Qualitative Analysis of Lecturers' Self-Efficacy and Perception of Artificial Intelligence Tools for Educational Practices in Higher Education

Synthesis: Muhammed and Yusuf (2026) interviewed 30 lecturers at a Nigerian university, six from each of five discipline categories, about their confidence in using generative AI for teaching and about its effects on their practice. Most described positive perceptions and high Self-Efficacy, naming chatbots and writing tools they use for research, grading, lecture notes and virtual assistants, while a minority — concentrated among professors and in language departments — used AI barely or not at all. Adoption tracked discipline: science and applied-science lecturers integrated the tools more readily, while humanities and language lecturers doubted their fit with cultural context, language nuance and creative work. What cut across every discipline was concern about student integrity, expressed as plagiarism, intellectual laziness and a shift from deep reading to immediate results. The study is a useful map of a Global South faculty in transition and a reminder that self-reported confidence is not evidence of competent use.

Key Findings

1. Thirty lecturers, one interview each, five discipline categories. Six lecturers from Sciences, Social Sciences, Humanities, Applied Sciences and Applied Social Sciences were interviewed one to one at a single Nigerian university.
2. Most reported positive perceptions and confidence. Lecturers described using chatbots and grammar tools for research, grading, simplifying lecture notes and virtual assistants, with one rating her proficiency 8 out of 10.
3. Adoption split by discipline, not by age. STEM and applied-science lecturers integrated AI as an extension of analytical workflows, while humanities and language lecturers reported caution or non-use.
4. Humanities objections were about fit, not fear. Lecturers questioned whether the tools handle cultural context, language nuance and the critical and creative demands of literature and discourse analysis.
5. A professor in environmental sciences barred both personal and student use. The stated grounds were intellectual laziness, plagiarism and a feared decline in critical thinking.
6. Integrity concern was universal and independent of personal use. Almost every participant raised academic dishonesty and over-reliance on immediate results, including lecturers who used the tools daily.
7. One professor had been credited with work he did not write. A lecturer reported that AI attributed external publications to him, a concrete case of the accuracy problem faculty fear in student submissions.

What confidence looked like in practice

Self-efficacy in this study is a belief statement, and the interviews show how thin that measure can be. Lecturers rated their own ability, described tools they had opened, and in one case named a proficiency score; none of the accounts is paired with an observed teaching task or a student outcome. The reported uses — summarizing lecture notes, drafting material, checking grammar, serving as a virtual assistant — cluster at the preparation end of teaching rather than in the interaction where Self-Efficacy and subject knowledge get tested. The authors read the pattern through Bandura's social cognitive theory, where mastery experience, vicarious experience, verbal persuasion and emotional arousal shape belief and, through belief, adoption. That framing fits the data, and it also implies the intervention: if confidence follows mastery experience, the way to build capability is guided use on real tasks rather than persuasion.

The disciplinary split and what it is really about

The finding that science lecturers adopt and humanities lecturers hesitate is easy to read as temperament and is better read as a claim about the object of study. A lecturer in literature or discourse analysis is judging whether a tool can handle ambiguity, register and cultural meaning, and their skepticism is a professional judgment about fit rather than resistance to technology — the paper itself notes that these lecturers felt the tools did not align with local cultural contexts or the creative and critical demands of their subjects. That interpretation has a consequence for faculty development: a generic workshop teaches everyone the same tool for the same tasks, and what this study describes is a faculty whose needs diverge by what they teach. It also explains the otherwise puzzling co-occurrence of high confidence and low use in the same interviews.

When the shared concern is integrity

Almost every participant, whatever their own practice, described students using AI to skip the work. The specific worries were plagiarism, intellectual laziness and the displacement of deep reading by immediate results, and they appeared alongside recognition that the same tools were useful for the lecturers themselves. The convergence is the paper's most portable finding, because it does not depend on discipline or on the Nigerian setting: faculty who use AI confidently still expect AI Governance to come from somewhere else. The paper's own recommendation is institutional rather than individual — systematic training workshops that let staff understand the tools well enough to recognize misuse, plus ethical frameworks written into university policy — which is the response that follows once integrity is understood as a policy and assessment problem rather than a character problem.

What this means for practice

• Faculty developers. Target development by discipline: run separate sessions for humanities and language faculty on where these tools fail on register, cultural context and argument, instead of one generic AI workshop for the whole faculty.
• Faculty developers. Build mastery experience on real course tasks. Confidence in these interviews tracked what people had actually done, so guided use on a live assignment is the lever, not advocacy.
• Administrators. Publish assessment and AI-use rules, because the integrity concern in this study came from lecturers who use the tools themselves and who expect the institution, not the individual, to set the boundary.
• Instructors. Redesign at least one Assessment to observe process, since the fear participants voiced was that students reach results without doing the reading that used to come first.
• Researchers. Pair self-reported confidence with observed practice, and report the tool versions lecturers name, because an 8-out-of-10 rating of a 2026 chatbot does not transfer to the next model.

Limitations

• The sample is 30 lecturers at one Nigerian university, recruited purposively with six per discipline category, so the disciplinary comparison rests on six people per group.
• Self-efficacy is measured entirely by interview self-report; no observation of teaching, student outcome or artifact was collected, and the paper reports no limitations section of its own.
• The paper gives no participant demographics beyond discipline, so the claim that professors and language lecturers were the non-users rests on the individual cases quoted.
• Coding and thematic analysis are described without naming the coders or reporting any check on agreement between them.
• The tool versions lecturers used are not reported, and the study predates a peer-reviewed version, so both the capability claims and the adoption pattern are time-stamped to 2026.

Citation

Muhammed, A. A., & Yusuf, M. O. (2026). Qualitative Analysis of Lecturers' Self-Efficacy and Perception of Artificial Intelligence Tools for Educational Practices in Higher Education. EdArXiv preprint.