Module 1 of 4
Intro to AI Ethics for the Classroom
By the end of this module you'll be able to name the four ethical questions to ask before using any AI tool with students, and spot the most common ways AI goes wrong in education.
Why ethics comes first
AI tools now draft lessons, grade essays, flag "at-risk" students, and answer homework questions at 2 a.m. Each of those uses touches something educators are trusted to protect: fair treatment, student privacy, and honest assessment of learning.
Ethics in this context isn't a philosophy seminar. It's a short set of questions you ask before a tool touches your students, so problems get caught in planning instead of in a parent meeting.
The four questions to ask about any AI tool
- Fairness: Could this tool treat some students worse than others? AI systems learn from past data, and past data carries past bias. Automated essay scorers have penalized non-native English writing patterns, and remote proctoring software has produced higher false-flag rates for students with darker skin tones and students with disabilities.
- Privacy: What student data goes in, and where does it go? In the U.S., FERPA governs education records — pasting a named student's grades, IEP details, or disciplinary history into a consumer chatbot can be a disclosure violation. Rule of thumb: de-identify before you paste.
- Transparency: Would I be comfortable explaining this use to a student or parent? If a grade, intervention, or opportunity was shaped by AI, the people affected deserve to know.
- Accountability: Who owns the outcome? An AI can draft, suggest, and flag — but a qualified human makes the call. This is often summarized as keeping a human in the loop.
Where AI most often goes wrong in schools
- Hallucination: generative AI produces fluent, confident text that can be factually wrong — invented citations, fake statistics, misattributed quotes. Fluency is not accuracy.
- Bias amplification: a model trained mostly on one demographic's writing or behavior quietly sets that group as "normal" and scores everyone else against it.
- Automation bias: the human tendency to over-trust a computer's output. The risk isn't just bad AI answers — it's tired humans accepting them without checking.
- Opaque decisions: if a tool can't explain why it flagged a student, you can't verify the flag or defend the decision.
Module 1 Quiz
4 questions. Pick the best answer, then select "Check my answers." You can retry as many times as you like.
1. A teacher pastes a full class roster with names and reading scores into a free public chatbot to "find struggling students." What's the primary ethical problem?
2. An AI essay scorer consistently gives lower scores to students who write in dialects other than standard academic English. This is best described as:
3. "Human in the loop" means:
4. Which use of AI carries the LOWEST ethical risk, as-is?
Module 2 of 4
Using AI for Teaching
By the end of this module you'll have a repeatable prompt structure, know which teaching tasks AI genuinely speeds up, and know what always needs your verification.
Where AI earns its keep
The best classroom uses share two traits: they're time-consuming for you but low-stakes if the first draft is imperfect, because you review everything before students see it.
- Lesson planning: generate a first-draft outline, warm-up, or exit ticket, then adapt it to your students.
- Differentiation: "Rewrite this passage at a 4th-grade reading level" or "Create a scaffolded version of this problem set" — minutes instead of an evening.
- Feedback drafting: paste a de-identified student paragraph and your rubric; ask for two strengths and one next step in an encouraging tone. You edit and personalize before it goes out.
- Assessment variety: turn one concept into a multiple-choice item, a short-answer prompt, and a real-world scenario.
- Parent & admin communication: first drafts of newsletters, reminder emails, and translated versions for multilingual families (verify translations with a fluent reader when stakes are high).
A prompt structure that works: R-C-T-F
Vague prompts get generic answers. Strong prompts have four parts:
- Role — who the AI should act as: "You are an experienced 7th-grade science teacher."
- Context — the situation and constraints: "My students just finished a unit on photosynthesis; several are English learners; I have 45 minutes."
- Task — the specific ask: "Create a review activity with tiered difficulty."
- Format — the shape of the output: "Give me a table: activity step, time, materials, and one check-for-understanding question."
What you must always verify
- Facts, dates, and statistics — models fabricate confidently. Check anything you didn't already know.
- Citations and quotes — AI invents plausible-looking references. Never pass one to students unchecked.
- Alignment to your standards — AI doesn't know your state framework or your students; it approximates.
- Tone in sensitive communication — behavior emails, grade disputes, and IEP-adjacent messages need your voice and your judgment.
Module 2 Quiz
4 questions. Pick the best answer, then select "Check my answers."
1. Which prompt is most likely to get a useful result?
2. An AI-generated worksheet includes the claim "The Great Barrier Reef was declared dead in 2016," with a citation to a marine biology journal. What should the teacher do?
3. Which task is the BEST fit for AI assistance under the "time-consuming for you, low-stakes if imperfect" rule?
4. The first output from an AI should be treated as:
Module 3 of 4
Guiding Students When They Use AI
By the end of this module you'll be able to set assignment-level AI expectations students actually understand, teach the three habits of responsible AI use, and handle integrity concerns without relying on unreliable detectors.
