Chew and Pour, Now with AI: Ghana, LLMs and the Fight for Depth of Knowledge
Every Ghanaian who went through our school system knows the phrase chew and pour. Read it, memorise it, pour it onto the exam paper, forget it by Monday. It was always the cheap tyre track our education left on most of us, and the thing every good teacher quietly fought against. The honest question of this moment is not whether AI is coming to Ghanaian classrooms; it is already here, in every student’s pocket, on WhatsApp bundles funded by their parents’ data plan. The question is whether AI makes chew and pour final, or whether it becomes the first tool in years that genuinely pushes children up past it.
What the machines are currently used for
The way large language models are used today, in most hands, is as the world’s fastest answer dispenser. Ask, receive, copy. A child who previously would have hunted textbooks at dusk after school for the model answer now gets something better than any model answer: an answer paraphrased to their level, delivered in seconds, with the tone of a kind uncle.
Notice what this actually changes. The old chew and pour at least required the chewing: some memory effort, some ownership of the material. The answer machine removes even that. Education researchers describe work at different levels of depth: recall at the bottom, then applying a skill, then reasoning through a problem, and at the top, sustained independent work. Teachers call these the Depth of Knowledge levels. The uncomfortable observation from my own classrooms at Mikrobot Academy, and in assessment papers generally, is that we were never strong above the recall level to begin with. An entire generation’s private study was concentrated at DOK 1; the machines have arrived precisely there, to do that bottom level for us.
The result is a drop that looks paradoxical: more access to information than any generation before them, less thinking per assignment than we did at their age. Not because children became lazier. Because the flexible machine does the part that was never the hard part of learning, and the hard part, the thinking, was already the part our system skipped most often.
What knowing how LLMs are used should teach us
Here is the twist worth sitting with: the same machine that flattens thinking in one configuration lifts it in another. It depends entirely on what you ask the AI’s role to be.
A plain assistant, the default everyone defaults to, is an answer dispenser. But you can configure the same model to be something else: an examiner, a scenario-setter, a coach that refuses to answer. That single difference is the whole ballgame, and it is the thing our education debate is missing. Banning the tool changes nothing (the phone is already in the pocket); redesigning the tool’s role in the room changes everything.
I have the receipts on this small truth, and I will share them honestly. In my own work building a Socratic tutor, a system that forces a student to restate the problem, explain what they know, what they tried, and where exactly they are stuck before it offers any help, I ran a deliberately adversarial test with a small local model: forty attempts per setup, four setups total, nothing at scale, my own equipment, my own rules of engagement. The plain assistant leaked the direct answer to the student behind the keyboard 9 times out of 10. The tutor that had to interrogate before it would help leaked zero. Two small numbers, not a research paper, but they carry the argument: the difference between answer-machine and thinking-coach is not the model’s intelligence; it is the design of the conversation around it. And a design is something a teacher, not a Silicon Valley firm and not even a decree from Accra, controls.
Teaching a child to think, rung by rung
If the machine must not hand over the answer, then its job at each level of depth is definite, and none of these jobs require the internet to be always on.
At the recall rung, the AI works as an examiner: quick questions, one at a time, checking the child against the reference, then drilling what was missed. At the application rung it works as a scenario-setter: it paints a concrete situation (a farmer, one control board, six water pumps) and waits for the child to commit to an answer and the reasoning behind it before it says anything. At the strategic-thinking rung it becomes the Socratic coach: it knows the correct path and asks the next-best question instead of giving the next step, refusing shortcuts politely and indefinitely. And at the top rung, the project where a child plans, builds, fails, reflects, the AI works as the critic on the review panel, demanding the three hardest questions about the child’s design weekly, never supplying the designs.
You will notice the pattern: at every rung, the child’s output is the reasoning. The machine’s output is a question. Swap that order and you get cheating with extra steps; keep that order and you get something our system has rarely managed to produce affordably at scale: an indefinitely patient examiner for every child, in a country where good teachers are scarce and class sizes are 40-plus.
The practical asks for Ghana’s education system therefore do not begin with banning chat apps. They begin with three moves. Redesign assessment so the reasoning, not the artifact, is what earns the marks: spoken defences, in-class builds, process logs. Train teachers in the coach role, because the same phone a teacher curses at during homework can become the drill-sergeant and the scenario machine in their lesson plan. Write the AI policy around roles, not devices. And above all, decide deliberately that we are aiming assessments and assignments at depth three and four, because whatever we aim at, the machine will follow the shortest path to it and our children will follow the machine.
The yardstick
I keep one yardstick for all of this: who is doing the thinking in this room right now? If the AI writes and the child submits, we have industrialised chew and pour and called it digital transformation. If the AI questions and the child reasons, even clumsily, even slowly, then the cheapest educational revolution available to us is already sitting in every student’s pocket, waiting to be pointed the right way.
Ghana has taken kids almost exactly where they needed to be on phones, payments and national ID in ways that skipped steps richer countries considered mandatory. Skipping the answer-machine stage of AI, jumping straight to the coaching stage, is exactly the same kind of leap. The tool is generic; the role we assign it is the choice that is still ours.
By John Awotwi, CSIR-INSTI