Home > News > Techscience

ChatGPT-4 Excels in Creativity Tests

Liu Xia Thu, Mar 07 2024 12:20 AM EST

Beijing, March 3rd (Xinhua) - Scientists at the University of Arkansas have conducted a recent study involving 151 human participants pitted against the artificial intelligence (AI) tool ChatGPT-4 in three tests to evaluate their divergent thinking. The results indicate that AI outperformed humans. The relevant paper was published in the latest issue of the journal "Scientific Reports."

Divergent thinking serves as a measure of creative thinking, characterized by the ability to generate unique solutions to problems without predetermined answers. In the recent study, GPT-4 provided more innovative and refined answers than human participants.

The three tests used in the study were "Alternative Uses Task," "Consequences Task," and "Divergent Association Task." The "Alternative Uses Task" required participants to propose creative uses for everyday items like ropes or forks; the "Consequences Task" asked participants to imagine potential outcomes of hypothetical scenarios, such as what would happen if humans no longer needed sleep; and the "Divergent Association Task" tasked participants with providing 10 nouns semantically distant from each other. The semantic distance between words like "dog" and "cat" is small, whereas the distance between "cat" and words like "ontology" is significant.

The research team evaluated the quantity, length, and semantic differences between words in the responses from both humans and the AI tool. The results showed that in each divergent thinking task, GPT-4's answers were both more creative and precise than those provided by humans. In other words, across the board of divergent thinking tasks, GPT-4 demonstrated higher levels of creativity.

One of the lead researchers, Kent Hubert, a psychology doctoral student at the University of Arkansas, noted some caveats to the study. For instance, the metrics used in the study measure creative potential, but engaging in creative activities or achieving accomplishments is another aspect of measuring creative ability. Additionally, AI relies on human users' assistance, and its creativity remains stagnant without prompts.

However, the research team emphasizes that large language models are rapidly evolving and surpassing humans in unprecedented ways. Whether they can replace human creativity remains to be seen. Nonetheless, the prospect of AI serving as an inspirational tool to aid human creative endeavors is on the horizon.