Pluralistic: Model collapse (12 Aug 2026)

Data-driven prediction systems, from AI to recommendation algorithms, drive “model collapse” and pervasive homogeneity across culture, urban design, and markets because their predictions become self-fulfilling prophecies, pushing everything towards a bland, collective mean.
This phenomenon is captured by concepts like “performativity” (economic models shaping markets), “model collapse” (ML models trained on themselves becoming bland), and “placelessness” (urban environments optimizing to a single template), exemplified by biased predictive policing, identical chain-filled high streets, and the decline of unique pubs.
The underlying issue is the “exploration vs. exploitation” tradeoff, where algorithms prioritize safe, exploitative recommendations, narrowing choices and obscuring true preferences. This creates systemic brittleness, as diverse options “decay out of consideration” and are never tested for desirability.
Counteracting this requires “forcing exploration back in” through regulatory interventions (e.g., mandating open or non-personalized recommendation systems) and promoting interoperability, as individual “adventurous consumption” is insufficient to overcome system-wide algorithmic homogenization.