What AI Can and Cannot Do in a Classroom

Classroom scenario

Fictional. Two teachers try the same experiment. One asks an AI tool for "a Class 7 science quiz" and receives ten fluent questions — two of which turn out to be factually wrong and one far above grade level. The other asks for the same quiz, checks every answer against the textbook, fixes the errors, and uses it happily. Same tool, very different outcomes. The difference was never the AI — it was the teacher's process.

Core explanation

Fact. Generative AI produces new text by predicting likely words from patterns in its training data. It does not look up verified facts, know your syllabus edition, or understand your class. Three consequences follow:

• Fluency is not truth. Confident, well-written output can be wrong — these errors are called hallucinations, and nothing in the tone warns you.
• It has gaps. A model may not know recent syllabus changes, your board's exact wording, or your local context — while sounding authoritative.
• It is a drafting assistant, not a source of truth. The expertise that checks the draft is yours.

What AI genuinely helps with: first drafts of lesson plans, worksheets, explanations at a chosen reading level, question banks to edit, translations to review, and routine non-sensitive paperwork.

What it cannot do: know your students, guarantee factual accuracy, make pedagogical judgements, or carry accountability. Decisions about children — marks, promotion, discipline, wellbeing — must never be delegated to it.

Weak vs improved

Weak prompt: "Make a Class 7 science quiz."
Problems: no topic, board, difficulty spread, or answer key — and no verification step planned.

Improved thinking (previewing SATHI): "CBSE Class 7 Science, chapter Nutrition in Plants (S). Mixed-ability class, simple English (A). 8-question quiz — 5 recall, 3 application — with an answer key I will verify (T). Use only textbook-level facts; I will check every answer (H). Then suggest one improvement to your own draft (I)."

Guided practice

Take one task you did this week (a worksheet, a notice, an explanation). Write one sentence: what could AI have drafted for you, and what would you still have had to check yourself?

Knowledge check

Q. An AI tool writes a confident, fluent paragraph about a science topic. What does the fluency tell you about its accuracy?

• (a) Fluent output is usually accurate.
• (b) Nothing — fluency and accuracy are unrelated; every factual claim still needs teacher verification. ✅
• (c) The tool would warn you if it were unsure.

Feedback. Correct: (b). Generative AI predicts likely words; it can be confidently wrong (a hallucination), and it does not reliably warn you. Verification is the teacher's job — this single habit underpins the whole course.

Responsible AI checkpoint

AI may draft; you verify, adapt and approve. Nothing goes in front of students without your review.

Key takeaways

• Generative AI predicts text; it does not know or verify facts.
• Fluency ≠ truth — expect and catch hallucinations.
• Use it for drafts of non-sensitive work; never for decisions about children.

Glossary

• Generative AI: software that creates new content (text, images) from patterns in training data.
• Hallucination: confident, fluent output that is factually wrong.

References / source notes

• Framing aligned in principle with the UNESCO AI Competency Framework for Teachers (human-centred, teacher-controlled use). Simplified for a foundation audience. Last reviewed: at authoring.