[论文解读] How Generative AI supports human in conceptual design

L. Chen, Yaxuan Song+4 more|ArXiv.org|Feb 1, 2025

Design Education and Practice被引用 3

一句话总结

该研究在四个概念设计阶段实证研究文本到文本（GPT-3.5）和文本到图像（Midjourney）生成式 AI 如何帮助初学设计师，发现 AI 主要帮助问题定义与创意生成，而评估仍由人类主导。

ABSTRACT

Generative Artificial Intelligence (Generative AI) is a collection of AI technologies that can generate new information such as texts and images. With its strong capabilities, Generative AI has been actively studied in creative design processes. However, limited studies have explored the roles of humans and Generative AI in conceptual design processes, leaving a gap for human-AI collaboration investigation. To address this gap, this study uncovers the contributions of different Generative AI technologies in assisting humans in the conceptual design process. Novice designers completed two design tasks with or without the assistance of Generative AI. Results revealed that Generative AI primarily assists humans in problem definition and idea generation stages, while idea selection and evaluation remain predominantly human-led. Additionally, with Generative AI assistance, the idea selection and evaluation stages were further enhanced. Based on the findings, we discuss the role of Generative AI in human-AI collaboration and implications for enhancing future conceptual design support with Generative AI assistance.

研究动机与目标

• Examine how Generative AI technologies contribute to conceptual design for novices.
• Identify which stages of conceptual design benefit most from AI assistance.
• Assess AI performance and designer perception across different AI tools.
• Characterize the prompt content strategies used by designers when interacting with AI.
• Provide implications for human-AI collaboration in early design phases.

提出的方法

• Four groups of novice designers performed two design tasks with either ChatGPT, Midjourney, both, or no AI assistance.
• Text-to-text (ChatGPT) and text-to-image (Midjourney) tools were used to generate conceptual outputs.
• Quantitative evaluation included participant surveys (speed, subject, diversity, novelty, triggering ideas, requirement satisfaction) and expert ratings on novelty, feasibility, usability, functional diversity, and cost.
• Prompts were collected and coded to identify design-stage strategies and collaboration patterns (affinity diagramming used for theme extraction).
• Inter-rater reliability for prompt coding: Cohen’s kappa = 0.74; expert rating inter-rater reliability: kappa = 0.66.]
• research_questions:[

实验结果

研究问题

• RQ1 RQ1: In which stages of conceptual design do Generative AI tools provide actual assistance?
• RQ2 RQ2: How do participants and experts evaluate the performance of Generative AI in the design process?
• RQ3 RQ3: What are the characteristic prompt-content strategies used by designers across stages?

主要发现

Group	Speed	Subject	Diversity	Novelty	Trigger more ideas	Requirement satisfaction
ChatGPT Group	6.0 (0.71)	4.6 (1.14)	3.8 (1.48)	3.8 (0.89)	4.8 (1.30)	5.0 (1.87)
Midjourney Group	5.0 (1.00)	5.6 (1.14)	5.2 (1.48)	5.4 (1.14)	5.6 (0.89)	4.8 (0.84)
Combined Group	4.6 (1.14)	5.2 (0.84)	4.6 (1.14)	4.0 (1.22)	4.6 (1.14)	3.8 (0.84)
Human Group*	N/A	4.8 (0.84)	3.4 (0.55)	3.6 (0.55)	4.6 (1.14)	N/A

• Generative AI most strongly assists problem definition (≈73.3%) and idea generation (≈86.7%), with idea evolution also aided (≈60%).
• Idea selection and evaluation remains predominantly human-led (≈86.7%), highlighting complementary roles between AI and humans.
• Midjourney achieved the highest expert overall score (mean 4.34, SD 0.75) with significant differences in novelty, cost, and overall performance (p<0.01).
• ChatGPT scored highest for speed (mean 6.0, SD 0.71) and requirement satisfaction (mean 5.0, SD 1.87) among AI tools.
• Combined AI use generally outperformed the Human group in expert ratings, indicating benefits from leveraging both tools.
• Compared with humans, AI-enabled groups expanded design option space and increased diversity and novelty in outputs