Content uploaded by Yu-Min Fang
Author content
All content in this area was uploaded by Yu-Min Fang on May 17, 2024
Content may be subject to copyright.
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 1
The Role of Generative AI in Industrial Design: Enhancing
the Design Process and Learning
(中譯標題：生成式人工智慧對工業設計的影響：設計程序與學習)
Yu-Min Fang (方裕民)
Department of Industrial Design, National United University
(教授。國立聯合大學工業設計學系)
FanGeo@nuu.edu.tw; Tel.: +886 37-381664; Fax: +886 37-355753
註：本文為中譯加註版，請以英文原意為準
Abstract
Generative artificial intelligence (AI) has emerged as a
cutting-edge technology that utilizes machine learning and
deep learning techniques to create new forms of content, such
as text and images, based on vast amounts of existing data.
What changes will it bring to the field of industrial design,
which heavily relies on text and image communication? This
study explores the impact of generative AI on practical design
processes and design knowledge learning. Generative text and
image methods are taught and applied in the core courses and
computer-aided courses of the Industrial Design Department at
National United University, with a total of 26 sophomore
student participants. Through hands-on activities, process
reflection, group discussions, and questionnaire surveys, this
study concludes that: (1) AI is highly beneficial for the
development of creativity and requires minimal time. As a
result, AI generation is well-suited for the early stages of the
design process, which is the divergent phase and an area where
AI excels. (2) The later stages, which involve detail
adjustments, design optimization, and final renderings
presented to clients, may require AI to spend several times
longer than existing design procedures. Therefore, using AI is
not recommended in the later stages. (3) The more cross-
departmental communication is required in the design process,
especially in the engineering drawing and data handover phase,
the more complex professional parameters are needed, such as
dividing parting lines, design constraints, the fit of internal
parts, and styling, etc. It is not recommended to use AI in this
phase. (4) Operating AI still demands a substantial amount of
knowledge. When inputting prompts, design history, style, and
trend knowledge are necessary. Design decisions still require
design experience knowledge, including evaluating the
rationality of ergonomics, feasibility of mass production, and
cost analysis. This knowledge has become even more crucial
under the influence of AI.
摘要：生成式 AI (Generative artificial intelligence)已成
為一種新興的技術，它利用機器學習和深度學習技術，根
據大量現有數據創建新形式的內容，例如文本和圖像。對
於大量使用文本和圖像溝通的工業設計領域而言，將會帶
來什麼樣的變化？本研究試圖以”實務設計流程”與”
設計知識學習”的角度出發，探討生成式 AI 所帶來的衝
擊。本研究將生成式文字與生成式圖像的方法教授與課堂
實作，帶入國立聯合大學工業設計學系的大二主軸課程及
電腦輔助課程中，參與學生為 26 人。經由課堂實際操作、
過程反思、群體討論、與問卷調查，本研究得出結論：(1)
AI 對於創意的發展非常有幫助，而且花的時間非常少。故
AI 生成適合放在設計階段的前端，畢竟這是設計［發散］
的過程，也是 AI 擅長的地方。(2) 後端［收斂］的過程，
就是細節調校、進行細節的優化、對客戶提出最終精描圖。
這一步驟可能 AI 要比現有設計程序花費數倍的時間。故
不建議使用 AI。(3) AI 生成越到設計程序的跨部門溝通階
段(移交階段)，越需要更多的專家經驗，比如、分模線、設
計限制、內部零件與造型的契合…等等複雜的專業參數，
這部分不建議使用 AI。(4) 操作 AI，仍須大量知識。輸入
prompt 時需具備設計史、風格與思潮的知識。設計決策
仍需設計經驗的知識，包含判斷人體工學合理性、量產製
造可行性、以及成本分析知識。這些知識在 AI 興起的衝擊
下，反而重要性有增無減。
Key words: AI-generated content, Industrial Design, Design
Process, Design Knowledge Learning
Introduction
The emergence of generative artificial intelligence (AI) has
introduced a new technology that utilizes machine learning and
deep learning techniques to create new forms of content [1],
such as text and images, based on a large amount of existing
data. In industrial design, the application of generative AI can
provide designers with new tools and techniques to generate
creative solutions, thereby enhancing the design process. By
training generative AI models on large datasets of existing
designs and design principles, designers can use these models
to generate new designs based on specific design prompts or
criteria. This helps designers to quickly explore a range of
design options and discover new possibilities that may not have
been apparent previously. Additionally, generative AI can be
used to create variations of existing designs, allowing
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 2
designers to easily test and refine their ideas.
