AI evidence extraction

Navigating opportunities and challenges of generative AI in higher education: towards responsible, equitable, and human-centered integration

Claudiu Coman, Vasile Gherheș, Anna Bucs, Ecaterina Coman, Cioca Victoria-Rodica, Diana-Cristina Bódi et al. · Frontiers in Artificial Intelligence · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

7/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

18.95

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.3389/frai.2026.1750978

Methodology & findings

Study design

Systematic literature review with qualitative thematic synthesis.

Sample

N = 27, 1 group

Primary method

This is a qualitative systematic review employing thematic analysis rather than statistical methods. The methodology used was Braun and Clarke's (2006) reflexive thematic analysis framework, which involved: repeated reading for familiarization, open coding of recurring concepts, iterative clustering of codes into provisional themes, and refinement through collaborative discussion. No inferential statistical tests were conducted. The search strategy employed Boolean logic (AND/OR operators) in Web of Science, but no quantitative synthesis or meta-analysis was performed.

Main result

The study found that "GenAI's benefits in improving personalized learning, self-regulated strategies, and adaptive feedback" are significant, yet "critical issues persist, particularly regarding academic integrity, data privacy, equity, and responsible governance." Core findings highlight that "when applied thoughtfully, GenAI holds significant promise in reshaping educational practices while safeguarding fundamental academic values."

Reports effect sizes.

Research paradigm

Critical realism with interpretivist elements

Author conclusions

The authors concluded: "When applied thoughtfully, however, AI can not only enhance learning experiences but also bring long-term benefits by optimizing administrative processes and reducing costs." Additionally, they state: "This paper aims to bridge that gap by systematically reviewing peer-reviewed literature on the integration of generative AI in higher education, with a particular emphasis on responsible governance, equity, and human-centered approaches." The authors emphasize that "the study underscores the need for stakeholder collaboration to ensure the ethical and effective integration of GenAI in education."

Risk of bias

Database selection bias: exclusive reliance on Web of Science may exclude non-indexed regional publications; Language bias: English-language publications only; Publication bias: peer-reviewed journals may skew toward positive or novel findings; Recency bias: most literature focuses on post-ChatGPT (2022+) implementation, potentially missing foundational work; Self-report bias: evidence on AI self-efficacy and responsible usage relies heavily on self-report surveys which may overestimate responsible patterns due to social desirability effects; Language bias: Exclusive use of English-language publications may exclude relevant non-English studies; Database bias: Single database (Web of Science) used; other databases not searched; Publication bias: Reliance on peer-reviewed journal articles may exclude gray literature; Social desirability bias: Self-report surveys on AI usage may overestimate responsible practices; Temporal bias: Rapid evolution of GenAI technologies means findings reflect a moving target; Selection bias: Exclusive reliance on Web of Science may exclude relevant studies from other databases or non-English publications; Reporting bias: The rapid pace of AI development means findings may quickly become outdated; Publication bias: Only peer-reviewed articles were included; gray literature was not searched; Methodological heterogeneity: The 27 included studies likely employ diverse methods, making synthesis complex; Social desirability bias: Many findings on AI self-efficacy rely on self-report surveys which may overestimate responsible usage patterns; No quantitative meta-analysis: Qualitative synthesis may reflect author interpretation of themes

Limitations

The authors stated: "This review is limited by the exclusive use of Web of Science and English-language publications, which may exclude relevant regional or non-English studies." Additionally, "given the rapid evolution of generative AI technologies, the findings reflect a dynamic and continuously developing research landscape." Furthermore, the authors note that "the literature reveals unresolved tensions between process-oriented reform and institutional feasibility
While calls for authenticity and competency-based assessment are conceptually strong, implementation may be constrained by class size, faculty workload, accreditation requirements, and resource disparities." They also highlight that "empirical evidence on redesigned assessment effectiveness remains emergent
Many proposals are normative or conceptual rather than empirically validated at scale."

Open questions raised

Lack of frameworks addressing responsible and inclusive embedding of GenAI into academic systems
Limited research on governance, ethics, and long-term societal consequences of AI adoption (most research emphasizes technical functionality or user perceptions)
Scarcity of comparative, cross-institutional studies evaluating learning outcomes under redesigned GenAI-integrated assessment models
Limited empirical evidence on effectiveness of redesigned assessment formats
Need for longitudinal and observational designs tracking actual usage behaviors over time (current evidence relies heavily on self-report surveys)
Gap in understanding cultural sensitivity and contextual variation in AI literacy and integration across diverse educational settings

Data: No external datasets are mentioned as available. The review synthesizes published literature only.Code: None mentionedExtracted from: pdfAgreement 59%

Explore related topics

Responsible AI Use AI Governance Literature Review