Encoding Culture:
Year: 2025
Type: article
Abstract: The formulation of a Swedish Cultural Canon by a committee of experts has sparked debate among cultural agents and the public: while some perceive it as controversial, others value its role in categorising and conserving key historical milestones of Swedish culture. “Encoding Culture” speculates that putting together cultural data as images and text for generative AI training may replicate the mechanisms of canon formation—both reflecting and reinforcing the underlying systems of cultural selection and valuation. AI models would embed the structural norms and biases that govern the intended data selection and organisation. The process of creating image datasets for Generative AI, often viewed as a neutral or purely technical process, encodes not only cultural narratives, but also biases and power dynamics. This paper explores the potential of these models as dynamic, generative systems that create outputs based on patterns, but also potentially embody and transmit cultural values. In doing so, the parameterisation of culture through dataset compilation emerges as a tangible phenomenon: as the algorithm seeks to capture shared patterns and structural rules among the samples, it parallels the concept of a “canon-as-model”. Presented at the 2024 Artistic Research symposium, this ongoing project foregrounds the interplay between programming practices and cultural encoding, where datasets act as reflections and agents of cultural normativity within AI-driven systems, with implications for how culture itself may be shaped in the age of AI.
Source: Parse Journal
Author: Paola Torres Nuñez del Prado
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4416290201