A recent study published in the journal Scientific Reports suggests that large language model (LLM) artificial intelligence chatbots may outperform ordinary humans in creative tasks, such as brainstorming alternative uses for common items—a reflection of divergent thinking. However, the highest-scoring individuals on these tasks still exceeded the results of the best-performing chatbots.
Divergent thinking is a thinking process often associated with creativity that emphasizes generating many different ideas or solutions to a specific task.
It is typically assessed through the Alternative Use Task (AUT), in which participants are asked to come up with as many alternative uses for everyday objects as possible in a short period of time. Responses were divided into four different categories: fluency, flexibility, originality, and elaboration.
Mika Koivisto and Simone Grassini compared the responses of 256 human participants with those of three AI chatbots (ChatGPT3, ChatGPT4, and Copy.Ai) and their AUT responses to four objects: a rope, a box, a pencil, and a candle. The authors assessed the originality of responses by rating semantic distance (how closely the response relates to the original use of the object) and creativity.
Semantic distance is quantified using computational methods on a scale of 0 to 2, while human evaluators, blind to the initiator of the response, subjectively rate creativity on a scale of 1 to 5. On average, chatbots generated responses that scored significantly higher than human responses for semantic distance (0.95 vs. 0.91) and creativity (2.91 vs. 2.47).
Human responses had a wider range on both metrics—the lowest scores were much lower than the AI responses, but the highest scores were generally higher. The human's best response outperformed each chatbot's best response in seven of the eight rating categories.
These findings suggest that AI chatbots can now generate ideas at least as well as human beings. However, the authors note that they only considered performance on a single task relevant to the assessment of creativity. The authors suggest that future research could explore how to integrate artificial intelligence into the creative process to improve human performance.