East Africa is drafting one AI course for a region of unequal classrooms

THE SIGNAL IN ONE SENTENCE

A shared university course can be a bridge. It can also be a very tidy document that never survives contact with a classroom. East Africa is now trying to build the bridge. The East African Community, the Inter-University Council for East Africa and the East African Science and Technology Commission have validated a draft AI Foundational Module for university students across academic disciplines. Representatives from universities, governments, regulators, development partners and the AI field reviewed the draft during a two-day workshop in Dar es Salaam, Tanzania. The Inter-University Council published the update on October 7, one day after the workshop date it reported. The ambition is refreshingly broad. The module is not meant only for computer-science students who already know their way around code. It is supposed to make AI literacy a basic university competence, whether a student is studying agriculture, law, health, literature, engineering, business or something that has not yet been forced into a chatbot demo. The official description emphasizes responsible use, critical engagement with generated information, data protection, ethical decision-making and practical application in technical and non-technical fields. That is the right neighborhood. AI literacy is not the ability to make a model produce a smoother paragraph. It is the ability to decide when the model belongs in the task, what evidence the result needs, what data should stay out, which people could be harmed and when a human must take responsibility. East Africa has a strong reason to develop that foundation regionally. Students and workers move across borders. Universities collaborate. Employers recruit from several education systems. Public problems such as food security, public health, climate resilience and regional trade do not stop at a national checkpoint. A common learning floor could give graduates shared language for discussing AI systems, even when their disciplines and local contexts differ. The EAC AI Alliance has already been building a regional network for AI education and research. The foundational module turns that regional idea toward the ordinary classroom. But the announcement is a milestone in curriculum design, not proof of classroom delivery. The module remains a draft. Pilot institutions have not been named. The publication does not provide a public syllabus, common budget, device requirements, language plan, lecturer-to-student ratios, assessment instruments or adoption timetable. It says institutions within the EAC Regional AI Network that demonstrate the required readiness will be considered for pilots. Readiness is doing a great deal of work in that sentence. One university may have reliable campus internet, modern labs and lecturers already teaching machine learning. Another may depend on shared devices, expensive mobile data, intermittent power and instructors carrying several courses at once. Some students will arrive with years of digital practice. Others may be meeting a generative AI system for the first time. The same module cannot pretend those differences do not exist. It can still preserve a common standard. The trick is to separate the non-negotiable learning outcomes from the delivery method. Every student should be able to question a generated claim, trace evidence, protect sensitive information, recognize uncertainty, identify bias, disclose appropriate use and explain who remains accountable. Those are shared outcomes. How students reach them can vary. One campus might use a live model in a connected lab. Another might use saved examples, printed transcripts and a small local server. A law class could examine a fabricated citation. An agriculture class could compare a generated crop recommendation with local extension guidance. A health student could practice refusing to enter patient data. A journalism student could trace an image or claim back to its source. The module should travel as a kit, not only as a website. That kit could include low-bandwidth materials, printable exercises, offline demonstrations, open examples, lecturer notes, accessibility guidance, assessment rubrics and a clear path for local adaptation. It should work on a shared laptop in a crowded classroom as well as on a personal device in a fast network. Otherwise the regional standard will quietly measure infrastructure privilege. Lecturer preparation may be the decisive constraint. The official plan recognizes this. It says the pilot will generate evidence about delivery approaches, institutional requirements, lecturer preparedness and learner engagement, while IUCEA will support institutional capacity development. That is more serious than dropping a slide deck into faculty inboxes. Still, preparation needs a definition. Lecturers do not all need to become model engineers. They do need enough confidence to demonstrate a failure, discuss data protection, connect the material to their subject and grade reasoning rather than prompt polish. They also need time to update examples as tools change. A train-the-trainer model could help, but only if support continues after the workshop. Office hours, peer communities, teaching observations, shared case libraries and small grants for course adaptation would matter more than a single launch certificate. Assessment deserves the same attention. A multiple-choice quiz can test vocabulary. It cannot show whether a student will catch a confident falsehood in a lab report, protect a community dataset or disclose that a model shaped an assignment. The strongest assessment would put students inside realistic dilemmas. Give them an AI-generated answer with a subtle error. Ask them to find and correct it. Give them a dataset containing sensitive details and ask what must be removed. Give them two tools with different access terms and ask which one fits the task. Ask them to explain a result to someone affected by it. Then publish how the pilot performed by discipline, institution type and delivery model. Regional averages can hide the students the program most needs to reach. Language also belongs in the design from the beginning. East African higher education operates across many languages and educational traditions. A course can use English in formal materials while still failing students whose strongest reasoning happens elsewhere. Local examples, multilingual discussion and terminology that lecturers can explain clearly are not decorative additions. They determine whether critical judgment is actually learned. The module should record where AI systems perform poorly in regional languages too. That is practical literacy. If a model is less reliable for a student's language, dialect or local reference, the course should make that limitation visible instead of treating the English-language demo as universal. The partner list adds both capacity and governance questions. The effort includes Germany and technical contributions involving GIZ, the Japan International Cooperation Agency, Alliance4AI and other regional and international experts. External partners can bring funding, research and implementation experience. The curriculum still has to remain answerable to East African institutions, educators and learners. That means publishing who owns the module, how revisions are approved, what teaching materials are licensed for reuse, which vendors influence examples and whether institutions can substitute tools without rebuilding the course. A foundational course should not become a customer-acquisition funnel wearing an academic lanyard. The official framing is explicitly local. IUCEA's acting executive secretary, Professor Idris A. Rai, said the goal is not simply to adopt material developed elsewhere, but to co-create a module that reflects East Africa's educational, technological and socioeconomic realities. The pilots will show whether that promise survives. The public evidence should begin with a simple readiness ledger. For each participating institution, publish the available power, connectivity, devices, class size, lecturer preparation time, accessibility support, languages, delivery format and local adaptations. Do not use the ledger to shame campuses. Use it to explain what support each site needs and which teaching methods work under which conditions. Then track more than enrollment. Measure who completes the module, who can demonstrate the target skills, how performance changes by discipline and language, where students need extra support and whether lecturers can deliver the course again without outside staff. Ask students whether the material helped them question systems, not only whether they enjoyed it. Follow graduates into internships and early work where possible. Did they disclose AI use appropriately? Did they protect data? Did they challenge bad outputs? Did employers notice better judgment? Those are harder questions than counting course seats. They are also the questions that make regional cooperation useful. A pilot in one well-resourced lab can show that the module is teachable there. A set of deliberately different pilots can show whether the foundation travels. Choose institutions with different disciplines, languages, geographies and technical conditions. Preserve the same core outcomes. Let delivery vary. Publish what breaks. Fund the repair. That would turn unequal classrooms from an awkward footnote into the central design test. East Africa does not need one frozen AI course copied across every campus. It needs a common promise: every graduate should leave with enough judgment to use AI without surrendering evidence, privacy or responsibility. The draft defines the direction. The signal will come from the first classrooms, especially the ones where the internet is slow, the examples need translation and the lecturer has to make the whole idea work with what is actually on the table.