| Abstract [eng] |
While several recent studies have demonstrated the impressive creative performance of Artificial Intelligence (AI) on divergent thinking tasks, there is currently insufficient research and understanding of how AI performs on convergent thinking and discernment activities, essential components of the human creative process. Creative problem-solving methods, in particular, present an intriguing means of researching AI's capacity to identify problems within ambiguous situations, while also generating relevant, useful solutions that respond to authentic, real-world issues in a meaningful way. This paper examined the performance of three separate AI entries generated by GPT-4o as compared with 68 human team entries from students in grades 7–9 within the competitive World Solutions Challenge offered by the Future Problem Solving Program International (FPSPI). The three entries were blind-scored by trained human evaluators for measures of effectiveness, impact, humaneness, creative strength, and development of action plan. Though all three AI entries scored in the top 15 % for all measures, including achieving the top scores for effectiveness, impact, humaneness, development of action plan, and overall performance, the AI entries were not found to be significantly different than the student control group on the measure of creative strength. The results suggest that AI models like GPT-4 may approach human-like abilities on certain aspects of creative performance during a more comprehensive creative process than divergent thinking alone, providing new insight into the current creative strengths—and limitations—of existing AI models. |
