Sign in Subscribe
Artificial intelligence

Generative Artificial Intelligence Outperforms Humans in Creativity Tasks

Strategic Insights

— 1 min read

Researchers at McMaster University have found that three state-of-the-art generative artificial intelligence (GenAI) models, ChatGPT-4o, DeepSeek-V3, and Gemini 2.0, outperform human participants in both divergent and convergent thinking assessments. The study compared the creative ability of human participants (n = 46) against the GenAI models using the Alternate Uses Task (AUT) and the Remote Associates Test (RAT). The results demonstrate the immense creative potential of GenAI, but also raise questions about the appropriateness of current creativity assessment methods.

Key Takeaways:

  • The study compared the creative ability of 46 human participants against three GenAI models, ChatGPT-4o, DeepSeek-V3, and Gemini 2.0, using the Alternate Uses Task (AUT) and the Remote Associates Test (RAT).
  • All GenAI models outperformed human participants in both divergent and convergent thinking tasks.
  • ChatGPT-4o consistently demonstrated the best scores on both tasks, outperforming human-generated ideas in terms of originality and convergence.
  • The findings highlight the immense creative potential of GenAI, but also raise questions about the validity of current creativity assessment methods.
  • The study suggests that GenAI models may be better suited for certain creative tasks, but also raises concerns about the potential for AI-generated content to be misleading or biased.

Statistics:

  • 46 human participants were compared against three GenAI models in the study.
  • The Alternate Uses Task (AUT) and the Remote Associates Test (RAT) were used to assess creativity in divergent and convergent thinking tasks, respectively.
  • The median and maximum originality scores on the AUT were compared to determine the level of originality of the 'average' and 'best' idea generated by humans and GenAI.
  • 57 RAT items were used to assess performance on convergent thinking tasks.
  • The study found that all GenAI models outperformed human participants in both divergent and convergent thinking tasks.

Sources:

  • Generative artificial intelligence models outperform students on divergent and convergent thinking assessments. Scientific Reports, 2025;15(1):36987.
  • NewsRx. Data on Artificial Intelligence Discussed by Researchers at McMaster University (Generative artificial intelligence models outperform students on divergent and convergent thinking assessments). Journal of Engineering. November 3, 2025; p 240.

Read more