Does AI homogenize culture the way mass media did?

Adorno's analysis of the culture industry in mid-century mass entertainment turned on the suppression of novelty under the appearance of variety. Pop songs, films, and magazines were not literally identical, but their structural sameness was concealed by surface variation. The audience experienced choice while consuming products that reproduced the same patterns. The function was social: maintaining a passive audience by feeding it familiar forms in slightly varied dress.

AI updates this analysis at a different productive level. Where the culture industry mass-produced identical commodities, AI mass-generates similar flows. The output is contextual rather than pre-stamped — each generation tailored to the specific prompt, audience, and use — so the surface appearance is even more individualized than the culture industry could achieve. But the underlying form is reliably homogeneous: the same rhetorical patterns, the same hedging structures, the same conclusion shapes, the same evidence-marshaling moves, all reproducing the center of the training distribution.

The artificial-hivemind effect is the mechanical evidence of this homogeneity (different LLMs independently converge on similar outputs in open-ended generation — the artificial hivemind effect). Models trained on overlapping corpora with similar RLHF regimes produce strikingly convergent output even when nominally competing. The surface is varied; the underlying form is the same. This is the culture-industry pattern precisely.

What the AI update does that the culture industry could not is suppress novelty more deeply, because contextual generation creates a stronger illusion of customization than pre-stamped variety could. The user consuming AI output is not aware of consuming a homogeneous product — the output is uniquely hers, generated for her prompt, addressed to her context. The homogeneity is invisible at the individual level and only becomes visible across users. This makes the suppression effect harder to perceive and therefore harder to resist.

The diagnostic implication: the cultural-criticism toolkit Adorno developed for mass entertainment transposes to AI with one update. Critique of the culture industry asked: what does this product suppress? What genuine cultural form does its homogeneity displace? Critique of AI asks the same questions of generated content. Does AI repeat the Enlightenment's reversal into its opposite? is the broader frame; this is the cultural-economic specification.

The strongest counterargument: AI output can be genuinely novel when prompted for novelty. Possible at the margin, but the centroid of AI output is the centroid of the training distribution, which is by definition where novelty is least.

Source: Tokenization of Intelligence - Dialectic of Enlightenment