Another potential benefit of using generative AI in industrial
design is its ability to facilitate design education. By generating
new designs and variations, generative AI models can help
novice designers identify and learn from patterns and trends in
their work. This can guide them towards a deeper
understanding of design principles and a more refined sense of
design aesthetics.
To investigate these potential advantages and disadvantages,
this study introduces generative text and image techniques into
the core curriculum of the Industrial Design Department at
National United University. A total of 26 sophomore students
participated in this study, which included both theoretical
instruction and hands-on practical exercises. The students
learned how to use generative AI models to generate new
designs and variations based on specific design prompts, as
well as how to evaluate and improve the output of these
concepts. The study also explores how the use of generative AI
can improve the industrial design process. In summary, the
research objectives include: (1) integrating industrial design
education to explore students' emotional responses, technology
acceptance, and identifying the advantages and areas for
improvement of incorporating generative AI; and (2)
implementing generative AI to validate design projects,
examining traditional design procedures, exploring the impact
of generative AI at each step, and proposing suggestions for
redesigning the design process.
前言：生成式人工智能（AI）的出現帶來了一種新技術，
利用機器學習和深度學習技術基於大量現有數據創建新
形式的內容 [1]，如文本和圖像。在工業設計中應用生成式
AI 可以為設計師提供新的工具和技術來生成創意解決方
案來增強設計過程。透過在大型設計和設計原則數據集上
訓練生成式 AI 模型，設計師可以根據特定設計提示或標
準使用這些模型生成新設計。這有助於設計師快速探索一
系列設計選項並發現可能以前不明顯的新可能性。此外，
生成式 AI 還可用於生成現有設計的變體，使設計師能夠
輕鬆測試和完善他們的創意。
使用生成式 AI 在工業設計中的另一個潛在好處是其促
進設計教學。通過生成新設計和變體，生成式 AI 模型可以
幫助新手設計師識別和從他們工作中的模式和趨勢中學
習。這可以引導他們對設計原則的更深入理解和更精緻的
設計美學感知。
為了調查這些潛在優點及缺點，本研究將生成式文本和
圖像技術引入到國立聯合大學工業設計系的核心課程中。
共有 26 名大二學生參加了本研究，包括理論教學和實踐
操作。學生們學會了如何使用生成式 AI 模型根據特定的
設計提示生成新的設計和變體，以及如何評估和改善這些
構想的輸出。研究還探討了在工業設計中，如何使用生成
式AI 以改進工業設計流程。總結以上研究目的，包括：(1)
導入工業設計教學，探索學生的情感反應、科技接受度、
以及導入生成式 AI 的優點點與改進建議。 (2) 實踐採用生
成式 AI 驗證設計專案，檢視傳統設計程序，探索生成式 AI
在每一個步驟的影響，並提出設計流程的改造建議。
Generative AI in Industrial Design Education
The traditional industrial design process is divided into four
stages: Definition, Divergence, Convergence, and Formation
[2]. The Definition stage outlines the scope and characteristics
of the product; Divergence generates ideas in different
directions to expand creativity; Convergence examines all the
generated ideas, selects, evaluates, and narrows the scope;
Formation carries out detailed design, transforms creativity
into a product, and proceeds to mass production. In this study,
we reviewed the existing industrial design process, modified it,
and added steps that can be assisted by generative AI to form a
five-step process. These steps are: Topic Definition,
Establishing Design Characteristics, Keyword Extraction, AI
Image Generation, Judgment and Correction (Fig. 1).
生成式 AI 對工業設計教育的影響：傳統的工業設計程序
分為：Definition, Divergence, Convergence, Formation
四大階段 [2]。Definition 將產品範圍與特性定義清楚；
Divergence 產生不同方向的構想，將創意擴大；
Convergence 檢視產生出來的所有創意，篩選、評價、縮
小範圍；Formation 則進行細部設計、將創意產品化，以
進行量產。本研究檢視現有的工業設計程序，修改後，加
上生成式 AI 可以輔助的步驟，形成五項步驟。分別為：題
目界定, 設計特點訂定, 關鍵字萃取, AI 影像生成, 決策與修
正 (Fig. 1).
