Artificial Intelligence in Libyan Academia: Adoption, Ethics, and Institutional Responses: A Mixed‑Methods Study

Abstract

AI has quickly become an important technology for knowledge creation in academia; however, there is little information about AI adoption in the context of non-Western countries that feature linguistic diversity. The present paper employs a mixed methods approach (sequential explanatory design) to examine the use of artificial intelligence among 130 Libyan researchers who work in public universities and institutes across various career levels and disciplines (STEM, medical sciences, humanities, business, and environmental sciences). Quantitative data were analysed using exploratory factor analysis (EFA), hierarchical cluster analysis (HCA), structural equation modelling (SEM), random forest (RF) classification, and social network analysis (SNA), complemented by thematic analysis of open-ended responses. Key findings: (1) AI adoption is near‑universal (96.2%, 95% CI [92.1, 98.6]), with ChatGPT as the preferred tool (85.5%). (2) Three distinct adopter clusters emerge: Enthusiastic Adopters (32%, daily use 76.2%, low ethical concern), Cautious Integrators (41%, strategic use, high verification), and Skeptical Minimalists (27%, minimal use, greatest ethical concern). (3) Despite widespread verification (73.8% always check AI output), only 11.5% consistently disclose AI assistance, a profound disclosure deficit. (4) Using SEM, we find that institutional policy clarity emerges as the most powerful predictor of disclosure (β = 0.45, p < 0.001), followed by individual ethical concern (β = 0.25, p = 0.008) and disciplinary norms (β = 0.18, p = 0.04). Productivity benefits have no significant effect on disclosure (β = 0.09, p = 0.21). (5) Finally, using random forest classification (accuracy = 78.4%), we find that policy clarity is the most important variable (importance = 0.32). Theoretical contribution: We present the Academic Technology Acceptance Model (ATAM), which builds upon existing TAM and TPB models by incorporating four novel variables: epistemic utility, disciplinary alignment, institutional legitimacy, and ethical compatibility, and illustrates how they impact the adoption intensity vs. disclosure. Practical implications: Training individual academics about the ethics of AI use is ineffective; what is needed are clearly defined, publicly available, and rigorously enforced institutional policies. Universities should develop comprehensive AI policies, make the disclosure process mandatory, and educate their students; journal editors should standardize disclosure templates and develop strict authorship policies; and researchers should adopt rigorous verification practices. Conclusion: AI is much more than just a tool; it is an epistemic infrastructure. Its responsible adoption necessitates institutional measures that ensure transparency, accountability, and ethical integrity across all stages of the research workflow. Without such structural safeguards, the disclosure deficit will persist regardless of individual awareness or productivity gains.