"Five African Universities' AI Pilots: Governance First, Chatbot Second"

Five universities in five African countries have started the same experiment at the same time, and the most interesting thing about it is what they refused to do first. They did not hand every student a chatbot and call that adoption. Led by WISE and Ashesi University's Education Collaborative, universities in Benin, Botswana, Kenya, Nigeria and Rwanda are running nine-month pilots, from October 2026 to June 2027, in which the governance comes before the gadget: each institution is testing one specific, bounded use of AI in teaching, student support or administration, with rules about data, human review and evidence written in from the start.

The five projects are deliberately different, and that is the point. Bells University of Technology in Nigeria is building an assistant and tutoring approach grounded in approved curricula, so the answers students get are supposed to trace back to what they were actually taught. Botho University in Botswana is pairing staff development and governance with an early-warning tool for student mentoring, the kind of system that can help a tutor reach a struggling student earlier — or quietly mislabel one whose attendance or assessment record is simply incomplete. KCA University in Kenya is trying adaptive learning checkpoints with alerts for lecturers. The University of Kigali in Rwanda is working on a conversational tutor built on course material, with a dashboard and human review. Université d'Abomey-Calavi in Benin is concentrating on the unglamorous backbone: institutional governance, AI champions, and declarations of AI use in dissertations.

I find this framing refreshing because it treats the usual campus AI debate as a false choice. Blanket prohibition and uncontrolled adoption are both weak strategies, and both are easy: one requires only a ban, the other only a licence. A bounded pilot is harder. It forces an institution to say what it expects the tool to improve, what evidence would justify expanding it, and what failure would look like. A tutor that answers quickly while encouraging shallow learning is a failure, even if students like it. An early-warning system that flags the wrong students is a failure, even if its dashboard looks impressive. The pilots come after research across 47 universities and 3,875 respondents, which found that only four in ten participating institutions had a comprehensive AI policy, while much of the reported use was already flowing through learning-management systems. In other words, the absence of a policy never meant the absence of AI; it just meant nobody was governing it.

There is also a harder question hiding in the data-protection paragraph of every such programme, and I am glad this one does not skip it. Student records can include performance, disability, financial circumstances and disciplinary information. A university that feeds those records into a system without minimising what it receives, documenting who can see it, and making sure a commercial provider cannot quietly reuse it has not adopted AI; it has outsourced a duty of care. Students, too, need to know when AI is involved in the advice they receive or the assessment of their work. Transparency here is not a courtesy, it is part of the pedagogy: a student who cannot tell where a judgement came from cannot learn to question it.

The consortium plans to publish case studies in July 2027 and share lessons at a later WISE summit. Those reports will matter most if they include the failures — the approaches that did not work, the staff workload, the cost of keeping each system running once grant-funded support ends. A pilot that only works with outside technical help is a demonstration, not a solution, and universities in particular should be allergic to demonstrations that cannot survive contact with an ordinary term. Cross-country comparison is the real prize here. Five institutions operating under different languages, regulations, connectivity conditions and institutional models, all asking the same questions about accuracy, adoption, staff confidence, appeals and cost, can produce guidance that a single showcase never could.

If there is a lesson for universities everywhere, it is this: the institutions that will use AI well are not the ones that move fastest, but the ones that decide, in advance and in public, what would count as working. These five have given themselves nine months to find out. I will be watching what they report in July — especially the parts that did not go to plan.