Fig. 1 Generative AI assists in the creative development of industrial
design process 生成式 AI 輔助工業設計程序中的構想發展
Generative AI creates new forms of content, including text
and images, based on vast amounts of existing data. In the
industrial design domain, which relies heavily on visual and
textual communication, this study presents detailed action
items for each of the five steps. As illustrated in Figure 2, these
action items include: (1) Inputting the product name and job
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 3
description of an industrial designer into ChatGPT, (2) Asking
AI to propose design features in the role of a senior industrial
designer, (3) Adjusting keywords through trial-and-error, (4)
Executing design decisions, and (5) Incorporating user
feedback to refine the design. By following these detailed
action items, this process can assist in the development of new
design ideas.
AI 生成根據大量現有數據創建新形式的內容，包含文本和
圖像。對於大量使用文本和圖像溝通的工業設計領域，本
研究將其導入前述的五項步驟。Fig. 2 繪出生成式 AI 可以
輔助構想發展的詳細做法，包括：(1) 將產品題目輸入
ChatGPT；(2) 描述工業設計師的工作內容，要求 ChatGPT
以資深工業設計師的角色提出設計特點； (3) 以嘗試錯誤
的方法調整關鍵字； (4) 設計決策。
Fig. 2 Detailed action items for generative AI to assist in the creative
development of industrial design 生成式 AI 輔助工業設計構想發展
的詳細做法(步驟)
In February 2023, I conducted a three-hour class on AI-
generated text and images in a Computer-Aided Industrial
Design course for sophomore students. The purpose of this
class was to explore the potential of using AI to aid in the
development of design concepts for a product design project,
which involved creating a "product with local cultural
symbolism." During class, students were required to submit a
proposal immediately. Around 26 students participated in the
class, and some of their works are presented in Figure 3.
The trial class included the following components:
generating design concepts with AI, operating ChatGPT,
writing design objectives, and identifying keywords
(approximately 1.5 hours). Students also worked with Stable
Diffusion and Midjourney (approximately 1.5 hours). The
class received enthusiastic responses from the students.
2023 二月，作者教授大二學生的電腦輔助工業設計課程，
建構三小時的 AI 生成文本與繪圖課程。這階段的目的是
嘗試輔助主軸設計的構想發展，題目是「帶有地方文化意
象的商品」，上課時學生需立刻繳交試做品。全班約 26 人
上課，圖三選出部分學生作品。
試教內容包含：AI 生成概念、ChatGPT 操作、撰寫設計
目標、找出關鍵字( 約1.5 小時)。操作 Stable Diffusion、
操作 Midjourney (約1.5 小時)。此課堂學生反應熱烈。
Fig. 3 Results of AI-Generated Text and Image Course for
Sophomore Students 大二學生作品：電腦輔助工業設計課程中的
AI 生成文本與繪圖課程，使用 Midjourney
In March 2023, Bing Image Creator was released. I asked
sophomore students to revisit their cultural product design
proposals from a month earlier and generate new images by re-
entering keywords, then compare them to their final products.
The class was divided into two parts: registration and
instruction on the procedure and commonly used prompts (20
minutes), and in-class practice and submission of the works (30
minutes). The following conclusions were drawn: (1) The use
of composite materials can be further enriched to stimulate
students' creativity. (2) The generated products have good
maturity, including parting lines, mechanical structure, and
reasonable styling. Some of their works are presented in Figure
4.
2023 三月， Bing Image Creator 發布。作者要求大二
學生將已經進行一個月後的文化商品設計提案，重新回頭，
再輸入關鍵字，生成圖像，與自己的最終作品比較。課程
分配如下：註冊與步驟導引、常用 prompt 的教導 (20 分
鐘)，課堂練習與作品繳交(30分鐘)。結論如下：(1) 大致
對學生，可以提供更豐富的複合材質運用刺激。(2) 生成的
產品成熟度不錯，比如分件、結構、造型合理性。
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 4
Fig. 4 Results of 2nd AI-Generated Text and Image Course for
Sophomore Students 第二次的大二學生作品： AI 生成文本與繪
圖課程，使用 Bing Image Creator
In the process of using generative AI to assist in design, the
operator's knowledge of design history terminology and design
style recognition ability are important factors for success. In
other words, entering the prompts required by generative AI is
essentially entering design jargon. Without a background in
design history, it is impossible to adjust the keywords, and
without sensitivity to design language, it is impossible to fine-
tune the AI. Teaching AI generation is essentially teaching
design trends or design history.
