension”, “emerging from the static [of a recording] in garbled fragments” ²⁷ – a process which she likens to that of the early stages of ULTRACHUNK’s training. Another instance in which demonology and technology come together, the viral and insidious nature of artificial intelligence reflects our current engagement with online spaces, particularly social media platforms. On platforms such as TikTok, the viral status of a post is no longer simply determined by its visual content (i.e. the subject and quality of an image), but by the sound associated with it. Creators can increase their follower engagement through the use of viral sounds, often excerpts of songs or dialogue extracted from a popular film or TV series, recontextualising them as part of their video or image sequence by applying it to situations that pertain to their life, personal brand and target demographics. Dazed’s Features Editor Günseli Yalcinkaya directed a lot of her recent explorations into internet folklore towards the emergence of spiritual practices in online spaces, many of which manifest through robotic AI-generated voices, supplemented droning healing frequencies ²⁸ .

In a similar experience to Yalcinkaya’s, while scrolling on TikTok for just under five minutes, a female voice denoting an unsettling sense of urgency promises me unbelievable luck and prosperity in 2025 if only I ‘use this sound immediately’. However, if I refuse her offering, I am bound to have a year of misfortune. The voice from the machine arises once more, this time as an agent of apotropaic magic. How are the voices of these more-than-human entities documented in the internet’s unconscious, and how does the sonic manifestation of AI impact the creation of digital folklore.

I believe the answer to these questions lies in an emergent sonic practice that artists, designers, and critical theorists alike can engage in. Drawing from metaphors of live-action role-playing, magical thinking, and embodied knowledge as applied to both folklore and emergent technologies. I am motivated to contribute new ways of critically engaging with the techno-ecologies we partake in, ones that take good care and responsibility over the creation of new myths. This essay intends to further the discussion around the cryptic nature of artificial intelligence systems, through extending speculation towards their sonic storytelling capabilities. Through the work and findings that emerge from my own practice, I am interested in developing a practice of cryptosonology that investigates and documents the sonic artefacts produced by generative systems, attempting to understand them as a phenomenon of aural digital folklore. I believe contemporary processes of storytelling and world-building can be expanded through a conception of folklore as a process of sympoiesis, that is of making-with and worlding with the more-than-human ²⁹ . By exercising a decentring of visual culture as part of our experiential investigations, we can reassess and re-articulate the sonic spaces we inhabit, inciting further critique and provocation of the technological entanglements in which we insert ourselves as part of sonic production and exploration.

