AI模型设计创意远不及人类设计师 新基准测试揭示差距

研究人员推出了设计创意基准（Design Creativity Bench），用于系统评估前沿AI模型在UI设计领域的创意表现1。该测试涵盖16个前沿AI模型在80个开放式UI任务上的能力1，通过原创性、创意范围和适当性三个维度来衡量AI的设计创意水平1。

评估结果显示，表现最佳的模型OpenAI GPT-6 Astra得分为65.5分，与人类设计师参考得分92.3分相比存在显著差距1。这一结果表明当前AI模型在创意设计任务上仍有相当大的提升空间。研究团队已发布论文、HuggingFace数据集和Github代码库供社区使用1。

Researchers have introduced the Design Creativity Bench, a new evaluation framework for assessing the creative capabilities of leading artificial intelligence models on user interface design tasks 1. The benchmark evaluated 16 state-of-the-art AI models across 80 open-ended UI design challenges, measuring creative performance across three dimensions: originality, creative range, and appropriateness 1.

The results reveal a substantial gap between AI and human designers 1. OpenAI's GPT-6 Astra, the top-performing model in the assessment, achieved a score of 65.5, significantly below the human designer reference score of 92.3 1. This score differential underscores the limitations AI systems currently face when tasked with creative design work 1.

The researchers have made their work publicly available, releasing the accompanying research paper, a HuggingFace dataset, and source code on Github 1.