Trust in AI shapes how generative tools boost engineering students’ creativity, study finds

Generative artificial intelligence has swept into engineering classrooms faster than almost any educational technology before it, and a new study suggests that whether these tools actually make students more creative depends less on the software itself than on what students believe about it. Research published in Current Psychology by Huifen Guo, Zhen Zhou, and Fengqi Ma of Guangzhou University examined how generative AI usage influences the creativity of engineering students, and found that the relationship runs through a chain of psychological intermediaries: students’ confidence in their own AI abilities, their grasp of what AI can and cannot do, and their perception of what the technology affords them. Trust in AI, the study shows, acts as a powerful switch that amplifies or dampens these pathways.

The research team built their investigation around a structural equation model, a statistical framework that allows researchers to test networks of hypothesized cause-and-effect relationships simultaneously rather than one variable at a time. Using SmartPLS software, the standard tool for partial least squares structural equation modeling, they evaluated how generative AI usage affects creativity both directly and indirectly through three mediating variables: AI self-efficacy, AI literacy, and AI affordance. AI self-efficacy refers to a student’s belief in their own capacity to use AI tools effectively, a concept rooted in Bandura’s social cognitive theory. AI literacy captures knowledge and competence concerning AI systems, including awareness of their limitations. AI affordance describes the perceived action possibilities that AI tools offer, the sense of what a user can actually accomplish with them.

The findings confirmed that generative AI usage has a statistically significant effect on engineering students’ creativity, with a test statistic exceeding the conventional significance threshold of 1.96 and a p-value below 0.05. The effect size fell in what researchers classify as the moderate range, between 0.15 and 0.35, meaning the relationship is real and meaningful but not overwhelming. More revealing than the direct effect, however, was the mediation structure. The impact of generative AI usage on creativity was transmitted primarily through AI self-efficacy and AI affordance, both of which showed moderate downstream effects on creative outcomes. AI literacy, by contrast, carried a notably weaker indirect effect, even though generative AI usage strongly influenced all three mediators.

This asymmetry between the mediators is one of the study’s most provocative results. Students who used generative AI tools developed stronger confidence, sharper literacy, and richer perceptions of affordance across the board, yet only confidence and perceived affordance translated efficiently into creative performance. Literacy, the researchers suggest, may function more as background knowledge than as an active engine of creativity. Knowing how a large language model works, in other words, does not necessarily mean a student will push it toward novel engineering solutions. What matters more is whether students feel capable of wielding the tool and whether they perceive it as opening genuine possibilities for design and problem-solving.

The study’s second analytical layer addressed trust, and here the design became more sophisticated. Rather than treating trust as a single average effect across all participants, the team split the sample into two groups: a trust-in-AI group and a distrust-in-AI group. Multigroup analysis, a technique that tests whether structural paths differ significantly between subpopulations, revealed that the effects of trust in AI on the relationships between generative AI usage and the three mediators were significantly stronger in the trusting group than in the distrusting group. In practical terms, students who trusted AI extracted more psychological benefit from every interaction with the technology, while skeptical students gained less from the same usage.

This trust-dependent architecture has implications that ripple well beyond engineering education. The result echoes a growing body of literature on the double-edged nature of generative AI in creative domains. Earlier work published in Current Psychology documented paradoxical tensions in how employees use generative AI, finding that the same tools can either support or undermine creative output depending on how they are deployed. Research on ChatGPT dependence has linked inert thinking to overreliance on chatbots, while studies of creative self-efficacy in STEM education argue that beliefs about one’s own creative capacity are central to whether students engage in creative work at all. The new study threads these findings together by showing that trust modulates the entire psychological pipeline from tool usage to creative belief to creative behavior.

The engineering context matters. Engineering creativity is typically characterized as functional creativity, the generation of novel solutions to complex technical problems, a concept developed in the systems models of Cropley and colleagues. Unlike artistic creativity, engineering creativity must satisfy constraints of feasibility, utility, and safety, which makes it a demanding test case for whether generative AI can enhance human ingenuity rather than merely accelerate routine output. Prior systematic reviews have identified fostering creativity in science and engineering education as a persistent challenge, and the arrival of large language models has forced educators to reconsider how creative skills are taught and assessed. The Guangzhou study contributes a psychological mechanism to this debate: generative AI does not act on creativity directly so much as it acts on students’ self-beliefs and perceived possibilities, which then act on creativity.

Methodologically, the research follows established conventions for survey-based structural modeling in behavioral research, including attention to common method biases, a concern documented extensively by Podsakoff and colleagues in their influential review of bias remedies in behavioral research. The study received ethical approval from the Ethics Review Committee of the School of Education at Guangzhou University, and informed consent was obtained from all participants. The authors report no conflicts of interest, and the work was supported by a 2023 Special Task Project of Humanities and Social Sciences Research by the Ministry of Education of China. The raw data supporting the study’s conclusions will be made available by the authors without undue reservation.

For educators and policymakers, the practical message is that simply providing access to generative AI tools is unlikely to be sufficient. If creativity gains flow through self-efficacy and perceived affordance, then training programs should target those beliefs directly, building students’ confidence through scaffolded practice and demonstrating concrete creative possibilities that the tools enable. The weak indirect effect of AI literacy suggests that literacy curricula, while valuable for risk awareness and ethical understanding, should not be expected to drive creative outcomes on their own. And the trust findings imply that institutional culture around AI, including transparency about how systems work and honest discussion of their failure modes, may shape how much psychological benefit students derive from usage.

The study also leaves open questions that future research will need to address. The moderate effect sizes indicate that generative AI usage explains only part of the variance in engineering creativity, leaving room for factors such as team dynamics, prior creative experience, and disciplinary context. The trust-based grouping raises the question of what causes students to trust or distrust AI in the first place, a topic explored in recent chain-mediation work linking AI literacy to perceived trust. As generative AI becomes embedded in engineering curricula worldwide, the Guangzhou findings suggest that the decisive variable may not be the sophistication of the model but the psychology of the student holding the prompt, a conclusion that reframes the debate about AI and creativity from a question about machines to a question about minds.

Subject of Research: The psychological mechanisms linking generative AI usage to creativity in engineering students

Article Title: Engineering creativity through generative artificial intelligence: the role of artificial intelligence self-efficacy, literacy, affordance, and trust

Article References: Guo, H., Zhou, Z., & Ma, F. (2026). Engineering creativity through generative artificial intelligence: the role of artificial intelligence self-efficacy, literacy, affordance, and trust. Current Psychology, 45(20), Article 1634. https://doi.org/10.1007/s12144-026-10177-7

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10177-7

Keywords: generative artificial intelligence, engineering creativity, AI self-efficacy, AI literacy, AI affordance, trust in AI, structural equation modeling, engineering education, Current Psychology, Guangzhou University, social cognitive theory, creativity research