What AI Can and Cannot Do for a Pakistani Classroom

The claims are enormous and the useful applications are narrow and specific. Both things are true, and knowing which is which saves schools a lot of money.
There is a lot of noise about AI in education, and most of it is written by people selling something. The useful version of this conversation is narrower: for a teacher with 180 students and five sections, what actually helps on a Tuesday?
Where it genuinely helps a teacher
First drafts of routine material. Ten short questions on a topic, a worksheet, three versions of the same question with different numbers, a summary of a chapter written at a simpler reading level. These are tasks where the teacher's skill is in reviewing and adjusting, and the blank page was the slow part.
The output always needs checking. It is frequently almost right, which is more dangerous than being obviously wrong — a plausible but incorrect definition in a worksheet will be learned by thirty students.
Rephrasing for a struggling student. Asking for an explanation of the same concept three different ways, or in simpler language, is genuinely useful when the textbook's version has not landed and you have four minutes.
Administrative writing. Letters to parents, notices, routine reports. Real time saved, low risk.
Where it helps a student
Explaining something at the moment of confusion, at eleven at night, when no teacher is available. For a motivated student who is stuck on one step, this is the most valuable thing on the list.
Generating practice questions to attempt. Note the order: generate, attempt, then check. Reading an answer is not practice.
Checking a piece of written work for obvious errors before submitting it.
Where it fails, specifically
It is confidently wrong. In subjects where precision matters — a formula, a date, a definition that will be marked against a scheme — plausible-sounding errors are common. A student without the knowledge to detect them is the most likely to be misled.
It does not know your board. It will produce a competent general answer to a Physics question and not the answer that collects marks under your board's marking scheme, which may require specific wording, specific steps, a specific diagram.
It cannot see the student. Most of a teacher's judgement comes from noticing — who has gone quiet, who is pretending to follow, who is struggling for reasons unrelated to the subject. None of that is available to a system that only sees text.
It does not motivate anyone. The reason most students underperform is not lack of access to explanations.
Homework and honesty
Schools are discovering that work done at home is now unverifiable. The sensible responses are structural rather than technological.
Weight assessment towards work done in front of you. Ask students to explain their homework aloud — sixty seconds reveals everything. Set tasks that are harder to outsource: your own data, your own experiment, your own local example.
Detection tools are unreliable in both directions and accusing a student on their word is a serious thing to get wrong. Design around the problem instead of policing it.
Before your school spends money
Ask what problem it solves that you have actually got, who will maintain it, what happens to your students' data, and whether it works at the connection speeds your school really has.
Most schools would get more benefit from reliable attendance records and a shared question bank than from anything with AI in the name — because those solve problems they genuinely have, every day.
Comments
Leave a reply