In this context, generative AI can indeed assist in learning
design trends. Therefore, introducing generative AI can be
taught alongside courses on "design trends/design
history/design appreciation and evaluation" and "design
language." Please refer to Figure 5 for details.
The following are the conclusions: (1) Generative AI is
very suitable for teaching students with strong and bold styles
or trends, such as Alchimia & Memphis, Zaha Hadid, etc. (2)
However, generative AI is more difficult to distinguish subtle
style trends. For example, it is challenging to discern the subtle
differences between Minimalism, Archetype, and Modernism.
Furthermore, the style of simplicity and subtractive design is
also difficult to obtain good results in practical AI testing.
真正進行生成式 AI 輔助設計的過程中，操作者是否有
足夠的設計史的術語知識、以及設計風格辨識能力，是成
功的重要因素。也就是說，輸入生成式 AI 所需的關鍵字，
根本就是輸入設計專有名詞。沒有設計史背景，就沒辦法
調校關鍵字；對設計英文不敏感，就無法微調 AI。教 AI 生
成 ， 根 本 是 在 教 設 計 思 潮 或 設 計 史 。
依此脈絡， AI 生成的確可以輔助設計思潮的學習。因此，
導入生成式 AI，可以跟「設計思潮/設計史/設計鑑賞與評
析」「設計英文」課程一起教。
以下改進以前的「設計思潮」教案，包含設計風格研究
與設計師/企業研究，請參考，詳見圖五。以下是結論： (1)
AI 生成很適合教導學生風格較強烈奔放的風格或思潮，比
如Alchimia & Memphis, Zaha Hadid,…。(2) AI 生成較
難分辨細微的思潮。比如比較出極簡主義、原型設計、現
代主義的細微差異。比如簡約、減法設計的風格，實測 AI，
也很難有不錯的結果。
Fig. 5 Generative AI can be used to teach design history, style
recognition, and shape analysis 生成式ＡＩ可以用於教授設計
史、設計風格辨識、及造型評析
Student reactions to AI-assisted design were investigated
through a questionnaire survey. The questionnaire consisted
of the following items: (1) Emotional response: The Self-
Assessment Manikin (SAM) scale was used [3], which
included affective valence (positive and negative emotional
responses) [4] and emotional arousal (intensity of emotional
response) [5,6]. (2) Technology acceptance scale: The scale
measured the level of user acceptance of new technology and
included indicators such as perceived usefulness, perceived
ease-of-use, perceived enjoyment, attitude toward using, and
behavioral intention to use [7,8]. (3) Efficiency and
preference. Short answer questions were also included, which
addressed the advantages and disadvantages of using the
technology, comparison with other methods, and the impact
on design. All items, except for the short answer questions,
were measured using a five-point Likert scale. A total of 26
participants completed the survey, with ages ranging from 19
to 21 years old. Of the participants, 15 (57.7%) were male and
11 (42.3%) were female. The results (TABLE I) showed that:
關於學生對生成式ＡＩ輔助設計的反應，問卷調查被執
行。問卷包含：(1) 情感反應：採用 SAM (Self-
Assessment Manikin)量表[3]，包含情緒效價(正、負向
的情感反應) [4]、與情緒喚起(情感反應的強度) [5,6]。(2)
科技接受量表：針對使用者對於新科技的接受程度，包
含指標為：Perceived usefulness, Perceived ease-of-
use, Perceived 有趣性, attitude toward using,
behavioral intention to use [7,8]。(3) 效率、喜好度。
以及簡答題：包含優缺點、成果比較、對設計的影響。
除簡答題外，以上皆採用五級 Likert 量表。最終共有 26
人完成問卷。參與問卷調查指的年齡介於 19~21 歲、男
生15 人(57.7%)，女生 11 人(42.3%)。最終結果為：
(1) Emotional response: The emotional valence and
emotional arousal were 3.85 (SD=0.67) and 3.85
(SD=0.88), respectively, indicating that users
experienced positive emotions when using the AI tool,
and the tool was helpful in eliciting emotional responses
from students.
