Ethics & data
An educational use of AI is also a decision about other people’s information, opportunities and ability to question a judgment. Examine those decisions alongside the proposed learning benefit.
UNESCO’s education guidance places human agency and ethical suitability at the center of institutional evaluation. NIST’s voluntary, cross-sector profile identifies privacy, harmful bias, intellectual property and human–AI arrangements among the areas to examine. These are guidance frameworks. They do not establish that a product complies with a particular institution’s requirements or improves psychology learning.
Follow the data through the whole task
Start with a proposed activity rather than a provider’s general privacy statement. For a feedback exercise, distinguish what the student submits, what the instructor supplies, what a connected tool retrieves and what is recorded afterward. Ask separately about retention, access and further use. A setting about model training does not, by itself, answer every question about the rest of this map.
The following is an original planning aid, not an audit of a service. Complete it for the account, application and connected tools actually being considered.
| Stage | Possible material | Question to resolve |
|---|---|---|
| Input | A question, submitted paragraph or attached file | What is necessary for this learning purpose? Can invented material answer the question? |
| Retrieval and tools | Course readings, a connected folder or an external search | Which sources can be accessed, and which service receives a request? |
| Output and decisions | Feedback, a summary or a proposed assessment judgment | Who can see, correct, contest or release the result? |
| Logs and memory | Conversation history, tool records or retained preferences | What is retained, who can inspect it, and when is it reviewed? |
| Further use | An evaluation dataset, model improvement or another teaching task | Is a new purpose involved? Which settings and institutional arrangements actually govern it? |
| Closure | End of the activity, account change or a deletion request | Which copies and derived records are covered, and who can verify the outcome? |
For a psychology lesson, an invented response may be enough to examine a misconception without supplying real student work. If real work is proposed, identify why it is needed. Do not treat removing a name as a completed assessment of whether the remaining information could identify someone. Avoid asking students to supply personal health histories to make an exercise feel authentic.
Consent, course requirements and institutional authority
Consider a student asked to accept a service’s terms to complete a graded assignment. What could they reasonably decline? Is the alternative comparable in learning opportunity and workload? What information would they need before choosing, and who could answer their questions?
Individual preference and institutional approval address different questions. A student’s willingness does not settle how a course should handle other students’ information, while an approved account does not settle the pedagogical case for requiring its use. Clarify which decisions belong to the learner, instructor and institution. Applicable requirements depend on the actual setting; the discussion here does not determine a lawful basis or contractual entitlement.
When support becomes surveillance
Imagine a proposed dashboard that summarizes where a class is struggling. Its educational purpose might be to help an instructor revisit a confusing concept. Examine whether a small amount of task-specific information could meet that purpose before collecting complete conversations.
Now consider using the same records to infer motivation, emotional state or misconduct. That is a different decision, requiring a different evidential justification and a clear opportunity to contest an interpretation. A pattern in platform activity should not silently become a judgment about a person. Ask what is measured, what the interpretation permits and what happens when it is wrong.
Bias, representation and the comparison that matters
For an original classroom example, compare proposed feedback on explanations that express the same psychological distinction in different language styles. Does the comment attend to conceptual reasoning, or does it mistake unfamiliar phrasing for a lack of understanding? Treat this as a question to investigate with suitable expertise and varied examples, not a finding about every model.
Compare the proposed arrangement with the course’s current practice. Keeping a task manual does not establish that it is fair, and a similar average score does not settle whether different students face different burdens. Decide which differences matter for the educational purpose, who helps define them and how a student can seek human reconsideration.
Ownership, attribution and the right to use material
Separate permission to read a work, permission to send it to a service and permission to redistribute a resulting artifact. An instructor might have access to a journal article while still needing to examine the terms governing an upload or a shared course package. Record the source and intended use; check the applicable license, institutional arrangements and jurisdiction before drawing a legal conclusion.
Student work deserves its own decision. Ask whether a submitted paragraph may be reused in a later class, included in an evaluation dataset or supplied to a different tool. Attribution of AI assistance is also distinct from establishing that the output is accurate or that underlying material may be reused. A disclosure statement answers a transparency question; it does not resolve every other question.
Accountability when agents can act
Follow an invented agent task from a draft comment to a course record. Who may authorize release? Who can reconstruct the information used and correct a mistaken change? Assign responsibility for the educational decision rather than leaving it with an interface that says the task succeeded.
A review trail should support a specific purpose. Saving every conversation indefinitely may create a new burden without improving that review. Consider what a student needs to understand and challenge a decision, what an instructor needs to investigate an error, and which details neither needs to retain.
End-of-course arrangements belong in the design. Identify who can request deletion, which records that request covers and what evidence of completion is available. Do not promise that one deleted chat removes every copy, tool record or derived item in an arrangement you have not examined.
Questions worth arguing through
These are editorial discussion prompts, not reports of evaluated interventions. Examine the strongest educational purpose, the information it actually requires and the alternatives available.
- Can individual choice settle a course-wide data decision?
- How much observation does useful teaching insight require?
- When does personalization justify more data?
- What should an accountable system remember?
For any proposed use, keep the learning question and the data decision visible together. A narrower design, an optional trial or an approach without AI may be the better justified choice. The answer depends on the people, evidence and arrangement.
Sources and reading scope
Reviewed October 24, 2026. UNESCO’s official overview and NIST’s publication metadata, introduction, selected risk definitions and governance passages were inspected. The linked source records state the limits. The original course examples and data map have not been evaluated as interventions.