Human-AI collaboration in higher education: Exploring the impact of technology expectations and distrust

Synthesis: Human-AI collaboration in higher education: Exploring the impact of technology expectations and distrust

Key Findings

• Drawing on expectation confirmation theory (ECT), the study models how students' expectations about generative AI and their confirmation of those expectations shape intentions to collaborate with GenAI.
• An online survey of 245 higher education students (59.2% female, 68.2% aged 20–25, including international students at a Danish university) who already use GenAI found positive and significant relationships between effort expectation and performance expectation (β = 0.530, p < .001), and from performance expectation to confirmation (β = 0.571, p < .001), as ECT predicts.
• Positive confirmation of GenAI collaboration significantly influences students' intentions to collaborate (β = 0.587, p < .001), which in turn positively impacts their behavior (β = 0.513, p < .001); the model explains 50.2% of the variance in performance expectation, 55.3% in confirmation, 47.2% in collaboration intention, and 31.2% in collaboration behavior.
• GenAI distrust negatively moderates two key links: (1) the relationship between effort expectation and performance expectation (β = −0.169, p < .01), and (2) the relationship between expectation confirmation and intentions to collaborate with GenAI (β = −0.099, p < .05); it did not significantly moderate the other three hypothesized paths.
• Perceived drawbacks center on the risk of plagiarism (72.2% agree or strongly agree), inaccuracy of responses (70.6%), and flawed referencing (58.0%), while time saving (89.0%), user-friendliness (79.2%), and enhanced productivity (76.7%) dominate perceived benefits — evidence that benefit and drawback perceptions coexist rather than exclude each other.
• The work extends the theoretical boundaries of ECT into the human-AI collaboration context and clarifies the distinct role of distrust in educational technology adoption.

Study Design & Method

The authors administered an online survey to higher education students who use GenAI at a Danish university, measuring the core ECT constructs — effort expectation, performance expectation, confirmation, and collaboration intention — along with a measure of distrust toward GenAI, using items drawn from prior expectation-confirmation and information-systems adoption research. The hypothesized model was estimated with partial least squares structural equation modeling (PLS-SEM) in SmartPLS 4 with 5000 bootstrap samples, with all constructs meeting reliability and validity thresholds and good model fit (SRMR = 0.070; NFI = 0.884). The moderation analysis tests whether distrust weakens the expectation-driven pathway to collaboration intention, and an importance-performance map analysis (IPMA) identified confirmation as the most important predictor of both intentions and actual collaboration while effort expectation showed the highest performance. ChatGPT was the most used tool, and reported collaboration concentrated on higher-level cognitive tasks — explaining and evaluating theories and concepts, editing and proofreading text, summarizing literature, and generating research ideas — suggesting augmentation of learning more than automation. This design positions the study as one of the first to bring ECT's confirmatory framework to bear on Human AI Collaboration in Higher Education rather than on conventional information systems.

What this means for practice

• Instructors. Manage expectations before capability: since distrust dampens the effect of confirmation on collaboration intention, onboarding, transparency about limitations and trust-building belong in any responsible deployment of Generative AI rather than being treated as optional.
• Instructors. Design feedback and Scaffolding that help students form accurate expectations, so that a positive first experience converts into sustained collaboration — a practical complement to AI Literacy teaching about evaluating AI output.
• Designers. Treat distrust and perceived benefit as separate signals: students acknowledged the tool's usefulness while remaining skeptical of its outputs, so expectation management and transparency are complements to capability, not substitutes for it.
• Administrators. Have responsible-use guidance address plagiarism risk, inaccuracy and flawed referencing directly for Student Experience design, rather than assuming that skepticism about the tool will resolve those problems on its own.

Limitations

• Expectations and first impressions cannot be separated. The authors note that the rapid evolution of GenAI makes initial expectations and early experiences occur simultaneously, and this study could not measure them apart.
• One institution's students, one country. The findings come from 245 students at a single Danish university and cannot be assumed to generalize to the global student population; multi-institutional and cross-national samples are needed.
• Trust measured with two items. The distrust construct rested on two items only, capturing skepticism toward the technology and lack of trust in its outputs.
• Two moderation paths left unresolved. Hypotheses H6c and H6e returned positive but non-significant coefficients, which the authors say warrant qualitative work, larger samples or experimental designs.

Citation

Razmerita, L., Zheng, X., & Allen, J. P. (2026). Human-AI collaboration in higher education: Exploring the impact of technology expectations and distrust.