TLDR: A pilot study at the Singapore University of Technology and Design explored how first-year engineering and architecture students engaged with generative AI in a design course. By introducing a “tool, teammate, or neither” taxonomy, the study found that students developed sophisticated AI literacy, using AI for accelerated prototyping and collaborative ideation, but also deliberately refusing it for tasks requiring human empathy or critical judgment. This approach fostered metacognitive awareness and transformed AI uptake into an assessable design habit, emphasizing the importance of teaching students *how* to think about using AI, not just *whether* to use it.
The rapid rise of generative artificial intelligence (Gen-AI) tools like ChatGPT has significantly impacted higher education, transforming how students approach learning across various disciplines. While initial responses from institutions often focused on concerns about academic integrity and skill erosion, a recent pilot study from the Singapore University of Technology and Design (SUTD) offers a fresh perspective, re-centering student agency in the integration of AI into engineering design education.
The study, titled “To Use or to Refuse? Re-Centering Student Agency with Generative AI in Engineering Design Education”, involved over 500 first-year engineering and architecture students in a 13-week foundational design course. The course was specifically designed to equip students with AI-based design skills through several interventions, including introducing AI-based design methods, incorporating an ‘AI x Design thinking’ competition, and providing access to premium Gen-AI services.
A Three-Way Lens for AI Engagement
A core aspect of the SUTD study was the introduction of a unique three-way taxonomy for students to reflect on their AI use: as a ‘tool’, a ‘teammate’, or ‘neither’ (deliberate non-use). This framework encouraged students to move beyond simply using AI for automation and instead to critically consider agency, ethics, and context in their design process. Students were required to log their AI usage, detailing the platform, its role, and the perceived effect on their design process in terms of time, quality, or ethics.
AI as a Tool: Delegating Drudgery, Unlocking Creativity
When students viewed AI as a ‘tool’, they typically assigned it bounded, efficiency-focused tasks. This included generating quick concept visuals, creating parts lists, or debugging code snippets. For instance, some teams used AI to translate rudimentary sketches into high-fidelity visual mockups within hours, significantly compressing the traditional design cycle. This saved time was then reinvested into more creative or empathetic aspects of their projects, allowing for a greater diversity of ideas and more iterations. Novice coders also found AI invaluable for troubleshooting, using large language models to spot errors and explain complex electronic concepts.
AI as a Teammate: A Collaborative Partner and Co-Designer
The ‘teammate’ role represented a deeper, more collaborative relationship with AI. Here, AI was treated as a dialogical partner capable of critique, debate, and even tutoring. A compelling example involved a team running the same design prompt through four different LLMs (Gemini, ChatGPT, DeepSeek, and Claude) and then collectively reviewing the responses. This practice, dubbed “hallucination fire-drills,” transformed AI from a black box into a collaborative entity, where disagreements among models prompted metacognitive inquiry and helped students challenge their own assumptions. This approach fostered recursive cycles of critique, mirroring metacognitive scaffolding where students evaluated alternative interpretations and anticipated flaws.
AI as Neither: Knowing When to Say No
Perhaps the most insightful finding was the deliberate choice by students to refuse AI at critical junctures, categorizing its role as ‘neither’. This wasn’t a rejection of AI’s value but a situated decision based on tasks demanding human presence, empathy, ethical judgment, or trust that AI couldn’t yet replicate. For example, during user interviews, stakeholder conversations, or empathy mapping, students intentionally excluded AI to preserve human nuance and ensure authentic insights. Some students also chose to avoid AI in technical aspects like coding or wiring to retain their learning opportunities. Misfires, such as AI recommending incompatible electronic parts, also reinforced the need for caution, leading students to cross-check AI suggestions with expert advice and real-world testing.
Also Read:
- Assessing Creativity: A Look at AI vs. Human Grading in Design Thinking Education
- AI Assistant Enhances Interdisciplinary Project Planning for Teachers
Implications for Education
The study’s findings suggest that explicitly teaching students to reflect on and categorize AI interactions develops more sophisticated decision-making frameworks than the traditional automation-versus-augmentation binary. The ‘neither’ category, in particular, highlights that deliberate non-use is a sophisticated form of technological literacy, challenging the assumption that AI adoption should always be maximized. This approach cultivates “hallucination-aware workflows” and moves beyond mere prompt engineering to foster deeper metacognitive skills. The SUTD study demonstrates that a structured integration of AI, combined with reflection requirements, role tagging, and public recognition, can transform AI from a potential threat to academic integrity into a powerful catalyst for deeper learning and design innovation.
Ultimately, the research argues that engineering education must evolve beyond merely teaching students *how* to use AI tools to teaching them *how to think about using them* – knowing when to engage AI, how to use it effectively, and crucially, when to set it aside. This re-centers student agency, preparing them not just for a world with AI, but for a world where strategic AI integration and selective non-use are fundamental professional skills.
