- Artificial Intelligence is transforming academic research, making it imperative for universities to establish clear ethical guidelines that distinguish responsible AI assistance from academic misconduct.
- As AI becomes increasingly integrated into postgraduate education, institutions must update their academic integrity policies to safeguard originality, critical thinking, and the credibility of university qualifications.
- Rather than banning AI, universities should promote its transparent and responsible use through clear policies, disclosure requirements, and AI literacy to preserve the integrity of scholarship.
For decades, universities have fought one of academia’s greatest enemies: plagiarism. Through strict citation rules and software such as Turnitin, institutions have established acceptable similarity thresholds to protect originality and scholarly integrity. A thesis with a 45 per cent similarity index would almost certainly be returned for correction, while one within an acceptable threshold would proceed.
But a new academic challenge has quietly emerged—one that plagiarism policies alone cannot address: Artificial Intelligence (AI). Across universities worldwide, postgraduate students are increasingly turning to AI-powered tools such as ChatGPT, Grammarly, and QuillBot to draft proposals, polish dissertations, summarize journal articles, and even generate literature reviews. What was once considered a futuristic technology has now become an everyday academic companion.
Unlike plagiarism, where clear institutional rules exist, AI remains largely unregulated in many universities. This silence is dangerous. Universities must now confront a difficult but necessary question: if plagiarism has an acceptable percentage threshold, should AI usage also have one? The answer is increasingly becoming yes.
To be clear, AI is not inherently academic misconduct. Just as calculators did not destroy mathematics and search engines did not kill research, AI is simply a tool. Its ethicality depends on how it is used. A student who uses AI to correct grammar, improve sentence flow, or simplify a complex explanation is not necessarily compromising academic integrity. In fact, such use may enhance communication, especially for students writing in a second language.
The problem begins when AI shifts from being an assistant to becoming the actual author. A postgraduate student who asks AI to “write a literature review on climate adaptation in arid Kenya” and submits it with minimal revision has crossed a dangerous line. Another who uses AI to generate a research methodology without understanding sampling procedures or analytical models is essentially outsourcing intellectual labour. Worse still, there are cases where AI tools generate non-existent citations or fabricate references—what experts now call “hallucinations.” This is where the integrity crisis begins.
Postgraduate education is fundamentally designed to test independent thinking, analytical reasoning, and original contributions to knowledge. A master’s or PhD thesis is not merely about producing pages; it is about demonstrating scholarly maturity. When AI does too much of the intellectual heavy lifting, the learning process itself is undermined. In such a case, there is every reason to question whether AI should effectively be treated as a co-author.
Consider this: if a PhD candidate cannot independently formulate a conceptual framework, critically analyse findings, or defend methodological choices without AI support, what exactly is the university certifying? A degree should represent competence, not technological dependency. This is why universities need to establish acceptable AI thresholds in research publications, much like plagiarism thresholds—not because AI is inherently problematic, but because boundaries matter.
It is important to note that there is currently no universal benchmark for what constitutes an acceptable AI percentage in academic publications. Most journals and publishers do not formally rely on AI-detection scores as a criterion for acceptance. Instead, the emphasis remains on originality of thought, accountability for authorship, transparency in disclosure, and the ethical use of technology.
Even so, a practical guide is beginning to emerge. Research writing that reflects between 0 and 10 per cent AI-likeness is generally considered the safest zone, as it often reads as authentically human-authored and rarely raises editorial concern, especially for high-stakes scholarly work. A range of 10 to 25 per cent may still be tolerable if the intellectual contribution remains clearly that of the researcher and the text has been carefully refined.
Beyond 25 to 30 per cent, however, the work enters a risk zone where journals may subject it to closer scrutiny, particularly where AI screening is in place, and questions about authorship integrity may arise. Once AI-likeness climbs to 40 or 50 per cent and above, the concern becomes far more serious, with many editors likely to suspect overdependence on machine-generated content and some institutions or publishers potentially requesting explanations or even rejecting the work on ethical grounds.
An AI threshold would not necessarily mean “how much AI wrote this paper,” since measuring that precisely remains difficult. Rather, it would serve as an ethical guide. For example, institutions could adopt a framework where AI assistance below 10 per cent is considered minimal and acceptable, especially for editing and language refinement. A range between 10 and 20 per cent could require disclosure by the student, indicating where AI was used. Anything beyond that could trigger closer scrutiny by supervisors or examiners. Such a framework would help distinguish assistance from dependence.
Of course, AI detection itself remains controversial. Tools such as GPTZero and Originality.ai claim to estimate AI-generated content, but they are far from perfect. They sometimes flag entirely human-written work as AI-generated while missing heavily AI-assisted texts altogether. This makes policy-making more complicated. But imperfection should not justify inaction.
After all, plagiarism-detection software is not flawless either, yet universities still use it as a guide. Human judgment remains central. Supervisors can often tell when a student’s writing suddenly changes in tone, complexity, or style. Oral defences, proposal presentations, and viva voce examinations remain powerful tools for testing authentic understanding.
The AI question, therefore, should not be reduced to percentages alone. It should be integrated into a broader academic integrity framework. Universities must update their postgraduate manuals to include AI-use declarations. Students should state whether AI tools were used for language editing, idea structuring, or data interpretation. Supervisors must be trained to distinguish ethical AI support from academic outsourcing. More importantly, institutions should teach AI literacy so that students understand both the power and limitations of these tools.
Banning AI altogether would be unrealistic and counterproductive. Today’s scholars are entering a world where AI will shape industry, research, and policy. Universities must prepare students to engage with it responsibly, not fearfully.
In Kenya, the urgency is even greater. Many universities are grappling with increased postgraduate enrolment, overburdened supervisors, and mounting pressure to publish. These conditions create fertile ground for overreliance on AI. Without policy guidance, students are left to navigate this new terrain blindly, while supervisors are forced to make case-by-case judgments without institutional backing.
This is where the Commission for University Education (CUE) should step in. Just as it sets standards for programme accreditation, institutional accreditation, and quality assurance, it can also provide a national framework on the ethical use of AI in postgraduate research. Such guidance would create consistency across institutions and protect the credibility of Kenyan academic qualifications.
The world of scholarship is changing rapidly. AI is no longer at the gate; it is already inside the classroom, the laboratory, and at the thesis-writing desk. Universities must stop pretending otherwise. The real question is no longer whether students are using AI—they are. The question is whether institutions will regulate it wisely before it quietly redefines what scholarship means.
Like plagiarism, AI needs boundaries. Because, in the end, a university’s duty is not merely to produce graduates, but to produce thinkers.
YOU MAY ALSO READ: Use of AI in Research and Postgraduate Supervision: Balancing Efficiency and Academic Integrity
The Author is a Professor of Chemistry; a former Vice-Chancellor; a Higher Education Expert; a Quality Assurance Consultant and Trainer. Contact: okothmdo@uoeld.ac.ke