(2) Technology acceptance scale: The overall mean score of
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 5
technology acceptance was 3.72 (SD=0.84), indicating
that most participants had a high degree of acceptance.
The indicator "behavioral intention to use" had a mean
score of 3.85, indicating that most participants thought
that AI-assisted design was worth participating in and
would use it frequently in the future.
(3) Efficiency: The mean score of efficiency was 4.00
(SD=0.89), indicating that participants were satisfied
with the time they spent using the AI tool. The mean score
of preference was 3.92 (SD=0.84), indicating that
participants had a high level of preference for AI-assisted
design.
(4) Regarding the advantages and disadvantages of AI-
assisted design and its impact on design, participants
believed that the feasibility of design proposals they
independently completed was higher, and that AI still had
some unreasonable aspects that required manual
correction. However, AI-designed proposals were
considered more avant-garde and innovative, with better
color and lighting rendering, higher precision, and faster
assistance in developing ideas.
(1) 情感反應：情緒效價與情緒喚起皆為 3.85(標準差分
別為 0.67, 0.88)，代表使用者使用生成式 AI 工具
時，獲得正向的愉悅感。喚起度也高，應對學生的
學習非常有幫助。
(2) 科技接受量表：科技接受度總平均值為 3.72(標準差
0.84) ，代表大部分參與者的接受度高。其中指標"
使用意圖”達 3.85，代表大部分參與者認為：生成
式AI 輔助設計是值得參與的，未來會經常使用 。
(3) 關於效率，平均高達 4.00 (標準差 0.89)，代表整體
上，參與者對生成式 AI， 完成任務所花費的時間是
滿意的。而參與者的喜好度平均值高達 3.92 (標準差
0.84)。
(4) 關於優缺點與成果比較，參與者認為：自己獨立完
成的設計提案的可行性較高、AI 仍有不少部分不合
理，還是需要人工手動修正；而 AI 設計出的提案比
較前衛與創新、顏色與燈光渲染上較佳、精緻度
高、輔助構想發展較快速。
TABLE I
Mean and standard deviation of questionnaire survey results
Emotional
response
Technology acceptance scale
(Mean=3.72、SD=0.84)
Effi-
ciency
Prefe
-
rence
Valenc
e
Arousa
l
Perceived
usefulnes
s
Perceive
d ease-
of-use
Perceived
enjoymen
t
Attitud
e
Intentio
n to use
Mea
n
3.85
3.85
3.42
3.81
3.81
3.73
3.85
4.00
3.92
SD
0.67
0.88
1.03
0.80
0.69
0.83
0.78
0.89
0.84
Generative AI in Industrial Design Process
To investigate whether generative AI can enhance the
industrial design process, I revisited a previously presented
hair dryer design proposal and implemented AI in its
development. I reconstructed the ideation and development
stages step by step, as well as the final design rendering
proposal, to see what impact AI would have. Figure 6 displays
the trial result, which took approximately two hours. The
preliminary conclusions are as follows: (1) If the generated
design is simple, geometric, and rational, it is generally less
appealing to humans. (2) AI provides significant assistance in
ideation development. For novices, it can expand their ability
to innovate and break free from existing frameworks. AI has a
high level of control over color and material. Sometimes, AI
even provides ideas whose manufacturing maturity, such as the
feasibility of dividing parts, and the processing of structural
components, exceed expectations. The lighting, details, and
even the atmospheric rendering generated by AI are worth
observing and learning from. (3) AI also has obvious errors,
such as the configuration of the power cord, which seems
illogical. The details, position, and relationship with hand grip
of buttons are also somewhat mixed up. These parts still need
the guidance of a senior designer to correct mistakes. (4) For
experienced designers, the ability to make judgments on
production, modify details, predict costs, and coordinate
communication for productization are all core competitive
strengths that AI cannot replace. Without these abilities,
decision-making under the guidance of AI may lead to
significant errors.
