Why creative crowds converge toward familiar styles

Creativity in groups is hard to study, but online art platforms offer a natural laboratory. Researchers analyzed 13 months of Artbreeder remix parties—where users iteratively modify images seeded from a single source—and found that images systematically simplify and converge toward thematic "attractors" like steampunk or alien architecture. Paradoxically, novelty works: more original parent images produce more novel children that earn more likes. Yet users preferentially build on simpler, less novel work, creating a tension between individual taste and collective drift. Larger remix parties generate more novelty overall but lower complexity. The findings parallel iterated learning experiments showing how cultural transmission distorts artifacts toward learners' biases, but extend that framework to networked human-AI systems.