SuperSkillsStack: Agency, Domain Knowledge, Imagination, and Taste in Human-AI Design Education

This study analyzes how 80 student design teams integrated generative AI into their creative process, revealing that while AI serves as a cognitive accelerator for early-stage tasks like brainstorming, human competencies in agency, domain knowledge, imagination, and taste remain essential for interpreting context, validating outputs, and refining design solutions.

The Gist

Imagine you are a chef trying to create a brand-new, award-winning dish. Suddenly, a super-smart robot assistant walks into your kitchen. This robot can chop vegetables in a millisecond, suggest 500 different spice combinations, and write a perfect recipe description in seconds.

But here's the catch: The robot has never actually tasted the food, smelled the ingredients, or seen the customers who will eat it.

This is exactly what the paper "SuperSkillsStack" is about. It looks at how design students (our "chefs") are learning to work with Generative AI (the "robot assistant") in their creative projects. Instead of letting the robot take over the kitchen, the researchers found that the best students use a specific set of human skills to keep the robot helpful without letting it run the show.

They call these four human skills the "SuperSkillsStack." Here is what they are, explained with simple analogies:

1. Agency: Being the Captain, Not the Passenger

The Analogy: Think of AI as a very fast, powerful car. Agency is your ability to hold the steering wheel.

• What the students did: They didn't just let the car drive itself to wherever it wanted. They decided when to turn the engine on, where to drive, and when to stop.
• In the study: Students realized that for certain parts of the job (like interviewing people or looking at a messy construction site), the robot shouldn't be driving at all. They chose to use the robot only for brainstorming or organizing notes, but they kept the final decision-making power in their own hands. They treated the AI as a tool, not a boss.

2. Domain Knowledge: The "Local Expert" vs. The "Tourist"

The Analogy: Imagine a tourist with a map (the AI) and a local resident who has lived there for 20 years (the student). The tourist's map might say, "Turn left at the big red building," but the local knows that building was demolished last week.

• What the students did: The AI is like the tourist; it has read millions of books and knows general facts, but it doesn't know the specific reality of the project.
• In the study: When the AI suggested a design for a park, it might have been technically correct but totally wrong for that specific neighborhood. The students used their own "local knowledge" (gained by actually visiting the site and talking to people) to catch these mistakes. They acted as the fact-checkers who said, "No, that won't work here because the ground is wet and the locals hate that color."

3. Imagination: The Spark Plug

The Analogy: Think of AI as a machine that can shoot out 1,000 fireworks in a second. Imagination is your ability to look at that explosion and say, "Hey, that blue spark looks like a dragon!" and then build a story around it.

• What the students did: They used the AI to break out of creative ruts. When they were stuck, the AI would give them a wild, weird idea.
• In the study: The students didn't just copy the AI's ideas. They used the AI's suggestions as a "jumping-off point." The AI provided the raw material (the fireworks), but the students provided the vision to turn those sparks into a coherent design. The AI expanded their options, but the humans chose which path to walk down.

4. Taste: The Final Taste-Test

The Analogy: This is the most important part. Imagine the robot chef has made 100 different soups. Taste is your ability to take a spoonful of each, close your eyes, and say, "This one is too salty," "This one is boring," and "This one is a masterpiece."

• What the students did: The AI is great at making lots of things, but it's terrible at knowing what is good or appropriate.
• In the study: Students constantly had to filter the AI's output. They rejected ideas that sounded fancy but were actually useless. They rewrote the AI's text to make it sound human and authentic. They used their "gut feeling" and experience to decide what was worth keeping and what should be thrown in the trash.

The Big Takeaway

The paper concludes that AI is a "Cognitive Accelerator," not a replacement.

Think of it like this:

• Before AI: A student had to dig through a library for 10 hours to find 5 ideas.
• With AI: The student gets 500 ideas in 5 seconds.
• The Result: The student saves time on the search, but they still have to do the hard work of judging which idea is the best.

The Lesson for Education:
We shouldn't teach students how to just "use the robot." Instead, we need to teach them how to be better captains, better local experts, and better judges. If we do that, AI becomes a super-powerful sidekick that helps humans create amazing things, rather than a tool that makes humans lazy.

In short: The robot can generate the noise, but only a human can make the music.

Technical Summary

Here is a detailed technical summary of the paper "SuperSkillsStack: Agency, Domain Knowledge, Imagination, and Taste in Human-AI Design Education" by Qian Huang and King Wang Poon.

1. Problem Statement

The rapid integration of generative Artificial Intelligence (AI) into creative practice and education has created a pedagogical tension. While AI tools offer unprecedented efficiency in generating ideas, synthesizing data, and producing visual artifacts, there is a critical concern that reliance on these tools may lead to cognitive outsourcing, where students bypass the development of fundamental human capabilities such as observation, empathy, and critical judgment.

The core research problem addresses the gap in understanding how students actually integrate AI into design workflows and whether they can maintain the "uniquely human" capabilities necessary for meaningful design practice. Specifically, the study seeks to determine how students balance AI-generated insights with human observation, contextual knowledge, and evaluative judgment.

