Before ChatGPT entered university classrooms, academic integrity debates largely focused on plagiarism. Today, the conversation has changed.
The real question is no longer whether students use AI but how universities should respond to a technology that is rapidly becoming part of everyday academic work.
While some students misuse AI by submitting machine-generated work as their own, bypassing learning in pursuit of grades, AI also has the potential to strengthen critical thinking, creativity and research when used responsibly.
Attempts to ban or police AI through detection software are proving increasingly ineffective as the technology evolves.
The rise and limits of AI detection
In response to growing concerns about AI-generated assignments, many African universities have turned to AI detection software. The systems attempt to determine whether a document was written by a human or generated by AI by analysing statistical characteristics of language.
Most AI detectors work by looking at how words are arranged in a piece of writing. AI often writes in a more predictable way than humans, the AI detectors try to identify writing that appears unusually consistent.
There has been success in AI detectors catching out AI text, as seen in Walters (2023) where 126 documents were evaluated and found that Copyleaks and Turnitin correctly classified some human-written assignments and AI-generated essays, proving how impressive some detection systems are.
However, the promising statistical results do not necessarily translate into reliable performance in real educational settings.
As AI systems become increasingly sophisticated and are trained on growing amounts of AI-generated content, distinguishing human and machine writing is becoming progressively more difficult.
Some scholars have predicted that human-written training data might be exhausted by 2028, meaning future language models will increasingly be trained on AI-generated content.
When detection creates more problems than it solves
Despite their widespread adoption, AI detectors remain probabilistic rather than evidentiary tools. They estimate the likelihood of AI use but cannot prove misconduct.
Under real-world conditions, no external evidence can conclusively establish whether flagged text was genuinely AI-generated.
False accusations remain unavoidable because human writing often shares characteristics with AI-generated text,.
Students adapt to the threat of being falsely accused of misconduct by deliberately introducing grammatical mistakes, awkward phrasing or stylistic inconsistencies into their assignments
False positives also disproportionately affect students whose writing differs from dominant English-language conventions. This is particularly concerning for African universities, where many students write in English as a second or additional language.
Liang and others (2023) found that widely used AI detectors falsely identified an average of 61.3% of human-written essays by non-native English speakers as AI-generated, raising serious concerns about fairness and the potential marginalisation of multilingual scholars.
Furthermore, the most accurate detection systems are frequently locked behind expensive subscription models, placing many African universities at a financial disadvantage and reinforcing inequalities in access to educational technologies.
Evidence also suggests considerable variation in detector performance. In a study conducted on the effectiveness of 16 software tools designed to detect AI-generated writing, Walters (2023) found that systems such as SEO.ai, Content at Scale and Crossplag produced high numbers of false positives because they tended to classify human-written text as AI-generated.
For these reasons, AI detector scores should never be treated as definitive proof of misconduct. At best, they provide one piece of supporting information that must always be interpreted alongside human academic judgement.
Moreover, African universities find themselves engaged in an escalating technological contest with AI developers. As detection software improves, new AI applications specifically designed to bypass detectors emerge at a swift rate.
This creates a self-defeating cycle. Universities prohibit AI use and invest in increasingly sophisticated detection systems.
Students respond by adopting AI tools specifically engineered to bypass the systems. Institutions then introduce new detectors, prompting the development of even more advanced bypass technologies.
The result is an endless technological arms race that neither side can realistically win.
More importantly, this policing model undermines the educational mission of universities which is to cultivate critical thinking, ethical reasoning, creativity, and the ability to solve complex real-world problems.
When institutional attention becomes dominated by identifying AI use rather than improving learning, the purpose of assessment shifts from education towards surveillance.
Moving from detection to responsible disclosure
A more constructive alternative is emerging. Rather than attempting to determine whether students used AI, African universities should focus on how AI was used and whether its use was academically appropriate.
Lecturers acknowledge the difficulty of distinguishing AI-generated writing from authentic student work during assessment. The recognition supports a growing international shift away from detection and towards transparency.
As such, disclosure frameworks encourage responsible use, attribution and accountability rather than suspicion. Instead of treating AI as something students must conceal, disclosure normalises honest reporting while allowing educators to evaluate whether AI was used appropriately.
The University of KwaZulu-Natal's AI Academic Guidelines offer a practical model by requiring disclosure based on the extent to which AI contributes to scholarly work.
Routine language editing or compiling a reference list requires little disclosure, while AI used for research design or analysis demands far greater transparency.
The approach recognises that not every interaction with AI carries the same academic significance.
Yet this creates an inconsistency because language improvement tools, such as QuillBot, often produce text that AI detectors classify as AI-generated, even when the ideas are the student's own.
Universities therefore need clearer distinctions between AI used for language refinement and AI used for intellectual content creation.
The approach is gaining traction. A history course in the UK required students to use GenAI in an assessment and critically evaluate its outputs, demonstrating that transparent AI use can strengthen learning rather than undermine it.
Rethinking assessment rather than policing students
Improving disclosure alone is not enough. Universities must also rethink how they assess learning.
Traditional essay-based assessments were designed for a time when students did not have access to sophisticated AI tools. Rather than relying more heavily on invigilated examinations, institutions should adopt assessments that evaluate the learning process as well as the final product.
Students can demonstrate their understanding through draft submissions, reflective commentaries, oral presentations, classroom discussions and project-based tasks that require them to explain and justify their thinking.
The focus should shift from asking: "Who wrote this?" to asking: "Does the student understand the work they have presented?"
At the same time, AI literacy must become an essential graduate skill. Students should learn to use AI responsibly, critically evaluate its outputs, recognise its limitations and biases and understand the ethical responsibilities that accompany its use.
Artificial intelligence is here to stay and African universities must adapt accordingly. Rather than relying on imperfect AI detectors, institutions should promote transparency, ethical AI literacy and assessment methods that reward critical thinking and genuine understanding.
The goal is not to stop students from using AI but to ensure they continue to think, question, analyse and create while using it.
Solomon Musonza is an emerging sociologist and an MA sociology candidate at the University of Johannesburg.