Human Collaboration Outperforms AI in Creative Brainstorming Study

Cognitive scientists Clin Lai and Roger Beaty conducted an experiment using the alternate uses task to compare the creative output of humans collaborating with other humans versus those working with generative AI. The study found that while participants using ChatGPT produced longer responses, human pairs generated ideas that were more creative, distinct, and diverse.

The researchers determined that AI outputs tend toward homogeneity because large language models predict the most likely next words based on similar training sets. This process leads to a convergence of writing styles and ideas across different users. In contrast, the creative benefits of human collaboration increased over time, whereas AI-assisted brainstorming appeared to narrow the range of explored ideas.

To mitigate this, the researchers tested persona prompting by assigning specific personalities to the AI. While this technique increased the diversity of ChatGPT's suggestions, the results still did not reach the level of human-led creativity. The authors concluded that AI is trained to generate average responses and cannot replicate the originality provided by idiosyncratic human experiences.