OpenAI's GPT-4 Outperforms Human Originality in Renowned Creativity Test

Study conducted by university researchers Dr. Erik Guzik, Dr. Christian Byrge, and Dr. Christian Gilde test GPT-4's creative abilities

ASHLAND, Ore., May 18, 2023 /PRNewswire/ -- In a new announcement at the University of Southern Oregon's Conference on Creativity, OpenAI's large language model, GPT-4, scored in the top 1% on the Torrance Tests of Creative Thinking (TTCT), outperforming humans in the areas of fluency, flexibility, and originality.

In a compelling study conducted by researchers Erik Guzik from the University of Montana College of Business, Christian Byrge from Vilnius University Business School, and Christian Gilde from the University of Montana Western, GPT-4 was tested eight times through ChatGPT using two variants of the TTCT, the most referenced and most widely used scientific test for creativity, and the gold standard for assessing creativity since the 1970s.

GPT-4 demonstrated its creative prowess by asking curious questions, guessing causes, contemplating potential consequences, generating ideas for product improvement, envisioning new alternative uses for existing products, and imagining future scenarios.

When the AI's responses were compared to those of humans, the results were groundbreaking. All eight tests saw GPT-4 score in the top 1% of all human responses for fluency — the ability to produce a large number of responses to each activity. For flexibility, the ability to produce a variety of ideas and to shift from one approach to another, three tests scored in the top 1%, two in the top 2%, and the remaining three in the top 7%. Most impressively, all eight tests saw GPT-4 score in the top 1% for originality, the ability to produce uncommon or unique responses that defy the commonplace and established norms.

The research team commented on the findings, saying, "These results are more than just impressive. They suggest that we need to reconsider how AI creativity can assist humans in solving complex and critical problems, especially during idea generation, evaluation, elaboration, and testing."

The research team believes these findings could prompt a reevaluation of how we define, assess, and develop creativity in education, organizations, and society. With AI models like GPT-4 demonstrating such high levels of creativity, the potential for partnership between humans and AI in tackling complex challenges is greater than ever.

The team's findings, including further research into training GPT-4 to improve its creative abilities in specific domains such as business innovation and entrepreneurship, is planned for publication this summer.

Contact:
Dr. Erik Guzik
406-243-6053
[email protected]

SOURCE Dr. Erik Guzik