Why Everything Is Starting to Look the Same

The essay argues that personalization can make the world more uniform rather than more diverse. Recommenders predict what people will choose, promote those predictions, then learn from the choices they helped produce. Because common prediction losses favor the average, uncertain systems pull toward popular options; repeated feedback makes those options still more dominant while unusual choices become harder to discover. The author connects this process to model-driven market behavior, urban “placelessness,” and cultural sameness, arguing that diversity is often the variation an optimizer is designed to minimize.

Evidence comes from several domains, with different results depending on what is being optimized. Across 640 London high streets and 18,000 food businesses, streets with fewer chains are more distinctive, while even a modest presence of repeated chains makes them more alike. Billboard hits have grown less varied within individual songs - using fewer unique words and repeating lines more - but the songs’ vocabulary has diversified across the chart, showing that flattening depends on the dimension targeted. Film ratings reveal concentrated attention: roughly 1% of titles attract nearly 40% of activity. Experiments with synthetic survey respondents similarly found that elaborate models could match averages while losing meaningful variation and social structure. The central concern is that apparent abundance and accurate averages can conceal narrowing choices and populations that represent no one in particular.