Encoding / Decoding
P. Bourdieu
Stuart Hall's model of cultural encoding and decoding explains why the same cultural texts are read differently depending on social position — and why AI models must be understood as specific decoding instances.
"There is no intelligible discourse without the operation of a code." Stuart Hall, Encoding/Decoding (1980), p. 131
The Encoding/Decoding Model
Hall argues against linear communication models: meanings are not contained in texts but produced through encoding and reproduced through decoding — or transformed. Three decoding positions are possible: dominant-hegemonic code, negotiated code, oppositional code.
AI as a Decoding Instance
Core Argument for HabitusMatch
The LLM model in HabitusMatch is not a neutral analysis machine — it decodes music lists from a specific cultural position determined by its training data. This position is primarily Anglophone, Western, urban, and academically educated. This is not a defect — it is an epistemological fact that must be made transparent.
Significance for Prompt Architecture
Ideological-Critical Perspective
Validation through the Survey
The discrepancy between AI profile (decoding by the model) and self-image (decoding by the user) is empirically measurable. The HabitusMatch survey captures exactly this difference systematically — as triangulation instrument between decoding positions.
To the Study
Your assessment of how well the AI profile matches is theoretically relevant data. Participate in the survey →
Hall's Model: Three Readings
Media content is ideologically encoded and can be decoded dominant-hegemonially, negotiated, or oppositionally. The audience's social position determines the reading.
Algorithms as New Encoders
Recommendation systems like Spotify or YouTube assume the encoding function and thereby reproduce existing genre boundaries and social hierarchies.
Cultural Bias in AI Systems
- To what extent does LLM-based cultural analysis (Claude) reproduce existing habitus classifications?
- Which reading does a user adopt when the AI classifies their profile as lower-middle-class?
- Empirically measurable symbolic violence through algorithmic classification
Airoldi, M. (2021). The techno-social reproduction of taste boundaries. Poetics, 101563.
Bonini, T., & Gandini, A. (2019). First Week Is Editorial. Social Media + Society, 5.
Hall, S. (1973). Encoding and Decoding in the television discourse.
Mehta, A. et al. (2025). Who Gets Heard? ArXiv abs/2511.05953.
Morris, J. (2015). Curation by code. European Journal of Cultural Studies, 18, 446–463.
Morreale, F. et al. (2025). Reductive, Exclusionary, Normalising. TISMIR, 8, 300–312.
Werner, A. (2020). Organizing music, organizing gender. Popular Communication, 18, 78–90.