A recent study found that AI language models, specifically OpenAI’s GPT-4, outperform humans on tasks that require divergent thinking – which involves generating unique solutions to open-ended questions, an important facet of creativity.
The study, conducted by Kent F. Hubert and Kim N. Awa, Ph.D. The study by students at the University of Arkansas and Darya L. Zabelina, an assistant professor at the same institution, involved 151 human participants. They were tested against the AI model on the Alternative Use, Consequences and Divergent Associations task. And the bad news for humans was that the AI model showed greater originality and detail in its responses, indicating higher creative potential.
Of course, these findings are not definitive proof of AI’s superior creativity. The study authors caution that while the AI models were more original, they were not necessarily more appropriate or practical in their ideas. AI’s creative potential also depends on human input, which limits its autonomy.
More research is needed
The study also found that AI used a higher frequency of repeated words compared to human respondents. Although people generated a wider range of responses, this did not necessarily result in greater originality.
The findings challenge the assumption that creativity is a uniquely human trait. The question remains, however, whether AI’s superior performance in creative tasks poses a threat to humans, now or in the future. While the results were undoubtedly impressive, the authors emphasize that the study only examines one aspect of divergent thinking. It doesn’t necessarily mean AI is more creative across the board.
The authors conclude that future research will need to take into account the usefulness and suitability of the ideas, as well as the applications of AI creativity in the real world.
The study, titled “The current state of generative language models for artificial intelligence is more creative than humans on diverse thinking tasks,” was published in Scientific reports.