Vinchon Florent, Gironnay Valentin, Lubart Todd
Laboratoire de Psychologie et d'Ergonomie Appliquée (LaPEA), Université Paris Cité and Univ Gustave Eiffel, F-92100 Boulogne-Billancourt, France.
J Intell. 2024 Dec 6;12(12):125. doi: 10.3390/jintelligence12120125.
This study examined generative artificial intelligences (GenAIs), as popularized by ChatGPT, in standardized creativity tests. Benchmarking GenAI against human performance, the results showed that ChatGPT demonstrated remarkable fluency in content generation, though the creative output was average. The random nature of AI creativity and the dependency on the richness of the training database require a reassessment of traditional creativity metrics, especially for AI. Our findings highlight the integral role humans play in guiding AI to foster genuine originality, suggesting the need for future research in human-AI co-creation and the development of robust AI creativity measurement mechanisms.
本研究在标准化创造力测试中考察了由ChatGPT推广的生成式人工智能(GenAIs)。将GenAI与人类表现进行基准对比,结果显示ChatGPT在内容生成方面表现出显著的流畅性,不过其创造性产出处于平均水平。人工智能创造力的随机性以及对训练数据库丰富程度的依赖,需要对传统的创造力指标进行重新评估,尤其是针对人工智能的指标。我们的研究结果凸显了人类在引导人工智能培养真正原创性方面所发挥的不可或缺的作用,这表明未来需要开展关于人类与人工智能共同创作以及开发强大的人工智能创造力测量机制的研究。