Invented Traditions: AI, Synthetic Folklore, and Cultural Appropriation

The digital ecosystem is accelerating the birth of new "invented traditions." Through the lens of "algorithmic folklore," we analyze how generative Artificial I

Artificial intelligence does not create folklore out of nothing. Like a tireless archivist endowed with superhuman combinatorial abilities, AI reshuffles textual fragments, visual styles, and pre-existing narratives, producing new forms that, if inserted into the right social circuits, can rapidly transform into shared beliefs. The heart of the contemporary debate on digital anthropology is not whether technology can or cannot generate fascinating stories, but how to distinguish a new tradition – openly and consciously declared as such – from a synthetic myth passed off as authentic historical memory.

The Invention of Tradition and Algorithmic Folklore

The idea that a tradition must possess a millennial lineage to be considered authentic is, for the most part, a historical fallacy. In the foundational text The Invention of Tradition, historians Eric Hobsbawm and Terence Ranger have extensively demonstrated that countless traditions, perceived by the collective as deeply rooted and untouchable, are in reality of relatively recent origin, often constructed deliberately to create a sense of reassuring continuity with the past. This academic awareness is the prism through which to read the present: the fact that a phenomenon is "invented" does not at all mean that it is false or socially irrelevant. Invented traditions, once adopted by a collectivity, become socially and emotionally real.

Today, this process of cultural creation is enormously accelerated by the digital ecosystem, giving rise to what researchers call Algorithmic Folklore. In a recent conceptualization published in the Open Library of Humanities (2026), algorithmic folklore is defined as an unprecedented repertoire of genres and practices born from the friction between the vernacular creativity of users and everyday automation. The algorithm no longer serves merely as a passive object for urban legends, but has transformed into an active medium that participates firsthand in the creation and circulation of content. The case of "Crungus," a monstrous cryptid generated by an artificial intelligence and rapidly transformed into a viral meme complete with its own mythology, empirically demonstrates AI's capacity to serve as an incubator for new popular mythologies.

This intersection is now a field of rigorous academic inquiry. The four-year project ALGOFOLK, conducted by the Center for Digital Narrative at the University of Bergen, systematically studies how recommendation engines and machine learning models are giving rise to new creative practices. The phenomenon embraces an entire sociological spectrum, ranging from urban legends about the supposed consciousness of systems like GPT-4, to popular conspiracy theories amplified on TikTok, to memes co-created in an incessant synergy between humans and neural networks.

The Myth Machine: Cultural Limits and Datasets

If on the one hand generative artificial intelligence can structure forms that echo the aesthetics of folklore, on the other its actual "understanding" of cultural roots remains profoundly limited and geopolitically unbalanced. An illuminating case study presented at ACL (2025) analyzed in depth the knowledge of Large Language Models (LLMs) regarding Japanese folk tales and the complex taxonomy of yokai, the supernatural creatures of Japanese folklore. The research highlights that the cultural competence of these models is almost exclusively calibrated on Anglophone communities and Western narrative paradigms. When a language model is queried about non-Western mythologies, it tends to marginalize them, brutally simplify their traits, or reinterpret them through entirely foreign cultural lenses. This constitutes technical proof that an artificial intelligence can replicate the syntax of a folkloric form without in any way understanding its original social context, its rigid ritual functions, or the taboos inherent in the narrative.

Academic efforts to mitigate this technological gap collide with insurmountable sampling barriers. A study from the University of Edinburgh (2024), focused on Irish and Gaelic folklore, demonstrated that subjecting language models to continuous pre-training on specific datasets of folkloric texts significantly improves their capacity for classification and representation. However, the final results remain heavily conditioned by the chronic scarcity of digitized data for low-resource languages. AI can undoubtedly support humanities institutions in the massive classification of immense narrative heritage, as demonstrated by studies on the use of GPT to annotate gender stereotypes and recurring cultural traits in literary motifs, a field in which the machine has reached levels of consistency comparable to or superior to those of human annotators. However, aseptically classifying a database of fairy tales is not equivalent to interpreting their spiritual essence, nor to preserving the vital memory that only the breath of a community can pass down.

Cultural Appropriation and Colonialism of Indigenous Data

The hybridization between neural networks and intangible heritage opens ethical and legal scenarios of extreme criticality, in which the boundary between technological innovation and cultural appropriation blurs until it disappears. The recent report Artificial Intelligence and Culture published by UNESCO (2025) issues an unequivocal institutional warning: the AI ecosystem urgently needs diversity, transparency, and accountability throughout the entire lifecycle of information systems, promoting participatory governance and data management practices that are directly guided by local communities. The organization explicitly warns that algorithms, when designed to impose a global aesthetic uniformity or structured to amplify disinformation for engagement reasons, inexorably end up weakening and flattening human cultural diversity.

This risk takes on dramatic contours when it touches the heritage of minorities and indigenous peoples. The new UNESCO guidelines on indigenous data sovereignty in AI development (2023) assert the indispensability of free, prior, and informed consent, as well as community control and direct participation of indigenous peoples in the technological supply chain. If a generative model – trained by scraping unauthorized historical archives or ethnographic diaries – begins to produce on an industrial scale "synthetic native mythologies," "imaginary ancient rites," or fake reproductions of religious symbols without any governance by the direct descendants, we are not witnessing the democratization of art. We are witnessing commercial extraction disguised as creativity.

As analyzed in an essay in the International Journal of Cultural Property (2026), unbridled automation in the digitization of intangible cultural heritage carries with it the shadow of a veritable "data colonialism." It has become crucial, under international law, to defend the legal and moral right of indigenous communities to maintain rigorous control over the collection, access, interpretation, and economic reuse of their data and their memory.

The central question that intellectuals, sociologists, and legislators must ask themselves today is not whether artificial intelligence can invent a plausible and captivating legend. The truly decisive issue concerns the allocation of epistemological power: who will have the authority and the capital to establish that that legend is true, historically grounded, or culturally sacred? A generative model can in a few seconds generate a frightening urban monster, stage an imaginary festival curated down to the smallest detail and songs, or trace the fake genealogy of a family of archaic symbols. However, it falls exclusively to the complex network of human relationships to decide whether that aesthetic form will transform into a true shared memory, a passing role-playing game, or whether it will become the umpteenth poisonous cog in mass disinformation.

Folklore does not die at the exact moment it is invented, hybridized, or contaminated; it dies only when people stop telling it, finding it meaningless. The most insidious risk posed by generative artificial intelligence does not lie in its immense capacity to create new and infinite stories, but in the potential to make us lose the fundamental difference between a narrative that grows organically together with a community and a synthetic narrative that is surreptitiously imposed upon it.

Bibliographic References and Sources

[Four Theses on Algorithmic Folklore — Open Library of Humanities, 2026]
[The Invention of Tradition — Hobsbawm & Ranger, Cambridge University Press]
[ALGOFOLK: Algorithmic folklore — University of Bergen, Center for Digital Narrative]
[Do Large Language Models Know Folktales? A Case Study of Yokai in Japanese Folktales — Findings of ACL, 2025]
[Evaluating and adapting large language models to represent folktales — University of Edinburgh, 2024]
[Using Large Language Models to Classify Cultural Heritage]
[Artificial Intelligence and Culture — UNESCO, 2025]
[New report and guidelines for Indigenous data sovereignty in AI developments — UNESCO, 2023]
[Indigenous data sovereignty in intangible cultural heritage governance — International Journal of Cultural Property, 2026]

Article by the Editorial Staff of La Bussola dell'IA