The emergence of artificial intelligence (AI) in design education is changing how students engage with meaning-making and visual ideation. With increased application of AI tools, particularly generative models like Midjourney, their influence on visual representation and semiotic interpretation in design requires increased attention. This paper presents a two-year study conducted in a second-year undergraduate course, Design Semiotics for Interior Architecture and Environmental Design, where nearly 40 student projects explored the integration of AI-generated imagery into the learning process. The aim of this study is to reveal how AI can support students in understanding the embedded cultural and social meanings within design while simultaneously challenging conventional notions of authorship and visual communication. The methodology of this study follows a structured, iterative process spanning five weeks.

Students worked in pairs through four key stages:
(1) formulating structured prompts that linked abstract concepts, spatial contexts, and semiotic terminology, then using Midjourney to generate corresponding images;
(2) engaging in peer critique sessions to analyze and interpret each other’s visuals;
(3) refining their prompts based on semiotic analysis and received feedback; and
(4) creating new images by blending previous iterations and further refining their concepts.

The final stage involved student presentations, where they reflected on their creative process, the affordances and limitations of AI, and the evolving relationship between human intention and AI-generated outputs. The data of this study consists of:
(1) visual outputs (AI-generated images produced through prompt iterations),
(2) textual reflections (self-assessments, insights on AI’s role, challenges),
(3) peer feedback (comments on AI-generated visuals), and
(4) instructor observations (notes on student engagement, struggles, and breakthroughs).

This study adopts a qualitative research approach, emphasizing experiential learning and iterative design. Thematic analysis is used for textual data including reflections and feedback, while image analysis is used for AI-generated outputs and iterative refinements. For holistic analysis, textual (reflections) and visual (images) data are linked, showing common and contradictory themes. The findings of this study highlight both opportunities and challenges in integrating generative AI tools into design education. The opportunities include how AI encourages new ways of thinking about semiotics by challenging conventional methods of image generation and interpretation, offers a unique space for experimentation by allowing students to engage with unexpected visual outcomes and to explore alternative ways of representing meaning, and enhances students’ ability to conceptualize abstract ideas, particularly in the context of semiotics. The challenges include how AI generates unpredictable outputs, has ambiguous prompt formulation, and offers questions regarding originality and authorship. The discussion of the paper addresses the role of AI as a creative collaborator to prompt new ways of thinking rather than a mere tool to replace human ideation and human-generated design. The conclusion of the paper not only reveals the possible implications of AI for design education to enhance critical thinking, imagination, and interdisciplinary study, but also the limitations such as the challenge to assess AI-generated works.

Keywords: Design education, artificial intelligence, text-to-image, semiotics.