Creative tendencies in artificial intelligence could surpass that of humanity

In a recent study outlined in ScienceDaily.com, titled “The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks,” published in Scientific Reports, researchers explored the creative potential of ChatGPT-4 in comparison to 151 human participants.

The study aimed to measure divergent thinking, a key aspect of creative thought. Three tests were employed to evaluate creative potential: the Alternative Use Task, which asked participants to devise creative uses for everyday items like a rope or fork; the Consequences Task, which prompted participants to envision possible outcomes of hypothetical situations (e.g., “what if humans no longer needed sleep?”); and the Divergent Associations Task, requiring participants to generate 10 nouns as semantically distant as possible. ChatGPT-4 consistently outperformed human participants across all divergent thinking tasks, providing more original and elaborate responses.

The study, conducted by Ph.D. students Kent F. Hubert and Kim N. Awa, along with assistant professor Darya L. Zabelina from the U of A, measured responses based on quantity, length, and semantic differences between words. Despite these findings, the study highlighted certain limitations. The measures used focused on creative potential, and the study did not assess the appropriateness of ChatGPT-4’s responses. The researchers emphasized that creative achievements or involvement in creative activities are separate dimensions of creativity that were not addressed in this study. Moreover, the study acknowledged that AI lacks agency and depends on human prompts, leading to a constant state of stagnation in its creative potential unless prompted.

The authors also pointed out that the motivation of human participants to provide elaborate answers might not have been high, raising questions about the operationalization of creativity in tests. While the study suggests that large language models like ChatGPT-4 are rapidly advancing and outperforming humans in certain aspects of creativity, it highlights ongoing uncertainties about the appropriateness of responses and the broader implications for human creative potential.

The authors envision future possibilities where AI serves as a tool of inspiration or aids in the creative process but emphasize the need for critical examination and further research on measuring divergent thinking in different individuals.