生成式 AI 與工業設計流程：為了實際驗證生成式 AI 是
否可以改進工業設計流程，作者將先前已經提案給客戶的
吹風機設計案重新採用 AI 生成軟體跑一次。作者一步一
步重作構想發展階段，以及最終的設計 rendering 提案，
看看到底 AI 會帶來何種衝擊？Fig. 6 為試作結果，約花費
兩個小時。初步結論為：
1. 若生成簡約、幾何、理性的造型，通常 AI 較不討喜。
2. AI 生成對構想發展，有重大幫助。對於一個新手，可以
擴展創新、跳脫框架的能力。AI 的色彩、材質，掌握度
甚高。甚至 AI 提供的部分構想，其量產成熟度、比如
分件的可行性、結構件的處理，有時也超乎預期。而渲
染的光影、細節打光、或甚至劇場氛圍，都值得讓新手
觀摩學習。
3. 當然 AI 也有顯而易見的錯誤：比如電源線配置，似乎
邏輯轉不過來。比如按鍵的細節、位置、跟手部握持的
關係，也是部分錯亂。這部分還是需要資深設計師的引
導、改正錯誤比較好。
4. 對於資深設計師：量產性判斷、細節修改技術、成本的
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 6
預判、產品化溝通協調能力，都是 AI 無法取代的核心
競爭力。缺乏這些能力，在 AI 的誤導下，決策時可能
會犯下重大錯誤。
Fig. 6 The use of Midjourney to input prompts of different design
styles, stimulating the development of innovative hair dryer design
proposals. Fang, Y. M. (2023). 使用 Midjourney 輸入不同設計風格
的提示詞，激發創新吹風設計方案的構想發展。 方裕民 (2023)。
Stable Diffusion holds potential as a collaborative tool for
industrial design, in addition to Midjourney. While Stable
Diffusion excels in generating images and boasts a wide range
of applications, precise direction control is crucial in industrial
design. To achieve this, the draft sketch of the idea can be
uploaded and the ControlNet extension can be used [9]. The
Preprocessor selects scribble, and the Model selects
control_sd15_scribble to obtain renderings, which are close in
quality to proposals that can be presented to the client (see Fig.
7).
除了 Midjourney 的應用外，Stable Diffusion 也適合最
為工業設計的協作工具。雖然 Stable Diffusion 產生圖像
的能力相當強大，應用也非常廣。但是對於工業設計而言，
為了控制產品發展的大致走向，可以上傳構想草圖手稿，
使用 ControlNet 的Extension。其中，Preprocessor 選
擇scribble，而 Model 選用 control_sd15_scribble，可
以獲得 rendering，其品質接近於可以提案給客戶的程度
(Fig. 7)。
Fig. 7 The draft sketch of the idea can be uploaded to Stable Diffusion
and the ControlNet extension can be used. 上傳構想草圖手稿至to
Stable Diffusion，使用 ControlNet 的Extension
Discussion
The following summarizes the advantages and disadvantages
of applying generative AI to industrial design, and presents the
results in Figures 8 and 9.
Advantages:
Generative AI offers considerable benefits in fostering
creativity while minimizing the time required. For example, it
can generate 50 concepts in only ten minutes, with one-third of
them being particularly useful for building upon existing ideas.
Moreover, generative AI is especially well-suited for
elaborating on more robust and expressive styles, such as the
design trends of Alchimia & Memphis and Zaha Hadid,
enabling designers to effortlessly break away from
conventional thought patterns.
Disadvantages:
However, generative AI faces challenges in discerning subtle
differences between design trends, such as "minimalism,"
"prototype design," and "modernism." This becomes
particularly evident when trying to generate results that adhere
to principles of simplicity and reductionism. Furthermore,
optimizing certain aspects may require a tenfold increase in
labor and effort, such as button layout, partial adjustments to
grip shape, and rationalization of detailed structures, making
this approach inefficient in terms of time and effort expended.
