Artificial intelligence (AI) could be capable of outperforming human creativity, according to a new study published in Scientific Reports last month.
The study, conducted by researchers from the University of Arkansas, pitted ChatGPT-4, the natural language processing model (NLP) by OpenAI, against 151 human participants in three tests designed to measure divergent thinking, or the ability to provide unique answers to questions that do not have one expected solution.
The questions were under three categories: “Alternative Use Task,” which prompted both human respondents and AI to provide alternative uses for objects such as a fork or a rope; “Consequences Task,” which asked for responses of hypothetical situations; and “Divergent Association Tasks,” which requested 10 nouns that are as semantically distant as possible, such as “cat” and “ontology.”
“Overall, GPT-4 was more original and elaborate than humans on each of the divergent thinking tasks, even when controlling for fluency of responses,” reads a quote attributed to the study’s authors in a media release.
“In other words, GPT-4 demonstrated higher creative potential across an entire battery of divergent thinking tasks,” the researchers stated in the release.
The study points out that a potential reason for major discrepancies between human and GPT-4 responses was the human tendency to ground responses in the real world. In other words, humans are more attuned to conventionality, which makes for more conventional responses, the researchers explained.
The study notes that AI still depends on human input to generate any content. As the authors explained in the release, “the creative potential of AI is in a constant state of stagnation unless prompted.”
According to the researchers, large language models now have the ability to outperform humans in ways they could not before, but the extent of their competitiveness with human creativity remains to be seen.
“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 stated.
