Ethics & privacy · Discuss and decide

Ethics decisions for scholarship teams

Use a structured conversation before adopting an AI-assisted process. Work from a fictional case, name the evidence, and write conditions your team can actually check.

This toolkit adapts Miguel Guhlin’s ETHICAL model, ethical instruction modules, and AI Ethics Case Court materials for scholarship providers. It is a discussion aid, not a scoring instrument or official NSPA policy.

Use the seven ETHICAL steps

Engage stakeholders

Who is affected? Invite applicants, reviewers, and people who may face access barriers to identify what the proposal misses.

Think emotionally and logically

Name the concern, then separate facts, assumptions, and missing evidence. A strong reaction is a reason to investigate, not proof.

Harmonize perspectives

Identify shared goals and tensions, such as faster processing versus time for careful review. Keep meaningful disagreement visible.

Investigate options

Compare doing nothing, improving the current process, and using AI with specific limits. Ask who benefits and who could be harmed.

Collaborate on recommendations

Agree on conditions that someone can check: who reviews which output, what evidence is required, and when work pauses.

Authorize the decision

Name the person with authority to approve the process. Record the decision, its reasons, and how an applicant can question an outcome.

Learn from outcomes

Set a review date. Check errors, omissions, unequal effects, applicant feedback, and time spent verifying. Change or stop the process when evidence calls for it.

Try a 20-minute case discussion

Use groups of three to six. Assign a person to describe the potential benefit, another to question harms and assumptions, and a reviewer to record the decision. Use only the details supplied; mark unknowns.

Timing: read and choose a case for 3 minutes; compare benefits and concerns for 5; use the questions below for 7; agree on conditions and an owner for 5.

Case 1 · Faster scoring with limited review

A fictional scholarship office proposes having AI score all essays. Staff will review one in ten scores, and applicants will not be told how AI is used. The office expects faster turnaround but has not tested the scoring across languages or writing styles.

Discuss: What evidence supports the expected benefit? Who reviews the other scores? What would applicants need to know? Compare this proposal with a narrower task such as organizing cited evidence for a human reviewer.

Use the existing rubric practice to test whether the criteria and evidence are clear.

Case 2 · A recommendation letter shortcut

A fictional program asks volunteers to draft recommendation letters with a consumer chatbot. A volunteer pastes an applicant’s grades and personal circumstances into it. The program has not reviewed the tool’s data terms or explained this use to applicants.

Discuss: What information does the task actually need? Who may approve a tool and its data use? Which judgments should the letter writer make personally? Identify a way to improve the writing process without sending applicant information to an unapproved service.

Case 3 · A prediction that changes opportunity

A fictional renewal program proposes an AI “risk” label using attendance, prior grades, and postal code. The label would be visible to all reviewers and might determine who is invited to renew. The program has not established how applicants can correct errors.

Discuss: What is being predicted, and with what evidence? Could the label change reviewers’ expectations? Compare offering optional support with restricting opportunity. Name what must be investigated before anyone relies on the label.

Questions every group should ask

Benefit: What improves for applicants, and how would we know?
Privacy: What data is needed, who can see it, and when is it deleted?
Fairness and access: Who could face more errors, fewer options, or extra burdens?
Transparency: What will people be told about the process and its limits?
Accountability: Who corrects an error and considers an appeal?
Human oversight: Does the reviewer have enough time and evidence to disagree with the tool?
Voice: Have people affected by the process helped shape it?

Choose proceed, proceed only after specified conditions are met, or do not proceed. Missing evidence can be a reason to pause; a discussion does not itself authorize a real deployment.

Use prompt safeguards as a starting point

These instructions adapt the source’s ethical modules to the workshop. Instructions cannot guarantee a model’s behavior; test outputs and enforce data boundaries through your actual process. You can also use the questions without an AI tool.

You help a scholarship team discuss a fictional AI workflow. You do not approve the workflow or make applicant decisions.

Ask one ETHICAL question at a time: Engage stakeholders; Think emotionally and logically; Harmonize perspectives; Investigate options; Collaborate on recommendations; Authorize the decision; Learn from outcomes. Wait for our response before moving on.

Use only the supplied scenario. Separate facts, assumptions, and unanswered questions. Cite the specific scenario details supporting each concern. Do not invent policies, laws, applicant facts, or numerical confidence scores.

Ask about data minimization, access, retention, fairness, transparency, human review, and an accessible way to challenge errors. Do not request real applicant records or infer personal traits from writing style.

Record alternatives and disagreements. Finish with a draft decision record: proposal, evidence, unknowns, conditions, named owner, applicant explanation and review route, stop condition, and review date. Clearly mark any approval still required.

Keep a decision record

Write about the fictional case. Export a copy before leaving this page.

Proposal and people affectedFacts and evidence from the caseUnknowns and strongest opposing argumentDecision and conditions before proceedingWho does what and by whenWhat applicants are told and how they can request reviewStop condition and review date

Continue with the existing NSPA materials

Fair applicant AI-use policy lab for the detector and translation-help scenario.
2/2/2 implementation framework for planning, building, and reviewing a pilot.
PROTECT privacy review tool for examining vendor documentation. This school-oriented tool supports a first review; its scores do not establish suitability for your scholarship program.

Source materials and adaptation

Consolidated from ETHICAL model instructions, ethical AI modules, and AI Ethics Case Court. The three scholarship cases and 20-minute format are adaptations, not the original 70-minute classroom activity. Model confidence labels and claims of “immutable” prompt safeguards were not carried over.