討論：
以下總結生成式 AI 運用在工業設計的優點及缺點，並
將結果呈現在圖 8與圖 9：
優點
：
AI 對於創意的發展非常有幫助，而且花的時間非常少。
比如十分鐘可以產生 50 個構想，而且有 1/3 對擴展概念
非常有幫助。
AI 生成很適合擴展風格較強烈、較奔放的造型。對設計
思潮理解的設計師，比如 Alchimia & Memphis, Zaha
Hadid Style，對於跳脫思考的框架，更是得心應手。
缺點
：
但是 AI 生成較難分辨細微的思潮。很難分辨「極簡主
義、原型設計、現代主義」之間的不同。尤其是簡約、
減法設計的風格，實測 AI 也很難有不錯的結果。
假使要進行某部分的優化，可能花費的精力要大十倍。
比如：例如按鈕的佈局、握把形狀的局部調整、 細部結
構的合理化….，工時成效比實在不高。
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 7
Fig. 8 Arguments for and against the use of generative AI in industrial
design. 對於生成式 AI 運用在工業設計的贊成與反對的論點
Fig. 9 The requirement of the use of generative AI in industrial design.
對於生成式 AI 運用在工業設計中，設計師的知識與經驗需求
The following are essential considerations when employing
generative AI in industrial design:
Limitations:
Notably, ChatGPT operates via language probability and can
engage in natural language conversations. However, it may
produce erroneous responses when dealing with logical
calculations, such as 999999 + 10. Likewise, Midjourney relies
on a keyword-based approach, searching for and combining
images with the highest probability of being associated with
relevant keywords. Despite this, Midjourney does not truly
understand the thought processes involved in industrial design.
以下是運用生成式 AI 時必須注意的地方：
限制
：ChatGPT 會以語言機率計算如何接話，可以自
然語言對話。但是對於邏輯算比如 999999+10 回答錯
誤。Midjourney 也是如此：MJ 並不是真的理解工業設
計的思考邏輯；反而只是依照關鍵字，尋找接近關鍵字
機率最大的圖片，擷取其中圖像，再融合一起。
Conclusion
Generative AI is well-suited for assisting in the development
of design concepts, particularly in the front-end of the design
process. After all, this is the stage of design "divergence," an
area in which AI excels. In contrast, the back-end
"convergence" stage of the process involves fine-tuning details,
optimizing specific elements, and presenting final detailed
sketches to clients. For example, rationalizing button layout
and making partial adjustments to grip shape may require ten
times the amount of time. Considering the opportunity cost, it
may be more efficient to maintain existing parameter-based
modeling and incorporate the designer's own experience into
detailed design, rather than relying on AI to accomplish this
task.
As generative AI progresses further into the design process,
it increasingly threatens the core competencies of industrial
designers. In essence, this involves inputting existing expert
knowledge parameters, such as informing AI of the parting line
location, specifying which parts should remain stationary,
determining the size of enclosed components, and identifying
available styling spaces. It is not recommended to use AI for
this aspect of the design process.
Utilizing AI still requires a wealth of knowledge. When
inputting prompts, a solid background in design history, style,
and trends is necessary. Design decisions still require
knowledge of design experience, including evaluating
ergonomics, assessing manufacturability, and analyzing costs.
Under the impact of the rising prominence of AI, the
significance of this knowledge has, in fact, become
increasingly important rather than diminishing.
Regarding user interface concerns, at present, generative AI
poses challenges for industrial designers lacking programming
expertise, primarily due to suboptimal interface design. As
professional graphics enterprises and productivity software
companies devote substantial resources to enhancing the user
interface and integrating intuitive, WYSIWYG features, the
learning and application processes are expected to become
increasingly streamlined.
Finally, in light of the swift progression of technology, it is
imperative to closely examine the ongoing impact of
generative AI. It is anticipated that researchers and industrial
designers will persist in investigating the potential of future
developments as technology continues to advance.