2. Methodology

The study employs a qualitative thematic analysis of reflective writings produced by student design teams.

• Data Source: Reflective submissions from 80 student teams enrolled in a human-centered design course at the Singapore University of Technology and Design (SUTD).
• Context: Students conducted design projects focused on real-world environments (e.g., public transport, healthcare clinics, hawker centers). They were encouraged but not mandated to use generative AI tools (e.g., ChatGPT, Gemini, Perplexity, image generators) throughout the process.
• Analytical Framework: The study utilizes the SuperSkillsStack framework (Huang et al., 2025) as the primary coding lens. This framework defines four interrelated human competencies required for effective human-AI collaboration:
  1. Agency: The ability to control and orchestrate AI tool usage.
  2. Domain Knowledge: Contextual understanding derived from fieldwork to validate AI outputs.
  3. Imagination: The capacity to generate and expand upon novel ideas.
  4. Taste: The ability to evaluate the aesthetic, conceptual, and practical quality of ideas.
• Coding Process: A two-stage coding approach was used:
  • Stage 1 (Deductive): Segments of text were coded against the four SuperSkillsStack categories.
  • Stage 2 (Inductive): Thematic patterns were identified to map the human-AI workflow across different design stages (observation, ideation, analysis, evaluation).

3. Key Contributions

The paper makes three primary contributions to the fields of design education and human-AI interaction:

1. Operationalization of the SuperSkillsStack: It provides empirical evidence validating the SuperSkillsStack framework as a viable model for analyzing human-AI collaboration in educational settings.
2. Empirical Workflow Mapping: It moves beyond theoretical speculation to document the actual sequence of human-AI interaction, revealing that students do not use AI as a replacement for creativity but as a specific accelerator for early-stage divergence.
3. Pedagogical Shift: It argues for a shift in design education from teaching technical AI literacy to cultivating higher-order cognitive capabilities (Agency, Domain Knowledge, Imagination, Taste) that allow students to act as "conductors" of AI rather than passive consumers.

4. Results

The analysis of 80 team reflections revealed consistent patterns regarding the application of the four competencies:

• Agency (Strategic Control): Students actively decided when to use AI and when to exclude it.
  • Finding: AI was frequently excluded during stages requiring deep empathy or direct observation (e.g., user interviews, site analysis) because students recognized AI's inability to interpret body language or physical context.
  • Action: Students experimented with multiple tools and refined prompts, treating AI as a subordinate instrument rather than an autonomous decision-maker.
• Domain Knowledge (Contextual Validation): This was the primary mechanism for detecting AI hallucinations and inaccuracies.
  • Finding: Students relied on fieldwork (site visits, interviews) to verify AI-generated diagrams and insights.
  • Action: When AI produced plausible but generic or inaccurate information (e.g., incorrect site layouts), students used their domain knowledge to filter and correct these outputs.
• Imagination (Cognitive Acceleration): AI was most effective during the divergent phase of design thinking.
  • Finding: Students used AI to overcome creative blocks, generate initial idea pools, and reframe problem statements.
  • Action: AI served as a "catalyst" to expand the search space of ideas, but the development and elaboration of these concepts remained a human task.
• Taste (Evaluative Judgment): This was the dominant capability in the convergent phase.
  • Finding: Taste appeared in every group reflection. Students consistently rejected AI outputs that were "polished but unrealistic" or lacked contextual nuance.
  • Action: Teams engaged in critical rewriting, manual sketching, and feasibility checks to refine AI suggestions into meaningful solutions.

Synthesis of Workflow: The study identified a recurring Human-AI Design Workflow:

1. AI generates initial ideas/summaries (Acceleration).
2. Students interpret outputs using Domain Knowledge (Validation).
3. Teams filter ideas through critical discussion (Taste).
4. Students refine designs through human judgment (Finalization).

5. Significance

The study concludes that generative AI functions primarily as a cognitive accelerator rather than a replacement for human creativity. Its significance lies in redefining the role of the designer in the AI era:

• From Producer to Orchestrator: The designer's role shifts from solitary idea generation to orchestrating the interaction between human insight and computational generation.
• Reinforcement of Human-Centric Skills: The findings suggest that AI integration does not diminish the need for human skills; rather, it increases the value of Domain Knowledge (to ground AI in reality) and Taste (to curate AI output).
• Educational Implications: Design curricula must evolve to prioritize:
  • Fieldwork and Experiential Learning: Essential for building the domain knowledge needed to audit AI.
  • Critical Evaluation: Teaching students to distinguish between superficial AI generation and meaningful design solutions.
  • Strategic Agency: Training students to decide when AI is a tool, a teammate, or a hindrance.

Ultimately, the paper posits that the future of design education lies in cultivating the SuperSkillsStack to ensure that human designers remain the ultimate arbiters of context, ethics, and meaning in an AI-augmented world.