A recent study published in Nature Human Behavior reveals a significant limitation in AI-generated creativity, particularly with ChatGPT: while it can produce a large volume of creative ideas, these ideas tend to be very similar to each other, limiting diversity in brainstorming sessions.
Key Findings:
-
In an experiment where participants brainstormed toy ideas involving a brick and a fan, 94% of the ideas generated by ChatGPT overlapped in concept, with nine participants independently naming their toy "Build-a-Breeze Castle".
-
Participants who relied on their own ideas combined with web searches produced the most unique and diverse concepts, outperforming ChatGPT in idea variety.
-
Although ChatGPT-4 generated more diverse ideas than ChatGPT 3.5, it still lagged significantly behind human creativity in terms of diversity.
-
This pattern of reduced diversity with AI assistance was consistent across five experiments and aligns with findings from other studies in creative writing, where AI-generated stories were more similar to each other than human-written ones.
Implications:
The researchers, including Wharton professors Gideon Nave and Christian Terwiesch, emphasize that the value of brainstorming lies in generating a broad ecosystem of distinct ideas, not multiple variations of the same concept. They clarify that the diversity discussed is about the novelty and difference in ideas, not demographic diversity.
Using ChatGPT Effectively:
Despite these limitations, ChatGPT can still be a valuable tool in the creative process if used thoughtfully. Terwiesch suggests combining AI-generated ideas with original human input and sourcing ideas from others to maximize diversity. He also recommends "chain of thought prompting," which involves asking the AI to generate multiple ideas that are explicitly different from each other.
OpenAI supports best practices for prompting ChatGPT to enhance creativity and provides guidance for writers and students on how to use the tool effectively.
Broader Context:
Other research shows mixed results on AI's impact on creativity. Some studies find that AI can boost creativity in everyday problem-solving tasks and produce ideas rated as more creative than human-generated ones. However, concerns remain about AI's effect on originality and creative writing skills, especially in educational settings.
Conclusion:
While AI tools like ChatGPT are advancing and can surpass humans in generating creative ideas individually, their outputs tend to be narrower and less diverse than ideas generated collectively by humans. This suggests that AI should be integrated as a complement to human creativity rather than a replacement, especially in brainstorming contexts where idea diversity is crucial.
Teaching future generations to use generative AI as a creativity booster, rather than a shortcut, could help balance AI literacy with concerns about overreliance or misuse.