總結：
1. AI 生成適合協助構想發展輔助階段，就是前端。畢竟
這是設計［發散］的過程，也是 AI 擅長的地方。
2. 後端［收斂］的過程，就是細節調校、進行某部分的
優化、對客戶提出最終精描圖。比如：按鈕的佈局合
理化、握把形狀的局部調整。這一步驟可能要花費十
倍的時間，以機會成本來看，建議保持現有的參數式
建模，將設計師自己的經驗投入細節設計，而非控制
AI 去做這件事。
3. AI 生成越到設計程序後段，越威脅工業設計師的核心
能力。其實這就是輸入現有專家經驗參數：比如告訴
AI 分模線在哪裡？哪部部位不準動？裡面包覆的零件
是多大？造型餘裕的空間有哪些？這部分不建議使用
AI。
4. 操作 AI，仍須大量知識。輸入 prompt 時需具備設計
IEEE conference: 2023 9th International Conference on Applied System Innovation (ICASI)
Page 8
史、風格與思潮的知識。設計決策仍需設計經驗的知識，
包含判斷人體工學合理性、量產製造可行性、以及成本
分析知識。這些知識在 AI 浪潮的衝擊下，反而重要性
有增無減。
5. 至於使用者介面問題，目前生成式 AI 對於無程式編寫
能力的工業設計師而言，操作上無法很順利。但是，這
是由於當前介面設計並無優化。當專業繪圖企業、生產
力軟體體公司大量投入資源，改善使用者介面後，加上
直覺、所見及所得的介面優化，學習與應用將會更簡易。
6. 最後，由於技術發展快速，生成式 AI 的影響，仍須繼
續觀察。期許學者與工業設計師，能隨著技術的演進，
持續進一步探索未來的可能。
Funding: This research was funded by the National Science
and Technology Council of Taiwan, grant number MOST 111-
2410-H-239-016-.
Conflicts of Interest: The author declared no potential
conflicts of interest with respect to the research, authorship,
and/or publication of this article.
References
[1] Cao, Y., Li, S., Liu, Y., Yan, Z., Dai, Y., Yu, P. S., & Sun, L.
(2023). A Comprehensive Survey of Ai-Generated Content
(AIGC): A History Of Generative AI From Gan To ChatGPT.
arXiv preprint arXiv:2303.04226
[2] Howard, T. J., Culley, S. J., & Dekoninck, E. (2008). Describing
the creative design process by the integration of engineering
design and cognitive psychology literature. Design studies, 29(2),
160-180.
[3] Fang, Y. M., & Huang, S. Y. (2021). Comparison of Digital
Applications and Conventional Equipment in Group and
Individual Recreational Activities: Social Psychology, Social
Interactions, Emotional Reaction, and Perceived Usability in
Middle-Aged and Senior Citizens. SAGE Open, 11(4),
21582440211065764.
[4] Wundt, W. " Lectures on Human and Animal Psychology".
Translated by JE Creighton and EB Titchener. 1894.
[5] Hadley,C.B.; Mackay,D.G. Does emotion help or hinder
immediate memory? Arousal versus priority-binding
mechanisms. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 2006, 32.1: 79.
[6] Mackay, D.G.; Hadley, C.B.; Schwartz, J.H. Relations between
emotion, illusory word perception, and orthographic repetition
blindness: Tests of binding theory. The Quarterly Journal of
Experimental Psychology Section A, 2005, 58.8: 1514-1533.
[7] Lederer, A. L., Maupin, D. J., Sena, M. P., & Zhuang, Y., 2000,
The technology acceptance model and the World Wide Web.
Decision Support Systems, 29(3), 269-282.
[8] Davis, Fred D. Bagozzi, Richard P. and Warshaw, Paul R., 1989.
User acceptance of computer technology: a comparison of two
theoretical models. Management Science, 35(8), 982-1003.
[9] Zhang, L., & Agrawala, M. (2023). Adding conditional control
to text-to-image diffusion models. arXiv preprint
arXiv:2302.05543.
備註
此論文發表於：
IEEE conference: 2023 9th International Conference on Applied
System Innovation (IEEE ICASI 2023) . Chiba, Japan, April 21-25,
2023.
https://2023.icasi-conf.net/
標號及領域：
J23058, IV7. AI applications of Engineering and Education
Citation 1 :
Fang, Y. M. The Role of Generative AI in Industrial Design:
Enhancing the Design Process and Learning. The 9th IEEE
International Conference on Applied System Innovation 2023 (IEEE
ICASI 2023). Chiba, Japan on 21–25 Apr 2023
Citation 2 (另收錄於資料庫 IET Digital Library)
Fang, Yu-Min. (2023) The Role of Generative AI in Industrial Design:
Enhancing the Design Process and Education. International
Conference on Innovation, Communication and Engineering (ICICE
2023), 2023 p. 135 – 136. IET Digital Library：DOI:
10.1049/icp.2024.0303.
