Integrating AI in Online Learning in Higher Education: A Literature Review

Synthesis: Lock, Arteaga & Johnson (2025) conduct a two-pronged critical literature review of how AI technologies are integrated into online learning environments in higher education, narrowing 207 initial citations to 63 for the final review (spanning 32 countries). Their thematic analysis yields four interconnected themes: (1) the types and purposes of AI integration; (2) pedagogical approaches, centered on AI literacy and self-regulated learning; (3) benefits of using AI in online learning; and (4) challenges spanning academic integrity, equity and bias, ethics/privacy/surveillance, and institutional policy gaps. The authors position AI integration as a complex, sociotechnical undertaking that must be anchored in pedagogy and human relationships rather than technology adoption.

Key Findings

• Types of AI integration. Fifty-four of the 63 reviewed citations reported specific AI uses in online learning, classifiable into: AI teaching assistants/machine teachers; automated AI monitoring (e.g., remote proctored exams); AI-driven surveillance (e.g., Proctorio, Turnitin, Perusall); algorithms and learning analytics; generative AI (intelligent tutoring, content creation, automated grading); language-support models (translators, voice assistants); large language models and chatbots (the most common, e.g., ChatGPT, Q-Module-Bot); machine learning; and predictive/adaptive early-warning systems.
• Pedagogical approaches cluster around AI literacy, strategy, and self-AI Regulation in Education. The review distinguishes AI literacy from digital literacy, framing the former as the skills and competencies to use AI effectively and the critical capacity to question design and implementation. Pedagogical strategies include reducing transactional distance, providing personalized and adaptive learning, role-play and Simulation (e.g., chatbots embodying patients for anamnesis practice), and supporting a heutagogical, learner-directed shift in teaching philosophy. AI was found useful for metacognitive, cognitive, and behavioral regulation but not for motivational regulation, where learner identity, activeness, and position were necessary complements.
• Benefits center on personalization, analytics, and instructor workload. AI tools personalize learning, provide automatic assessment and timely feedback, run predictive analytics, and reduce instructor time on repetitive tasks so instructors can focus on higher-order needs. Notably, students who used ChatGPT alongside teacher tutoring perceived greater learning and skill improvement than those using ChatGPT alone — a design-contingent finding consistent with hybrid human–AI collaboration.
• Challenges are multidimensional and interconnected. (i) Academic integrity: pandemic-era remote proctored exams and plagiarism software proved inadequate against unsupervised AI access; students criticized punitive surveillance and memory-based assessments displacing authentic learning, while content generation without prior learning "masked" learning and obstructed cognitive development. (ii) Equity and bias: AI nudges rarely benefit all learners, algorithmic scoring can penalize non-native and culturally diverse input, unfamiliar complex systems add cognitive load, and biased metrics risk unfair practice. (iii) Ethics, security, privacy: heavy data collection normalizes surveillance, threatens autonomy, and raises privacy/security risks, especially for vulnerable students. (iv) Institutional support and policy gaps: institutions often lack frameworks for accountable AI deployment, and ambiguous data-ownership/retention policies enable misuse.
• The automation-without-human review critique is vivid. The authors quote a teaching-assistant vignette of "click[ing] release scores" without reading AI comments, arguing such automation without human-in-the-loop undervalues instructor–student interaction and the positive role of human engagement.

Research agenda

Shift focus from creating and deploying AI to how learning is designed, facilitated, and assessed within AI-enhanced environments, including how the instructor's role and the student–instructor–technology relationship change.

It complements the knowledge base's other online-learning syntheses by emphasizing the overlapping, interconnected nature of AI-integration themes in online higher education.

What this means for practice

• Instructors. Keep a human in the workflow: pair GenAI with teacher tutoring rather than substituting it, since the review reports that students who used ChatGPT alongside teacher tutoring perceived greater learning and skill improvement than students using ChatGPT alone, and never release AI-generated feedback unread — the authors use a teaching-assistant vignette of "click[ing] release scores" to argue that automation without human-in-the-loop review devalues instructor–student interaction.
• Instructors. Teach AI literacy as the critical capacity to question how AI is designed and implemented, not just as tool-operation skill, and design assignments that require critical thinking rather than content generation. Give instructors educational development in the areas the review names — introductions to AI, using AI for course development and assessment, using AI in teaching, image generation and customization, and student-focused training on AI use.
• Instructors. Keep motivational regulation with the learner. The review found AI useful for metacognitive, cognitive, and behavioral regulation but not for motivational regulation, where learner identity, activeness, and position were necessary complements.
• Administrators. Close the policy gap the review documents: institutions frequently lack frameworks for accountable AI deployment, and ambiguous data-ownership and retention policies enable misuse.
• Policymakers. Treat equity and surveillance findings as design requirements, not caveats: AI nudges rarely benefit all learners, algorithmic scoring can penalize non-native and culturally diverse language, and heavy data collection normalizes surveillance.

Limitations

• This is a critical (narrative) two-pronged literature review, not a systematic or scoping review; the authors explicitly recommend a future systematic or scoping framework and narrowed 207 initial citations to 63.
• Database coverage was limited to six ProQuest, ERIC, and EBSCO databases plus a hand-built list of 16 educational-technology journals, which the authors state should be broadened for a wider scope.
• Inclusion was restricted to published English texts in refereed journals, books, chapters, and conference proceedings from 2020 to 2025 with a higher education focus, and 5 of the 63 citations reported no geographic location for the study.
• Themes were derived by the author team's own spreadsheet synthesis rather than through a coded protocol with inter-coder agreement checks.

Citation

Lock, J., Arteaga, S., & Johnson, C. (2025). Integrating AI in Online Learning in Higher Education: A Literature Review. International Journal on Innovations in Online Education, 9(1), 59–79.