Artificial Intelligence and Higher-Order Thinking: A Systematic Review of Educator and Student Experiences and Perspectives in Higher Education

Synthesis: Patrick et al. (2025) consolidate 31 qualitative and mixed-methods studies — the reported views of 251 tertiary educators and 1,585 higher education students — to ask how AI use in higher education shapes higher-order thinking. Positive experiences outweighed negative ones for both groups: educators co-created lesson content and assessment material with AI, while students described an interactive, experiential partner for idea generation and self-directed learning. Both groups independently raised the same worry, that easily obtained answers produce superficial comprehension and over-reliance. The review's distinctive contribution is structural: it separates three areas of educator–student agreement from a single divergence, and locates that divergence in academic integrity, where educators argue about the credential and students about the assignment.

Key Findings

1. Positive experiences outweighed negative ones for both groups. Educators co-created lesson content, texts, ethical-dilemma scenarios and clinical question banks with AI; students described it as an interactive, experiential partner for generating perspectives and ideas.
2. Three areas of commonality organize the synthesis. Both groups treated AI as a launchpad for content and idea generation, credited it with richer classroom environments, and reported the same worry that easy answers yield surface-level comprehension.
3. Academic integrity is where the two groups split. Educators' concerns were macro-level — ethics, plagiarism, and students' inability to judge the reliability of AI-generated output — while students worried about misinformation inside the work they were producing.
4. Over-reliance and depth of thinking recur across themes. Educators reported students submitting AI output without critical engagement; students described complacency, weakened argumentation, and loss of independent problem-solving after relying on instant answers.
5. The corpus is thin on measured higher-order thinking. Of the 31 included articles, 15 were qualitative, 13 mixed methods and three case studies, and only three employed a specific measure of critical thinking.
6. The sample skews toward student voice. Six papers examined educators' perspectives, one examined both, and the remaining 24 focused on students, consolidating the views of 251 educators and 1,585 students across disciplines and study levels.
7. Both groups ask for clearer governance. Educators and students supported AI integration but emphasized consistent guidelines, institutional policy, and AI competency training built into the curriculum for both groups.

Where educators and students agree

Three claims hold across both groups. First, AI is a launchpad for content and idea generation: educators co-created lesson content with AI (generated texts and prompts for writing tasks, ethical dilemma vignettes, word clouds, clinical scenario videos with question banks), and students used it to surface novel viewpoints and step outside their own habits of thought. Second, both groups reported a richer classroom environment — educators through individualized and timely Feedback, progress monitoring, and more tailored instruction; students through the conversational, human-like quality of the interaction that made exploring a topic systematic and supported Experiential Learning. Third, and least comfortable for advocates, they share the same failure mode: students use AI output without critically analyzing it, and the result is superficial comprehension. The review names this the "slippery slope" between using AI to augment learning and complete dependency on it.

The integrity divergence

Academic integrity was the one clear divergence. Educators' concerns were macro-level: plagiarism and cheating, the absence of any mechanism to validate whether AI-generated information rests on sound evidence, and students' insufficient knowledge to identify the reliability of generated content. Students' concerns were narrower and practical — that inaccuracy would damage the work they were actually producing, through false references, outdated material, or claims that silently omitted necessary information. The review reads the gap as a difference in stake rather than in understanding: educators are accountable for the credential, students for the assignment in front of them. Its proposed remedy is instructional — discipline-specific dialogue about the ethical and responsible use of AI, so students are taught not only how to use the tools but why ethical use matters — which ties the integrity problem to AI Literacy and reduces expectations for detection.

Over-reliance and what the corpus does about it

The over-reliance theme generates the review's most concrete teaching strategies. Rather than restricting tools, the included studies assigned tasks that forced engagement: students compared and contrasted AI-generated text with their own writing (Strobl et al. 2024), critiqued and reflected on AI-generated information in discussion boards (Lin et al. 2024), and compared traditional sources against AI content (Gonsalves 2024). One educator framed the conversational tool as a source-evaluation partner that asks critical questions about credibility. The student-side implication is a balanced approach: routine verification and cross-checking, wide reading, and practicing on real-world problems, because the automatic answer is what weakens metacognitive engagement. The review cites the phrase "metacognitive laziness" (Fan et al. 2025) for exactly this erosion, and treats Problem Solving and argumentation as the capacities that decay first when verification stops being routine.

Assessment and the institutional response

The review's institutional recommendations follow from assessment risk rather than from classroom technique. Its first implication is that traditional essay formats are losing validity: multimedia assignments, viva voce, and real-time oral presentations offer immediate feedback, are harder to steer with injected prompts, and assess performance similar to real workplaces, which connects to Authentic Assessment and Oral Assessment. It recommends flipped learning so that critical work happens in class, AI competency training embedded across the curriculum for students and educators alike, and an AI ethics committee that includes educators and students from several faculties. On this account the institutional question is no longer whether to permit the tools but how to govern them: the review argues that AI integration requires clear, consistently implemented guidelines and policy, and that the universities' governance gap is itself a barrier to developing higher-order thinking in teaching.

What this means for practice

• Instructors. Assign the adjudication, not just the answer: the included studies that reduced over-reliance required students to critique, compare, and fact-check AI-generated output, so build that task into every assignment where AI is permitted.
• Instructors. Co-create with the tool before policing it: the educator theme is AI as a lesson-design partner for texts, prompts, vignettes, and question banks that also supplies individualized feedback and progress monitoring.
• Learners. Make verification routine and keep the reasoning: cross-check AI output against other sources, read widely, and practice on real-world problems, because students themselves linked instant answers to weakened argumentation.
• Administrators. Supply consistent, curriculum-embedded guidance rather than a per-course policy: both groups supported AI use but asked for clear rules and AI competency training, and the review proposes an AI ethics committee that includes educators and students.
• Researchers. Treat self-reported perception as a starting point, not as evidence of cognitive gain: only three of the 31 included studies measured critical thinking with a specific instrument.

Limitations

• Inclusion required a qualitative component, so quantitative and pre-post designs that could measure change in higher-order thinking are outside the sample by construction.
• Of the 31 included articles, 15 were qualitative, 13 mixed methods and three case studies, and only three employed a specific measure of critical thinking, so most evidence is perceptual.
• Data are largely self-reported — surveys, semi-structured interviews, focus groups, and reflective writing — so the reported effects on higher-order thinking are what participants said, not what was assessed.
• Quality appraisal was uneven: the CASP criterion on the researcher–participant relationship was met by only 1 of the 31 studies, and just 22 addressed ethical issues.
• Searches ran in August 2024 with a supplementary search in January 2025 and a cutoff of 31 December 2024, so the corpus reflects the early ChatGPT era of tool use.
• The review could not capture how institutions' own AI policies shaped what participants reported, which the authors identify as a factor left outside the analysis.

Citation

Patrick, P. M., Yip, S. Y., & Campbell, C. (2025). Artificial Intelligence and Higher-Order Thinking: A Systematic Review of Educator and Student Experiences and Perspectives in Higher Education. Higher Education Quarterly, 79, e70069.