Dynamics of Collective Creativity | Cornell Computational Cognition Lab

Collective creativity is increasingly shaped by interactions between people and generative AI systems. This project studies how creative artifacts evolve when they are produced, selected, modified, and transmitted through social networks composed of humans, AI agents, or hybrid human-AI groups.

Using a creative writing paradigm, participants join networked chains in which short stories are repeatedly selected and transformed over many iterations. This allows us to measure not only the quality of creative products, but also the dynamics by which creativity, diversity, and semantic structure change over time.

Read the paper PDF

Experimental framework for studying collective creativity. (A) Participants join social networks and engage in a creative writing task where short stories are selected, modified, and transmitted over many iterations. (B) The study compares human-only, AI-only, and human-AI network configurations. (C) Story creativity is evaluated by a separate group of human raters.

The experiments compare three kinds of creative systems: fully human networks, fully AI networks, and mixed human-AI networks. This design lets us ask whether AI amplifies creative quality, whether it narrows or broadens the diversity of cultural output, and how hybrid systems differ from either humans or AI alone.

Creativity and diversity dynamics across human-only, GPT-only, and human-AI conditions — Dynamics of collective creativity. (A) Mean creativity ratings of stories over time, as evaluated by human participants. (B) Story diversity, measured as inverse similarity, over time. The 25 iterations are grouped into five sets of five iterations. Error bands show variability across participants. (C-D) Creativity and diversity gains quantify improvement from the first to the last iteration.

The results show that creative systems differ not only in final quality but also in their trajectories. GPT-only networks can rapidly increase rated creativity, while human-AI networks preserve and amplify diversity in ways that differ from both human-only and AI-only systems. These dynamics suggest that hybrid creative societies may support distinct forms of exploration, recombination, and cultural differentiation.

UMAP projection of story embeddings and semantic clusters — Semantic structure of creative output. A UMAP projection of a shared semantic embedding space highlights clusters of story content, with word clouds showing characteristic terms associated with different regions of the creative landscape.

What we do

Build controlled creative transmission experiments in human-only, AI-only, and human-AI social networks.
Measure how creativity and diversity change across repeated selection and modification.
Analyze semantic trajectories using embeddings, clustering, and term-level dynamics.
Identify when AI systems amplify creativity, when they homogenize output, and when hybrid groups preserve richer diversity.

Term dynamics by condition across story iterations — Term dynamics by condition. Words are plotted along the horizontal axis and generations along the vertical axis. Each circle indicates that a word appeared in a story at a given iteration; line segments indicate use across successive iterations; circle size reflects word frequency within that iteration.

Why it matters

Generative AI is becoming part of everyday creative work, but its effects on collective creativity depend on the structure of interaction. This project treats human-AI creativity as a cultural-evolution process, showing how network composition and transmission dynamics shape both the quality and diversity of creative output.

(Shiiku et al., 2025)

What we do

Why it matters

Related Publications

2025