What is AI in education? A plain-language explainer for Pakistani teachers and principals

"AI in education" is now said in staff meetings, parent chats, and ministry announcements as though everyone agrees what it means. Most people don't, and that gap causes both needless fear and needless trust. This is a plain explanation for teachers and principals: what the term covers, how the tools you'll actually meet produce their answers, where they're used in schools, and where they reliably go wrong.

Start with the words. Artificial intelligence is a broad label for software that performs tasks we'd normally associate with human thinking, such as recognising speech, translating, recommending, or writing. Machine learning is the main way modern AI is built: instead of being given rules, the software is shown huge numbers of examples and learns patterns. Generative AI is the kind that produces new content, such as text, images, or audio. And a large language model, or LLM, is the generative system behind chatbots and most classroom writing tools. When someone says "we use AI to make lesson plans," they almost always mean an LLM.

How does an LLM produce an answer? Very simply, it has learned from vast amounts of text which words tend to follow which, and it builds a response one piece at a time by predicting what fits. It is not looking up a fact in a database, and it doesn't "know" your syllabus. That explains both its strengths and its failures. It writes fluent, well-organised text quickly, which is why it's good at drafting. And because it optimises for what sounds right, it can state something false with complete confidence, which is why every output needs a human check. It also reflects the text it was trained on: UNESCO has noted that current models are trained on data that reflects the values and dominant norms of the Global North, which is one reason examples can feel foreign.

Where does AI show up in education? Four places cover most of it. First, teacher-side content: drafting lesson plans, worksheets, quizzes, and parent messages. This is the lowest-risk and most common use, because a teacher reviews the result before students see it. Second, tutoring and student-facing help: chatbots that answer students' questions or explain a concept. This raises more difficult questions about accuracy, dependence, and age. Third, assessment and feedback: tools that mark or comment on student work, where errors and fairness matter most. Fourth, school administration: scheduling, reporting, and communication.

In Pakistan, all four are beginning to appear, unevenly. A UNESCO programme with Pakistan's National Commission for UNESCO trained 150 teachers, 25 head teachers, and 25 education coordinators from federal schools between September and December 2025 in practical AI tools. In Khyber Pakhtunkhwa, the chief minister gave in-principle approval to a first AI Education Authority, according to Business Recorder, alongside a pilot of virtual schools in 46 government schools and an AI Teacher system covering English, Mathematics, Physics, Chemistry, and Biology in Urdu, English, and Pashto. These are pilots and early programmes, not the state of an average classroom, so a teacher shouldn't assume any of them has reached their school.

What can these tools not do? They can't replace a teacher's judgment about a particular child, a class's mood, or what will land on a Tuesday afternoon. They can't guarantee accuracy, especially in Urdu; researchers who built UrduFactBench and UrduFactQA point out that fact-checking resources are mostly developed for English, and found persistent challenges for open-source models' factual reliability in Urdu. They can't protect privacy on their own, which is why what you type matters; see our post on AI data privacy. And they can't tell you whether a piece of writing came from a student or a machine, however confident a detector looks.

Given all that, how should a school start? Begin on the teacher side, where a human review sits between the AI and the students. Set a short policy: no student-identifying data in general tools, every draft reviewed by a teacher, and clear disclosure. Try one tool on one real task, such as next week's worksheet, and measure whether it saved time and whether the result was good. Only then consider anything student-facing, and check the age guidance: UNESCO's guidance on generative AI in schools suggests an age limit of 13 for using AI tools in the classroom.

Muallim is a small, concrete example of teacher-side generative AI. It's built by DIGIT Pakistan for teachers and school admins, not students, and uses Google's Gemini API, as our Privacy Policy states, to generate lesson plans, worksheets, and quizzes from the class, subject, and topic you provide. The Lesson Planner, the Worksheet Builder, and Quiz & Assessment are scoped to your board and produce Urdu, Roman Urdu, or English, and each result opens in an editor before export. The LLM is doing the drafting; the teacher is doing the deciding. That division of labour, more than any feature, is what makes the use responsible.

If one thing sticks, let it be this: AI in education isn't a single thing, and the useful question is always about a specific use. Who is holding the tool, what goes in, what comes out, and who checks it? A school that can answer those four questions for every AI use it allows, in a sentence each, is in a stronger position than one that has bought the most impressive product. Start there, and revisit it as the tools change.