Studio Aletheia Presents · Clarity by Design
An AI Faculty Learning Series
Responsible AI in Teaching, Assessment, and Academic Integrity
Course 03 of 08 · Policy Meets Pedagogy
This seminar supports faculty in navigating authorship, assignment design, transparency, and academic integrity in a way that protects learning goals and preserves professional judgment. The point is not enforcement by tool, it is clarity by design, communicated in language students can follow.
Guiding stance: policy follows pedagogy, not the reverse.
Core move: design assignments that reveal process, then set transparent disclosure expectations.
Success signal: fewer integrity disputes, consistent grading, preserved trust.
Designed for faculty and department chairs shaping course-level AI policy, built to support departments with differing norms while preserving institutional alignment.
Principles that keep policy learning-centered.
Clarity, fairness, autonomy, and learning-centered boundaries.
Learning Goals First
AI allowances and restrictions should map to the skill being assessed, not to the existence of a tool.
Transparency Over Suspicion
Disclosures and process evidence reduce conflict more effectively than detection or guesswork.
Consistency With Flexibility
Departments share a baseline language, while faculty retain discipline-specific choices and rationale.
Practical faculty decisions with clear student-facing language.
- Define authorship as a spectrum (idea, outline, draft, revision, final).
- Clarify what students must do themselves for the learning goal to be valid.
- Use discipline norms, not universal slogans, to define "ownership."
- Require process artifacts (planning notes, checkpoints, drafts, reflections).
- Use lived context, course readings, lab data, field observations, or unique prompts.
- Build oral defense, annotation, or decision logs into grading, proportionally.
Resilient Pattern
"Claim + evidence + why this evidence fits this claim (course-specific)."
Resilient Pattern
"Version history + revision rationale + what changed after feedback."
- Low-stakes disclosure: brief tool-use note, what it helped with.
- High-stakes disclosure: structured log (tool, purpose, prompt summary, edits made).
- Normalize disclosure as academic honesty, not as confession.
- Focus on the learning evidence in the work, not vibes about style.
- Use checkpoints and process requirements to reduce ambiguity.
- Adopt a restorative first response when boundaries were unclear.
- Map each assessment to the skill it measures, then decide what assistance is compatible.
- Write student-facing policy language in plain terms (allowed, limited, not allowed).
- Ensure policies support equity and do not punish students for access differences.
Pick a model that matches the stakes and learning goal.
Model A: Simple Statement
One sentence disclosure describing tool use and purpose. Best for low-stakes formative work.
Model B: Structured Log
Tool, purpose, prompt summary, and what the student changed. Best for medium-stakes submissions.
Model C: Process Evidence
Drafts, checkpoints, reflections, and brief defense. Best for high-stakes or capstone assessments.
Assignments that reveal learning even when tools exist.
Local Context Anchors
Require course-only readings, local data, lab results, or class discussions that AI cannot invent without being obvious.
Checkpointed Process
Break into staged submissions: proposal, outline, draft, revision notes, and final. Grade the process lightly but consistently.
Decision Logs
Students explain key choices (sources selected, claims removed, revisions made) and justify trade-offs.
Oral Defense Lite
Short, respectful check-ins that confirm understanding. Use randomly selected prompts or annotated passages.
Consistency, evidence, and a defensible process.
- State expectations in advance, in student language.
- Use assignment design evidence (checkpoints, drafts) rather than detection claims.
- Separate "policy violation" from "learning gap" and respond proportionally.
- Why not detectors: AI-detection tools carry meaningfully higher false-positive rates for multilingual and neurodivergent writers, treating a detector score as evidence risks a fairness problem bigger than the one it is meant to solve.
- Clarify: ask for process evidence and disclosure.
- Teach: correct misunderstanding of allowed use.
- Revise: allow a restorative resubmission when appropriate.
- Escalate: only when there is clear evidence and consistent policy language.
Check yourself against the guiding principles.
A short reflection, plus the self-check this course promised at the start.
One Assignment I Will Make More Resilient
Outcome Guarantee
One assignment. One expectation.
Name one assignment element you will make more AI-resilient this term, and one disclosure expectation you will state clearly to students. Small and specific beats broad and vague.