Zhou Eric B, Lee Dokyun, Gu Bin
Questrom School of Business, Boston University, Boston, MA 02215, USA.
Computing & Data Sciences, Boston University, Boston, MA 02215, USA.
Sci Adv. 2025 Sep 5;11(36):eadu5800. doi: 10.1126/sciadv.adu5800. Epub 2025 Sep 3.
Artists are rapidly integrating generative text-to-image models into their workflows, yet how this affects creative discovery remains unclear. Leveraging large-scale data from an online art platform, we compare artificial intelligence (AI)-assisted creators to matched nonadopters to assess novel idea contributions. Initially, a concentrated subset of AI-assisted creators contributes more novel artifacts in absolute terms through increased output-the productivity effect-although the average rate of contributing novel artifacts decreases because of a dilution effect. This reflects a shift toward high-volume, incremental exploration, ultimately yielding a greater aggregate of novel artifacts by AI-assisted creators. We observe no evidence of a human-AI effect above and beyond the productivity effect. The release of open-source Stable Diffusion accelerates novel contributions across a more diverse group, suggesting that text-to-image tools facilitate exploration at scale, initially enabling persistent breakthroughs by select "masterminds," driven by increased volume, and subsequently enabling widespread novel contributions from a "hive mind."
艺术家们正在迅速将生成式文本到图像模型整合到他们的工作流程中，但这对创造性发现的影响仍不明确。利用一个在线艺术平台的大规模数据，我们将人工智能（AI）辅助创作者与匹配的未采用者进行比较，以评估新颖想法的贡献。最初，一部分集中的AI辅助创作者通过增加产出——即生产力效应——在绝对数量上贡献了更多新颖的作品，尽管由于稀释效应，贡献新颖作品的平均速率有所下降。这反映了向大量、渐进式探索的转变，最终AI辅助创作者产生了更多新颖作品的总量。我们没有观察到超出生产力效应的人机效应的证据。开源的Stable Diffusion的发布加速了更多样化群体的新颖贡献，这表明文本到图像工具促进了大规模的探索，最初由数量增加驱动，使选定的“策划者”实现持续突破，随后使“群体思维”产生广泛的新颖贡献。
