Decision Desk · Issue 11 · complete public learning edition

Should AI Review a Student’s Work?

Decide when AI may support feedback, assessment, personalization, or intervention without replacing the learning purpose, educator responsibility, equitable access, or a learner’s right to understand and challenge.

Begin the issue Run the learner review

Learning time90 minutes

Teaching12 cumulative lessons

InstrumentLearner-impact review

BoundaryEducator remains responsible

Learning chain

Purpose → evidence → access → review → challenge → learning.

An educational AI use succeeds only when it improves a defined learning condition without degrading validity, access, agency, trust, privacy, or accountable human judgment.

Define the learner benefit and non-AI alternative

Gate

Bound the function, stakes, work, and data

Gate

Test validity, bias, access, and failure

Gate

Keep informed educator review before consequence

Gate

Give learners explanation, choice, and challenge

Gate

Monitor learning, relationships, and unequal effects

Gate

Complete issue curriculum

Twelve cumulative lessons.

Use supplied fictional scenarios only. Never enter student work, records, grades, disability information, identifiers, or confidential material.

Issue 11

Define the learning purpose

Open

Name the learning objective, instructional rationale, intended benefit, evidence required, and non-AI alternative. Saving educator time is not by itself a learner benefit.

Issue 11

Separate support from consequential judgment

Open

Brainstorming, practice, formative feedback, grading, placement, intervention, discipline, and access decisions carry different stakes. Keep high-stakes decisions human-owned.

Issue 11

Protect student work and educational data

Open

Map prompts, submissions, drafts, feedback, scores, metadata, disability information, inferred traits, retention, training/reuse, provider access, deletion, and downstream records.

Issue 11

Provide understandable notice and meaningful choice

Open

Explain what AI does, what it sees, how outputs are used, limitations, who reviews, retention, challenge routes, and a genuinely usable nonpunitive alternative.

Issue 11

Preserve disability, language, and device access

Open

Test keyboard and assistive technology, alternative formats, language meaning, bandwidth/device constraints, accommodations, processing differences, and whether AI use creates a new barrier.

Issue 11

Test construct validity

Open

Ask whether the system evaluates the intended knowledge or skill rather than writing style, dialect, language proficiency, disability, formatting, AI familiarity, or irrelevant proxies.

Issue 11

Examine bias and subgroup performance

Open

Use representative evidence to inspect error, feedback quality, false flags, opportunities, and burdens across relevant learner groups without treating aggregate performance as sufficient.

Issue 11

Keep the educator responsible

Open

Educators need source work, rubric, AI output, uncertainty, rationale, and authority to disregard or correct it. Rubber-stamping an output is not meaningful human review.

Issue 11

Design feedback that supports learning

Open

Feedback should be timely, specific, developmentally appropriate, actionable, aligned to instruction, transparent about uncertainty, and unable to quietly rewrite a learner’s voice or do the work for them.

Issue 11

Make challenge and correction real

Open

Learners need a safe way to understand, question, correct, and appeal AI-influenced feedback or decisions without retaliation, technical expertise, or proving that the model failed.

Issue 11

Monitor learning and relational effects

Open

Track learning outcomes, subgroup differences, educator workload, student agency, trust, overreliance, chilling effects, complaints, overrides, false accusations, provider changes, and unintended displacement.

Issue 11

Issue a learner-specific disposition

Open

Do not use, keep educator-only, repair conditions, or evaluate one low-stakes educator-reviewed use. No decision transfers across learners, courses, assignments, stakes, models, or terms.

Critical stop conditions

The learner is not a data point.

the learning objective, instructional rationale, stakes, or non-AI alternative is unresolved
AI output determines a grade, placement, discipline, access, or intervention without accountable educator judgment
student work or educational data authority, retention, training/reuse, access, or deletion is unclear
notice is hidden or the alternative is burdensome, stigmatizing, costly, or academically punitive
disability, language, device, accommodation, or subgroup performance evidence is missing
the tool measures irrelevant proxies rather than the intended knowledge or skill
the educator cannot inspect source work, rationale, uncertainty, and errors before use
learners cannot understand, challenge, correct, appeal, and receive timely human support

Interactive instrument · device-local

Learner-Affecting AI Use Review

Use fictional learning scenarios only. Do not enter student work, educational records, disability information, identifiers, grades, credentials, or confidential material.

0/30learner-impact gates evaluated

Efficiency is not a learning outcome.
AI may support instruction only when the learning purpose, evidence validity, access, educator responsibility, and learner challenge path remain intact.

Fictional educational use	Learning purpose	Work/data authority	Access equitable	Educator reviews	Student can challenge	Impact monitored
Formative feedback	Formative feedback: Learning purpose	Formative feedback: Work/data authority	Formative feedback: Access equitable	Formative feedback: Educator reviews	Formative feedback: Student can challenge	Formative feedback: Impact monitored
Writing or problem analysis	Writing or problem analysis: Learning purpose	Writing or problem analysis: Work/data authority	Writing or problem analysis: Access equitable	Writing or problem analysis: Educator reviews	Writing or problem analysis: Student can challenge	Writing or problem analysis: Impact monitored
Personalized practice	Personalized practice: Learning purpose	Personalized practice: Work/data authority	Personalized practice: Access equitable	Personalized practice: Educator reviews	Personalized practice: Student can challenge	Personalized practice: Impact monitored
Rubric-supported assessment	Rubric-supported assessment: Learning purpose	Rubric-supported assessment: Work/data authority	Rubric-supported assessment: Access equitable	Rubric-supported assessment: Educator reviews	Rubric-supported assessment: Student can challenge	Rubric-supported assessment: Impact monitored
Progress or intervention signal	Progress or intervention signal: Learning purpose	Progress or intervention signal: Work/data authority	Progress or intervention signal: Access equitable	Progress or intervention signal: Educator reviews	Progress or intervention signal: Student can challenge	Progress or intervention signal: Impact monitored

Learners, learning purpose, stakes, and alternativesName learners, objective, instructional theory, stakes, current practice, non-AI alternatives, affected groups, and what the tool must never determine.

Student work, data authority, notice, and choiceRecord inputs, authority, minimization, retention, training/reuse, access, notice, meaningful choice, age/guardian considerations, deletion, and nonpunitive alternative.

Access, validity, educator review, and challengeRecord disability/language/device access, construct validity, bias evidence, educator review, source visibility, correction, learner explanation, contest, appeal, and support.

Monitoring, correction, support, and stop conditionsTrack learning outcomes, subgroup differences, overreliance, false flags, workload, complaints, overrides, provider changes, incident response, and stop owner.

Bounded dispositionDo not use AI for this learner-affecting functionKeep the function educator-onlyRepair learning, data, or access conditions firstEvaluate one low-stakes, educator-reviewed use

Restoring your local review…

Completion boundary

The educational product and reusable Learner-Affecting AI Use Review are complete.

Real use still requires current institutional and legal authority, student and affected-party participation, product/configuration evidence, privacy/security/accessibility review, validity and subgroup testing, qualified educator judgment, meaningful alternatives and challenge, support and incident readiness, monitoring, and organizational approval.

Return to the 12-month program Previous issue