Start from reality: students are already using it
The question isn't whether students will use AI — most already do. The question is whether they'll use it secretly and passively, or openly and skillfully. A blanket ban mostly teaches concealment. Clear, assignment-specific expectations teach judgment.
The traffic-light model: set expectations per assignment
One classroom policy can't fit every task, because different assignments assess different skills. Label each assignment:
| Level | Meaning | Example |
|---|---|---|
| Red | No AI. This task measures what you can do unaided. | In-class essay assessing writing fluency; math skills check |
| Yellow | AI allowed for specific steps, with disclosure. | Brainstorm or outline with AI, write the draft yourself, note how AI was used |
| Green | AI encouraged; the skill being graded is your process and judgment. | Use AI to critique your draft, then submit the revision plus a reflection on which suggestions you accepted and why |
Teach the three habits: verify, disclose, own it
- Verify: AI output is a claim, not a fact. Students check important claims against real sources before using them — the same skill as evaluating any website.
- Disclose: normalize a one-line AI-use note ("I used AI to outline and to check my grammar"). Disclosure kills the secrecy that turns AI use into cheating.
- Own it: whatever a student submits, they're accountable for — including AI's errors. "The AI said so" is never a defense, which is a powerful reason to verify.
About AI detectors — an honest note
AI-writing detectors are not reliable enough to serve as proof. They produce false positives — flagging genuine student work — and research has shown they disproportionately flag writing by non-native English speakers. Several major detection tools' own guidance says scores shouldn't be the sole basis for action.
Stronger approaches: compare the work to the student's known writing and in-class output, ask the student to walk you through their thinking or revise a section live, and design assessments where the process is visible — drafts, version history, in-class components, oral defenses.
Module 3 Quiz
4 questions. Pick the best answer, then select "Check my answers."
1. In the traffic-light model, a "yellow" assignment means:
2. An AI detector flags a student's essay at "92% AI-generated." The student says they wrote it themselves. The best next step is:
3. Why does requiring a short AI-use disclosure note help academic integrity?
4. Which assignment design makes learning MOST visible in an AI-saturated world?
Module 4 of 4
Tracking Student Metrics for Learning
By the end of this module you'll know which metrics actually indicate learning, how AI can help you spot patterns early, and the guardrails that keep student data safe and decisions humane.
Measure learning, not just activity
Dashboards make it easy to collect numbers that feel informative but aren't. The key distinction:
- Activity metrics — logins, minutes on task, pages viewed, videos watched. Easy to count, weakly tied to learning. A student can be logged in for an hour and learn nothing.
- Learning metrics — mastery of specific skills, growth between assessments, quality of work over time, ability to transfer a skill to a new problem.
- Leading indicators — early signals that predict trouble before a grade drops: missed submissions, a sudden drop in engagement, declining quiz scores on prerequisite skills, reduced participation.
Where AI genuinely helps with data
- Pattern-spotting at scale: "Which quiz questions did most of the class miss, and what concept do they share?" AI can cluster errors into a reteach list in seconds.
- Early-warning triage: flagging students whose leading indicators are slipping so outreach happens in week 3, not after the failing grade posts. The flag starts a human conversation — it is never the verdict.
- Narrative summaries: turning a semester of de-identified data into a plain-language summary for your own planning or a grade-level meeting.
- Item analysis: identifying quiz questions that even your strongest students miss — often a sign the question, not the students, is the problem.
Guardrails for student data
- Minimize: collect and share only the data needed for the decision at hand. De-identify before using any tool that isn't district-approved.
- Context before conclusions: a flagged student may be caring for a sibling, working nights, or switching schools. Data starts the conversation; the student's story completes it.
- Beware self-fulfilling labels: an "at-risk" tag can quietly lower expectations. Frame flags as "needs outreach now," not "unlikely to succeed."
- Close the loop: a metric only matters if it changes an action — a reteach, a check-in, a redesigned question. Track whether the intervention worked, too.
Module 4 Quiz
4 questions. Pick the best answer, then select "Check my answers."
1. Which of these is a LEARNING metric rather than an activity metric?
2. An AI early-warning system flags Marcus as "at risk" based on two missed assignments and lower login frequency. The right response is:
3. Most of the class — including your strongest students — missed question 7 on the unit quiz. The most likely explanation to investigate FIRST is:
4. "Closing the loop" on learning data means:
Wrap-up
Your results
Complete all four knowledge checks to see your overall score and your take-home toolkit.
Course in progress
Finish the four module quizzes to unlock your results. Your progress shows in the outline on this page.
- F-P-T-A: Fairness, Privacy, Transparency, Accountability — the four-question ethics screen for any AI tool.
- R-C-T-F prompts: Role, Context, Task, Format — and treat every first output as a rough draft.
- Traffic-light assignments: red / yellow / green AI expectations, taught with verify–disclose–own it.
- Process over detectors: visible drafts and conversations beat unreliable AI-detection scores.
- Act on learning metrics: growth and leading indicators over login counts — and every flag ends in a human conversation.