Popularity Feedback Constrains Innovation in Cultural Markets

TL;DR

This study investigates how popularity feedback influences cultural evolution by running a large-scale online experiment where participants both select and modify images in evolving markets. By comparing conditions with and without popularity information, the authors show that access to popularity accelerates cumulative advantage in selection, reduces diversity, and biases creation toward expanding existing features rather than disrupting them. These dynamics slow early innovation and alter the trajectory of aesthetic and creative improvements, though long-run effects may reverse as high-quality ideas persist. The findings highlight how producer–consumer feedback loops can shape not only market states but also the form and direction of cultural innovation, with implications for understanding creative processes in real-world cultural systems.

Abstract

Real-world creative processes ranging from art to science rely on social feedback-loops between selection and creation. Yet, the effects of popularity feedback on collective creativity remain poorly understood. We investigate how popularity ratings influence cultural dynamics in a large-scale online experiment where participants ($N = 1\,008$) iteratively \textit{select} images from evolving markets and \textit{produce} their own modifications. Results show that exposing the popularity of images reduces cultural diversity and slows innovation, delaying aesthetic improvements. These findings are mediated by alterations of both selection and creation. During selection, popularity information triggers cumulative advantage, with participants preferentially building upon popular images, reducing diversity. During creation, participants make less disruptive changes, and are more likely to expand existing visual patterns. Feedback loops in cultural markets thus not only shape selection, but also, directly or indirectly, the form and direction of cultural innovation.

Figures (10)

• Figure 1: Selection and creation steps. Participants observe a market of 12 images, choose one image, change $\geq 1$ and $\leq 24$ pixels, and add their own creation to the market.
• Figure 2: Evolution of two chains starting from the same image, over the first 35 generations, in conditions with popularity information (PI, to the left) and without (NPI, to the right). Markets include the 12 last images created in a chain before the current generation (each column in the tables shows the contents of a market at a given generation). This view emphasizes the perspective of the participants, who choose among images in the market. Phylogenetic trees reflect the ancestry relationships between images in a chain, emphasizing the propagation and decay of ideas. The semantic space (recovered via Vision-Transformer embeddings of the images blip, and reduced to two dimensions via UMAP for the sake of the illustration) provides another perspective on the exploration process in each chain, by revealing cultural exploration dynamics. Contours indicate the density in semantic of all images from a given condition (thus representing the landscape of their respective cultural output).
• Figure 3: Average market diversity in the conditions with popularity information (PI) and without (NPI). Diversity is measured as the average distance between images in a market at any generation. Error bars indicate standard errors ($\pm 1$SE). Statistical significance is evaluated via a non-parametric permutation test pairing chains with identical initial images.
• Figure 4: Average chain autocorrelation in semantic space $\bar{\rho}(\tau)$ under the conditions with popularity information (PI) and without (NPI) and for different time-lags $\tau$.
• Figure 5: Quality of images throughout generations and across conditions. Error bars indicate standard errors ($\pm 1$SE).