In three divergent thinking tests, ChatGPT-4 was compared against 151 human participants. The results showed that the AI was more creative. The tests, which evaluated the capacity to come up with original replies, revealed that GPT-4 was able to provide more creative and detailed responses.
While acknowledging the limitations of AI’s agency and the difficulties in evaluating creativity, the study highlights the growing potential of AI in creative fields. Even though AI has the potential to be a tool for boosting human creativity, concerns concerning its function and possible integration with creative processes still exist.
Important Details:
The Creative Edge of AI
In activities involving divergent thinking, ChatGPT-4 fared better than human participants, exhibiting superior originality and elaboration in responses.
Notes on the Study:
Researchers point out that despite AI’s remarkable abilities, it lacks agency and that human contact is necessary to unleash its creative potential.
AI’s Place in Creative Futures:
The results imply that AI might be a motivating instrument that supports human creativity and helps it overcome conceptual rigidity, but it’s still unclear how much AI can actually replace human innovation.
The ability to come up with a novel answer to a question without a predetermined one, like “What is the best way to avoid talking about politics with my parents?” is a hallmark of divergent thinking. GPT-4 gave more thoughtful and unique responses to the study than did the human participants.
The authors of the study, “The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks,” are Kim N. Awa and Kent F. Hubert, Ph.D. candidates in psychological science at the University of Arizona, and Darya L. Zabelina, an assistant professor of psychology science and the head of the Mechanisms of Creative Cognition and Attention Lab. The study was published in Scientific Reports.
Three tests were used: the Consequences Task, which asks participants to imagine possible outcomes of hypothetical situations, such as “what if humans no longer needed sleep?”; the Divergent Associations Task, which asks participants to generate 10 nouns that are as semantically distant as possible; and the Alternative Use Task, which asks participants to come up with creative uses for everyday objects like a rope or a fork. For example, there is very little semantic difference between the words “dog” and “cat,” but a lot between words like “cat” and “ontology.”
Response length, quantity, and semantic variation across words were all taken into consideration while evaluating the answers. In the end, the researchers concluded that even after adjusting for answer fluency, GPT-4 outperformed humans overall on all of the divergent thinking tasks in terms of originality and elaborateness. Stated differently, GPT-4 showed greater creative potential on a battery of activities involving divergent thinking.
There are certain limitations to this discovery. As the authors put it, “It is important to note that the measures used in this study are all measures of creative potential, but another aspect of measuring a person’s creativity is their involvement in creative activities or achievements.”
The aim of the research was to investigate creative potential at the human level, rather than focusing on individuals who may possess formal creative qualifications.
Hubert and Awa go on to say that AI is “dependent on the assistance of a human user” and “does not have agency, unlike humans.” As a result, until something prompts it, AI’s creative potential remains untapped.
Furthermore, the suitability of the GPT-4 replies was not assessed by the researchers. As a result, even if the AI might have responded more often and creatively, the human participants might have felt limited by the requirement that their answers be based on reality.
While acknowledging that there may not have been much human reason to compose lengthy responses, Awa added that there are still unanswered problems regarding “how do you operationalize creativity? Is it truly true that the results of these tests on humans may be applied to other individuals? Is a wide range of innovative ideas being evaluated? Therefore, I believe it requires us to critically evaluate the most widely used metrics for divergent thinking.
It’s not really important whether the tests accurately reflect the creative capacity of people. The key point is that big language models are developing at a breakneck pace, surpassing human performance in previously unheard-of ways. It remains to be seen if they pose a threat to human ingenuity.
“Moving forward, future possibilities of AI acting as a tool of inspiration, as an aid in a person’s creative process or to overcome fixedness is promising,” the authors continue to believe for the time being.