Human-AI Collaboration Pushes Design Students in Two Directions at Once
Generative AI is no longer sitting at the edge of design education. It has moved into the center of idea generation, scheme iteration, and decision-making, where students now use it as a collaborator rather than a passive tool. A PLOS ONE study in 483 design students in East China shows that this shift does not produce a single educational effect. It can strengthen design competence, but it can also erode it through a separate, equally measurable pathway.
The study framed this tension through human-AI collaboration and Stress Appraisal Theory, asking how students interpret AI-supported work: as a challenge that expands capability, or as a hindrance that raises confusion, overload, and anxiety. Those appraisals mattered more than the technology itself. A tool that speeds production can also reshape how students think about their own skill, effort, and authority over the creative process.
One finding was straightforward. When students experienced AI collaboration as a challenge, their competence development improved. The positive pathway ran through challenge appraisal, which rose with human-AI collaboration and then predicted better competence development. The indirect effect was statistically significant, and the path coefficient for challenge appraisal was positive at β = 0.283, with competence development following at β = 0.363.
Another finding cut in the opposite direction.
When AI collaboration was experienced as a barrier, students reported hindrance appraisal, and competence development fell. The negative path was also statistically significant, with human-AI collaboration predicting hindrance appraisal at β = 0.271 and hindrance appraisal predicting competence development at β = −0.229. The result is a partial mediating structure with one positive route and one negative route, rather than a simple net benefit from AI use.
Why the Same Technology Can Build Skill and Drain It
The logic behind the dual effect is clinically interesting in a psychological sense, even though the setting is educational rather than biomedical. Generative AI reduces friction in visual production, lowers the technical threshold for beginners, and allows rapid iteration. Those features can make students feel more capable because they can test more ideas, see more options, and move faster from concept to output. In appraisal terms, the technology becomes a resource.
The same system can also create cognitive strain. If students lean too heavily on AI-generated suggestions, they may face uncertainty about originality, fear of making the wrong choice, or a sense that the machine is steering the design process. That can push the interaction into a hindrance frame, where AI no longer feels empowering but destabilizing. In practical terms, the tool may save time while simultaneously raising the psychological cost of judgment.
The study’s authors used a sample of 483 design students from East China and analyzed the data with SPSS 24.0 and Amos 23.0. Reliability and validity testing, regression analysis, and Bootstrap mediation analysis were all part of the statistical approach. The design does not measure objective studio performance or portfolio grades. It does, however, map how students cognitively process AI-rich learning environments, and that process appears to matter for competence development.
Teacher AI Leadership Buffered Risk, Not Creativity Gains
The most asymmetric result involved teacher AI leadership. Stronger teacher AI leadership reduced the negative impact of human-AI collaboration on hindrance appraisal, with a moderating effect of β = −0.112. In plain terms, teachers who understand AI and can guide its use appear to help students manage cognitive load, uncertainty, and anxiety around the technology.
The effect did not extend to the positive pathway. Teacher AI leadership did not significantly amplify the challenge-appraisal route, with β = 0.001 and a nonsignificant p value above 0.05. That is a narrow but important result. Teacher leadership appears to work less as a creativity multiplier and more as a risk-control mechanism, helping students avoid the cognitive drag that can emerge when AI becomes too opaque or too dominating.
For design education, this distinction carries real consequences. An instructor who can explain how to prompt, critique, revise, and verify AI output may protect students from passive dependence. The educator becomes a cognitive guide rather than a content dispenser. That shift aligns with the study’s argument that teaching should move from knowledge delivery toward cognitive facilitation in AI-augmented classrooms.
A Partial Mediation Model With Concrete Classroom Consequences
The central takeaway is not that human-AI collaboration is good or bad. It is that the psychological outcome depends on how the collaboration is appraised. Challenge appraisals support competence development; hindrance appraisals suppress it. Both are present in the same educational environment, and both are statistically meaningful.
This creates a very practical problem for design schools. If curricula expand AI use without building teacher capacity, students may gain speed but lose confidence, judgment, or depth of learning. If instructors are trained to understand AI systems and to integrate them into assessment, critique, and iteration, the harmful pathway may weaken. The study suggests that institutional response matters as much as software access.
The authors also point to a broader assessment issue. Traditional evaluation systems built around static outputs may not capture whether a student used AI in a way that strengthened understanding or merely outsourced the hardest parts of the task. A more AI-aware education model would need to assess process, reflection, and decision-making, not just finished work.
The findings arrive at a time when AI tools are already embedded in major design platforms and in university-level creative training. The study does not claim that AI inevitably improves competence, nor does it argue that AI is inherently corrosive. Instead, it shows that the educational value of AI collaboration is filtered through students’ psychological appraisals and through the competence of the teacher guiding the process.
What remains unresolved is how these patterns would hold across other disciplines, other age groups, or more objective measures of design performance. This study is anchored in one student cohort and in self-reported constructs, which leaves open the question of whether the same dual pathway appears in actual studio output, long-term skill retention, or professional practice.