According to a recent study published in Scientific Reports, ChatGPT could be capable of outperforming humans in a test of creativity. Source: Pexels.
According to a recent study published in Scientific Reports, artificial intelligence (AI) could be capable of outperforming humans in a test of creativity.
The study was conducted by researchers from the University of Arkansas, who pitted 151 human participants against ChatGPT-4 — a natural language processing (NLP) model developed by OpenAI — in three tests designed to measure divergent thinking, which is an indicator of a person’s creative thinking potential.
Divergent thinking is characterised by the ability to provide multiple creative answers to a problem that does not have one expected solution.
The tests gave three categories of questions; the Alternative Use Task, which asked participants to provide creative uses for everyday objects (such as a fork or a rope); the Consequence Task, which asked participants to imagine responses for hypothetical situations; and the Divergent Associations Task, which asks participants to name 10 nouns that are as semantically different from each other as possible.
Answers were evaluated across three dimensions: fluency (number of responses), originality (response originality), and elaboration (length of response).
The study reported that in comparison to human participants, “GPT-4 was more original and elaborate . . . on each of the divergent thinking tasks, even when controlling for fluency of responses”.
“In other words, GPT-4 demonstrated higher creative potential across an entire battery of divergent thinking tasks.”
The study notes that while AI measured for a greater creative potential, this is not a “guarantee for creative achievement”.
“To comprehensively examine creativity requires not only an assessment of originality, but also of the usefulness and appropriateness of an idea or product,” said the authors.
AI does not have “agency” and is still very “dependent on the assistance of human user to elicit responses”. The study adds that human participants may have also felt constrained to ground their answers in the real world.
The study concludes that large language models are promptly outperforming human creativity in ways they have never seen before, and their extent of threatening human creativity remains to be seen.
