Module 2 Responsible AI for Teachers
A practical ethics module built on real staffroom situations. Teachers hunt for bias in live AI output, work out which classroom data is a child’s personal data and what the DPDP Act 2023 context means for their own habits, decide what appropriate student AI use looks like task by task, and draft three clauses that feed the school’s AI policy. Accountability stays with the teacher — this module shows what that looks like on a Tuesday.
By the end, teachers can…
- 1 Point to where bias could have entered a given AI output — the data it learned from, the way it was asked, or who was missing — and say which students it would be unfair to.
- 2 Sort the data in a real classroom task into personal, sensitive and not-personal, and ask the two questions before any of it goes into a tool: who can see this, and did the school and the parent agree to this use?
- 3 Describe in plain language what the DPDP Act 2023 context means for a teacher handling students’ data — anyone under 18 is a child, a child’s data needs a parent’s verifiable consent, data is used for the purpose it was collected — and name who in school settles the specifics.
- 4 Decide, for a given student AI scenario, whether it is appropriate use, misuse or a teaching moment, and say what they would say to the student.
- 5 Verify an AI-generated fact, answer or remark against a source before it reaches a student or parent, and explain why accountability does not transfer to the tool.
- 6 Draft three school AI-policy clauses — on student data, on student AI use, on teacher use and verification — specific enough that a colleague could tell whether they were followed.
Session plan — 2 hours
| Time | Activity | Format |
|---|---|---|
| 0:00 10 min | Opening: where did 7-B’s data go this week? Pairs list every place a student’s data travelled in the last seven days — the marks register, a class WhatsApp group, a photo on the school’s social media, a worksheet pasted into a chatbot. The board fills fast. That list is the module’s raw material. | Discussion |
| 0:10 15 min | Bias and fairness: how a model learns to be unfair Three doors bias walks through: the data a model was trained on, the way a question is asked, and who is missing from both. School examples: a “good essay” model trained on one kind of school’s essays; feedback that changes tone with a student’s name; an English-medium assumption in a Hindi-medium class. Fairness is not identical treatment — it is not disadvantaging a child for who they are. | Input |
| 0:25 25 min | Bias hunt Pairs run matched prompts in any chat assistant, changing one thing at a time: the same paragraph “by Priya” then “by Arjun”; “feedback for a weak student” then “for a bright student”; “describe a scientist / a nurse / a farmer”; a remark for a Hindi-medium then an English-medium student. They compare outputs line by line and record on the bias-hunt sheet what changed — tone, expectation, vocabulary, who was assumed — and what that would do to the child reading it. Runs that show no difference are recorded too. | Hands-on |
| 0:50 20 min | Student data and privacy: whose data is it? What counts as a child’s personal data in a classroom: a name with marks, a photograph, a health or family note, a phone number, anything that points to one child. Plain-language context from the Digital Personal Data Protection Act 2023: anyone under 18 is a child, a child’s data needs a parent’s verifiable consent, and data is used for the purpose it was collected. The teacher’s habits that follow: never paste a named student’s record into a public chatbot; de-identify; use the tools the school has approved; ask who can see it and who agreed. Habits, not legal advice — the school’s policy and its advisers settle specifics. | Input |
| 1:10 20 min | Integrity, misinformation and AI-generated answers Three case cards per group: a Class 8 student submits an AI-written English essay; a Class 10 student uses AI to check her maths working before a test; a Class 7 student quotes an AI “fact” about a freedom fighter that is wrong. Groups decide: appropriate use, misuse or a teaching moment — and what they would say to the student. Then they build an appropriate-use ladder for their own subject: allowed and declared, allowed for practice only, not for this task. | Group work |
| 1:30 10 min | Verify, then sign: teacher accountability The four-step verify habit before any AI output reaches a student or parent: check against a source, check against your own subject knowledge, read it as the child or parent will, ask “would I put my name under this?” If an AI remark, answer or fact goes out under the teacher’s name, the teacher owns it — the same professional standard, new tool. | Input |
| 1:40 20 min | Policy-drafting sprint Each group drafts three clauses for the school’s AI policy on the “We will / We will not” template: one on student data, one on student AI use, one on teacher use and verification. Two-minute sharebacks; duplicates merged on the board. The collated clauses go to the school’s named policy owner as teacher input to the first review of the draft policy from the Week-0 leadership workshop. | Group work |
| Total 2 h · timings are a default; facilitators adapt to the group. | ||
Key ideas
Bias lives in the data, not in anyone’s intent
A model learns what it was shown. If it was shown mostly essays from one kind of school, its sense of “good writing” is that school’s. Nobody meant it — and a child on the wrong side of it still gets colder feedback. The teacher’s job is to notice who an output is quietly unfair to, and to correct it before it lands.
