School

We have learned what AI is, how it works, when to trust it, where its bias comes from, and how to ask it good questions. This lesson takes those five and puts them to work in the place where most Pakistani learners actually use AI: schoolwork. We are going to walk through five subjects and look at what AI can do well in each one, what it cannot do, and where the line between using AI and outsourcing your thinking sits. The line is the entire point. The first cost of being unable to do your own thinking is not exam marks. It is what happens to you when you arrive at university or at a job and the AI is not the one being assessed.

Mathematics. AI is good at explaining a concept two or three different ways. If you do not understand factorisation when the textbook explains it as area, ask AI to explain it through grouping, then through the distributive property, then through a real-world example. Three ways are better than one and AI gives them fast. AI is bad at correcting your work in detail. It will tell you the answer is wrong without locating where you made the slip. For that, work the problem out loud with a peer or a teacher. Use AI for the concept layer. Use a person for the diagnostic layer. The combination is faster than either one alone.

Science. AI can list the steps of an experiment. AI can describe the theory behind why the experiment works. AI cannot do the experiment for you, and the experiment is the point. The Pakistani science curriculum, especially at FBISE and PCTB level, is heavy on memorisation and light on practical work. The practical work is exactly where understanding lives. Use AI to prepare for the practical: predict what should happen, write down why, and only then do the experiment with whatever you can put together at home or in school. The gap between your prediction and what actually happened is the lesson. AI cannot give you that gap. Only doing it can.

English. AI is fluent in English. That fluency is not free for you. If you ask AI to write your English essay, you submit fluent English and you keep the level of English you walked in with. The exam in three years' time will not be in front of an AI. It will be in front of you. Use AI to identify your specific patterns. Paste a paragraph you wrote, ask the model to flag every place a native writer would phrase it differently, and study what changed. This is hard. It is also the path that produces a learner who can write English well. The shortcut produces a learner who depends on the shortcut.

Urdu. The model's Urdu is uneven. It can produce grammatical Urdu sentences. It will sometimes produce a turn of phrase that does not exist in any actual Pakistani register. It will sometimes mix Urdu of Lucknow with Urdu of Lahore in a way no real speaker would. This is not the model's fault and it is not yours. The training data of Urdu, online, is small relative to English. So the rule for Urdu schoolwork is reversed from English. Trust your own Urdu before trusting the model's. Use the model for translating English source material into a starting Urdu draft, then rewrite the draft in your own voice. The course returns to this in the lesson on language and identity.

Pakistan Studies. Here AI is at its weakest. Pakistan Studies asks for specific dates, specific people, specific quotes, specific local context. The model has read an English Wikipedia version of these and a small Urdu portion. It will mix dates, attribute quotes to the wrong person, and present a confident sentence that is partly wrong in three different ways at once. For this subject the rule is verify everything against your textbook before you write it down. AI is a brainstorming partner here, not a source. The information you submit must come from your textbook or a verifiable source. The model is the assistant. The textbook is the authority. Reverse this and you will lose marks and, more importantly, you will not learn the country you live in.