Generative artificial intelligence models outperform students on divergent and convergent thinking assessments

Area of Science:

• Cognitive Psychology
• Artificial Intelligence
• Creativity Studies

Background:

• Generative artificial intelligence (GenAI) is increasingly used in creative fields, prompting scientific inquiry into its creative capabilities.
• Previous research has evaluated GenAI on creativity, but lacked direct comparisons with humans on both divergent and convergent thinking.
• The study addresses the gap in understanding GenAI's creative performance relative to human benchmarks.

Purpose of the Study:

• To compare the creative abilities of humans and state-of-the-art GenAI models.
• To assess performance on both divergent and convergent thinking tasks.
• To evaluate the effectiveness of current creativity assessment methods for GenAI.

Main Methods:

• Human participants (n=46) were compared against three advanced GenAI chatbots: ChatGPT-4o, DeepSeek-V3, and Gemini 2.0.
• The Alternate Uses Task (AUT) was used to measure divergent thinking (originality of 'average' and 'best' ideas).
• The Remote Associates Test (RAT) was employed to assess convergent thinking (performance on 57 items).

Main Results:

• All GenAI models significantly outperformed human participants on both divergent and convergent thinking tasks.
• GenAI-generated ideas demonstrated higher originality ('average' and 'best') than human ideas on the AUT.
• GenAI models showed superior performance on the RAT compared to humans, with ChatGPT-4o achieving the highest scores among AI models.

Conclusions:

• Current state-of-the-art GenAI models exhibit remarkable creative potential, surpassing human performance in key cognitive tasks.
• The findings suggest that existing creativity assessment methodologies may need re-evaluation for their suitability in studying AI creativity.
• Further research is needed to refine and develop appropriate frameworks for evaluating the creativity of artificial intelligence.