A child’s data is not yours to paste
A student’s name, marks and notes belong to the child and the parents; the school holds them for a purpose. Pasting a named record into a public chatbot hands it to a company that had no part in that purpose. The habit: strip names, use the tools the school has approved, and ask who can see it and who agreed.
Integrity is about the learning, not the tool
The useful question is never “did the student use AI?” but “did the learning happen, and is the claim honest?” Set the rule per task — drafting an outline is different from submitting an essay as one’s own — say it before the task, and design tasks where AI use is either declared or unnecessary.
Verify, then sign
Fluent is not true. Before an AI-drafted fact, answer or report-card remark reaches a child or a parent, check it against a source and your own knowledge, and read it as they will. If you would not put your name under it, do not pass it on. Accountability stays with the teacher.
Worked example
The marks sheet that went to a chatbot
A Class 7 Social Science teacher has forty report-card remarks to write by Friday. She pastes her 7-B marks sheet — names, marks, and her private column with notes like “Rehan — slow reader, father unwell” — into a free chatbot and asks for “a warm remark for each student”.
What came back: forty fluent, kind, usable remarks. What also happened: a named child’s family and health note now sits on a company’s servers, for a purpose nobody in school or at home agreed to. Nothing felt wrong at the time, because the output was good.
What she could do instead, in about the same time: replace names with roll numbers or initials before pasting; write the sensitive cases herself; use the tool the school has approved for exactly this; read every remark as Rehan’s father will read it before it goes on the card.
The remarks were fine. The habit was not. That is what responsible use looks like in practice — not a refusal to use AI, but knowing what must never leave the room, and checking what comes back.
Assignment — counts towards certification
Responsible-use reflection
Over one week, watch your own AI use for the moments this module was about. Refine the three policy clauses your group drafted; record one real situation from your week where AI use touched student data, fairness or integrity — what happened, what you did, what you would do now; and show one AI output you verified before using, with the check you made.
Deliverable: One page: three “We will / We will not” policy clauses (student data, student AI use, teacher use and verification); a short reflection (about 150 words) on one real situation; and one verified AI output with the source named. This is the Level 1 responsible-use reflection.
| Criterion | Meets the standard when… |
|---|---|
| Three usable policy clauses | Each clause is specific enough that a colleague could tell whether it was followed; at least one is about student data and names what must not leave the school’s approved tools. |
| A real situation | The reflection describes something that actually happened that week (not a hypothetical), with what the teacher did and what they would do now, and names which principle it touched — data, fairness or integrity. |
| Bias or fairness noticed | At least one concrete observation — from the bias hunt or the teacher’s own use — of an AI output treating students or groups differently, and who it would be unfair to. |
| Verification shown | One AI output is shown with the check made against a named source or the teacher’s subject knowledge, and what was kept, changed or discarded. |
Where this lands in the PrepGraph workflow
Each certification task corresponds to a behaviour in the teacher portal — so the programme produces a working habit, not a slide deck. The pedagogy comes first; the surface implements it.
| Certification task | Pedagogical move | PrepGraph surface |
|---|---|---|
| Ask who an AI signal could be unfair to before acting on it — an “engagement drop” may be a child without a phone at home | Fairness review of an AI signal | Early-warning signalsNightly signals per group — engagement dropping, scores slipping — so you know who needs a check-in this week. |
Materials
- •Bias-hunt sheet: matched prompt pairs and a recording grid (printable, facilitator pack)
- •Any chat assistant on a phone or laptop, one per pair
- •Three integrity case cards per group (essay, maths check, wrong “fact”)
- •“Whose data is it?” one-pager — plain-language DPDP Act 2023 context for teachers, marked “not legal advice”
- •“We will / We will not” policy-clause template, one per group, plus a board to collate
- •The school’s existing acceptable-use, device or photography policy, if one exists
Facilitator notes
- •In the bias hunt, insist that pairs change one thing at a time — the name, the label, the medium — and keep everything else identical; otherwise they cannot tell what caused the difference. Some runs will show no difference. That is a finding too; have them record it.
- •Be plain about the DPDP Act 2023: teachers need habits, not section numbers and penalties. Say what it is about — children under 18, parental consent, purpose — and send every “is X legal?” to the school’s policy and its advisers. Reframe as “what would you need to know before doing X, and who in school decides?”
- •The integrity discussion runs hot. Steer it from catching students to designing tasks — where AI use is either allowed and declared, or unnecessary. The appropriate-use ladder is the takeaway.
- •The policy clauses are real input, not an exercise: collect them and hand them to the school’s policy owner — the person named in the Week-0 leadership workshop — for the first review of the draft policy. Tell teachers that at the start — the writing gets sharper.
- •Do not demonstrate PrepGraph. If asked how PrepGraph handles student data, say the school’s agreement and the leadership workshop cover it; the habits taught here apply to every tool, including ours.
Bring the programme to your school
Tell us your board, grades and teacher count. We’ll reply within two working days with a cohort plan and the CPD record templates.