核心概念
圖像生成式 AI (IGAI) 在參與式設計中,應優先考慮生成能夠促進更豐富對話的「不完美」圖像,而非追求準確但可能阻礙討論的「完美」圖像。
摘要
文獻類型:研究論文
書目資訊:
Guridi, J. A., Hwang, A. H., Santo, D., Goula, M., Cheyre, C., Humphreys, L., & Rangel, M. (2024). From Fake Perfects to Conversational Imperfects: Exploring Image-Generative AI as a Boundary Object for Participatory Design of Public Spaces. In Proceedings of the ACM on Human-Computer Interaction, 8(CSCW2), 1–34. https://doi.org/XXXXXXX.XXXXXXX
研究目標:
本研究旨在探討圖像生成式 AI (IGAI) 如何支持設計師和利益相關者共同設計公共空間的參與式過程。
研究方法:
研究團隊與洛杉磯的景觀設計公司 Studio MLA 合作,針對洛杉磯一個公園的升級改造項目,進行了三個階段的研究:
- 推測階段: 分析先前訪談記錄,探討 IGAI 在參與式設計中的機遇和挑戰。
- 試點階段: 進行內部訪談,測試使用 IGAI 的流程和方法。
- IGAI 介導訪談階段: 對年輕合法移民進行 IGAI 介導的訪談,利用 IGAI 生成圖像,以引出他們對公園設計的想法和需求。
研究團隊收集了訪談和工作坊的音頻、屏幕錄像以及研究人員的反思記錄,並使用親和圖和軸心編碼法進行分析。
主要發現:
- IGAI 生成圖像的成功標準應從「準確性」轉變為「促進對話」。與追求「虛假完美」相比,「對話式不完美」的圖像更能激發參與者的想法,揭示隱藏的需求,並促進更深入的討論。
- IGAI 促進了空間感知對話。參與者在 IGAI 生成的圖像的幫助下,能夠更具體、更準確地描述空間和體驗,並更有效地溝通複雜的概念。
- IGAI 生成的結果取決於引導者管理反應和互動的能力。引導者需要善於應對 IGAI 生成的意外結果,並引導參與者進行有意義的討論。
研究意義:
本研究為 IGAI 在公共空間參與式設計中的應用提供了實證依據,並提出了設計 IGAI 參與式工具的實踐建議,例如:
- 工具設計應側重於促進引導者與參與者之間更豐富對話的界面和功能。
- 引導者需要具備 AI 素養、技術專長和軟技能,才能有效地使用 IGAI 進行參與式設計。
研究局限與未來方向:
- 本研究的樣本量較小,且參與者背景相對單一。未來研究可以擴大樣本量,並納入更多不同背景的參與者。
- IGAI 技術本身存在偏差和可解釋性等問題,需要進一步研究如何減輕這些問題對參與式設計的影響。
引述
“It took at least four or five changes to get a satisfactory picture. Sometimes, we felt a later picture was even worse than the first one.” (D1)
“There is a limited range of manipulation of images with prompts, potentially leading to significant differences [from expectations] and omissions of key elements” (D2).
“This allows us to know the nuances of landscape elements important to the experience of space. Maybe they are particularly shaped by their cultural values, upbringing, or migrant identity. This allows richer conversations and their voices to be heard.” (F5)
“Whimsical unrealistic images can be generative in helping participants to think more creatively” (F10)
“Even absurd images can be analyzed through a spatial lens to enrich design thinking for transformation beyond banal reproductions of what is already known” (F2).
“[Using IGAI] introduced interactive and iterative elements to engagement, allowing the creation of experiential dynamics and reactions in real-time. These more interactive direct dialogues and activities help form bonds with people more authentically, promoting one-to-one interactions.” (F3)
“Image creation helps dissipate tension or nervousness; it shifts the focus from a binary dynamic where the person asking questions becomes less prominent” (F2).
“Reaching mundane design outcomes and not knowing what to do with them can close the conversations” (F1).
“There were instances the interviewees did not comment on images to make further changes after only a few iterations, which did not reveal much information” (F5).
“resistance or alienation from the participant when resonance between the interviewee and the image did not happen” (F2).
“protocols could help to improve the process with suggestions on how to prompt” (F8).
“We don’t know if the AI is driving too much (...) there is obscurity in where the images are selected from and how they build upon” (F2).
“Images without a starting point seem idealized and US-looking. To what extent can it work as an artifact to represent memories for migrants?” (F8).
“The use of partial arrangements on a chosen reference image perhaps limited the possibilities to unveil imaginaries; images showed some common sense but accepted and normalized an existing situation. For example, it was hard to go beyond limited expressions of translocality (e.g., colors, flags) after the first results (F1).