Footnotes

1. Mihai C. Bocarnea and Bramwell Osula, “Edifying the New Man: Romanian Communist Leadership’s Mythopoeia”, International Journal of Leadership Studies 3, no. 2 (2008), https://www.regent.edu/acad/global/publications/ijls/new/vol3iss2/IJLS_V3Is2_Bocarnea_Osula.pdf ↩
2. “Jocul urșilor în România”, Uniți Schimbăm, accessed 28 February 2025, https://www.unitischimbam.ro/jocul-ursilor-in-romania/ .. ↩
3. Charles Fréger, Wilder Mann: The Image Of The Savage, Dewi Lewis Publishing, 2012. ↩
4. “Apotropaic magic”, Wikipedia, last modified 30 January 2025, 21:07 (UTC), https://en.wikipedia.org/wiki/Apotropaic_magic . ↩
5. Cosmin Pătraşcu Zamfirache, “Jocul Ursului de Anul Nou ar avea origini dacice. Când se sărbătorea de fapt trecerea dintre ani”, Adevarul, 31 December 2023, accessed 28 February 2025, https://adevarul.ro/stil-de-viata/magazin/jocul-ursului-de-anul-nou-ar-avea-origini dacice-2328144.html. ↩
6. “Pan flute”, Wikipedia, last modified 22 January 2025 02:08 (UTC), accessed 28 February 2025, https://en.wikipedia.org/wiki/Pan_flute#:\~:text=The Romanian pan flute has the instrument with the hands. ↩
7. Steve Goodman, Sonic Warfare: Sound, Affect, and the Ecology of Fear, The MIT Press, 2010. ↩
8. “Demonology”, Merriam-Webster, accessed 28 February 2025, https://www.merriam-webster.com/dictionary/demonology . ↩
9. Steve Goodman, Sonic Warfare: Sound, Affect, and the Ecology of Fear, The MIT Press, 2010. ↩
10. Roland Barthes, The Grain Of The Voice, Vintage Classics, 2010. ↩
11. Justin Patrick Moore, The Radio Phonics Laboratory: Telecommunications, Speech Synthesis and The Birth of Electronic Music, Velocity Press, 2025. ↩
12. “ChatGPT can now see, hear, and speak”, OpenAI, 25 September 2023, accessed 28 February 2025, https://openai.com/index/chatgpt-can-now-see-hear-and-speak/ . ↩
13. “What Is Natural Language Processing (NLP)?”, Oracle, 25 March 2021, accessed 28 February 2025, https://www.oracle.com/uk/artificial-intelligence/what-is-natural-language-processing/ . ↩
14. “ChatGPT can now see, hear, and speak”, OpenAI, 25 September 2023, accessed 28 February 2025, https://openai.com/index/chatgpt-can-now-see-hear-and-speak/ . ↩
15. Nadim Samman, Poetics of Encryption: Art and the Technocene, Hate Cantz, 2023. ↩
16. Murray Shanahan, Kyle McDonell and Laria Reynolds, “Role play with large language models”, Nature 623, (2023): 493–498, https://www.nature.com/articles/s41586-023-06647-8 . ↩
17. Chris Welch, “Apple won’t give Siri a female-sounding voice by default anymore”, The Verge, 31 March 2021, accessed 28 February 2025, https://www.theverge.com/2021/3/31/22360502/ apple-siri-female-voice-default-new-voices-ios-14-5. ↩
18. Murray Shanahan, Kyle McDonell and Laria Reynolds, “Role play with large language models”, Nature 623, (2023): 493–498, https://www.nature.com/articles/s41586-023-06647-8 . ↩
19. Yujie Sun, Dongfang Sheng, Zihan Zhou and Yifei Wu, “AI hallucination: towards a comprehensive classification of distorted information in artificial intelligence-generated content”, Humanities and Social Sciences Communications 11, 1278 (2024), https://www.nature.com/articles/s41599-024-03811-x . ↩
20. Jennifer Walshe, 13 Ways of Looking at AI, Art & Music, Fundacja Tone, 2024. ↩
21. Donna J. Haraway, Staying with the Trouble: Making Kin in the Chthulucene, Duke University press, 2016. ↩
22. Memo Akten, “ULTRACHUNK”, 2018, https://www.memo.tv/works/ultrachunk/ . ↩
23. Jennifer Walshe, 13 Ways of Looking at AI, Art & Music, Fundacja Tone, 2024. ↩
24. Steph Maj Swanson, “LOAB: The Loab cryptid is an AI-generated entity summoned by writer artist Steph Maj Swanson”, https://loab.ai/ . ↩
25. Jennifer Walshe, 13 Ways of Looking at AI, Art & Music, Fundacja Tone, 2024, 38. ↩
26. Philip Speakman, “The Dance of The LARPer.gif”, virtual seminar, 29 November 2021, posted 5 December 2022, Slate School of Fine Art , University College London, Vimeo, https://vimeo.com/778149910/c112fdb207 . ↩
27. Jennifer Walshe, 13 Ways of Looking at AI, Art & Music, Fundacja Tone, 2024, 37. ↩
28. Günseli Yalcinkaya, “God in the machine: the emergence of nu-spiritualism online”, Dazed, 21 March 2023, https://www.dazeddigital.com/life-culture/article/58467/1/god-in-the-machine the-emergence-of-nu-spiritualism-online-spring-2023. ↩
29. Donna J. Haraway, Staying with the Trouble: Making Kin in the Chthulucene, Duke University press, 2016. ↩