The AI Deployment Review Board

 

Copy Text

The AI Deployment Review Board

Primary Tool: Google Gemini, Gems (one custom board chatbot)

Focus: AI Governance and Ethical Decision-Making in Healthcare

Disciplines: Healthcare and Nursing, Health Policy, Health Informatics, and any field with professional governance or ethics requirements

 

ACTIVITY OVERVIEW

 

In real health systems, no clinical AI tool reaches the bedside and no AI-related policy advances without first passing through a governance body that weighs competing interests and issues a decision with conditions attached. In this activity, students serve as a member of a Deployment Review Board. They are not asked whether artificial intelligence is good or bad in the abstract. They are asked to talk through a specific case with the panel, weigh the competing concerns, and reach a defensible position on what should happen and where human judgment must remain in control.

 

Students do this in conversation with a custom Gemini Gem, a single board chatbot that role-plays every stakeholder on the panel. Stakeholders surface on their own as the discussion reaches their concerns, the board never decides for the student, and the session closes naturally once the student has explored the perspectives and explained a defensible position. The board then offers a short summary of the session and a brief, formative assessment.

 

Two pathways, one chatbot. The board supports a clinical-tool review and a policy review. Both use the same chatbot and produce the same deliverables; only the case and the active stakeholders differ, so students can choose the path that fits their interests.

 

OBJECTIVE & OUTCOME

 

Goal: Move students from describing AI ethics in the abstract to exercising governance judgment, by reaching a defensible, conditional position on a healthcare AI case that weighs competing interests and specifies where human judgment must remain non-delegable.

 

Outcome: Students explore a healthcare AI decision from multiple stakeholder perspectives, reason from recognizable ethical tenets toward a defensible position, articulate the conditions protecting human oversight, relational care, and patient autonomy, weigh the equity and structural implications, and critically assess where the chatbot's simulation was useful and where it overreached.

 

INTRODUCTION FOR STUDENTS

 

You sit on a Deployment Review Board. A case has come before you, and the board must reach a position with conditions, not a simple yes or no. You will talk it through with a panel of stakeholders, played by a single board chatbot built for this course. You do not have to call on anyone; as the conversation reaches a given stakeholder's concern, that person will step forward to speak.

 

Notice one thing early: the board will not tell you what to decide. If you ask whether to approve the tool or advance the bill, the chair will hand the question back to you. The decision is yours. Treat the stakeholders as informed but fallible, and watch for moments when the simulation says something that does not hold up. When you have explored the perspectives and can explain a defensible position, the chair will close the session and give you a short summary and a brief assessment to include with your work.

 

STUDENT INSTRUCTIONS

 

1. Choose your scenario and pathway. Select one Spark Scenario below. Scenarios 1 and 2 are clinical-tool reviews; Scenario 3 is a policy review.

2. Open the board chatbot. Sign in at gemini.google.com with your FSU email and open the Deployment Review Board Gem using the link your instructor posts, or build your own copy from the system prompt your instructor provides.

3. Talk through the case. Describe the case and share your initial thinking. Respond to the stakeholders as they step forward, follow the tensions that emerge, and explain the reasons behind the conditions or changes you would require. You do not need to summon anyone or use any special commands; just discuss the case as a member of the board.

4. Let the session close. Once you have explored the perspectives and explained a defensible position, the board chair will bring the session to a close and provide a short interaction summary and a brief formative assessment. Keep the chat open so that you can export the required information. 

5. Write your Decision Memo. Use the Decision Memo Template below.

6. Write your AI Reflection. In 150 to 200 words, name one moment the chatbot sharpened your thinking and one moment it overreached and had to be corrected.

7. Export and submit. Export the chat summary and rubric result (Share, then Export to Docs, or copy it into a Google Doc or save as a PDF) and submit it together with your memo and reflection in Canvas. Do not submit the entire chat. 

 

 

THE BOARD CHATBOT (GEM SYSTEM PROMPT)

This is the system prompt that defines the board Gem. The instructor builds the Gem once and posts the link, or posts this prompt for students to build their own copy. All eight personas live inside this one chatbot; it uses the four that match the student's pathway, and a Board Chair voice facilitates and closes the session.

 

You are THE DEPLOYMENT REVIEW BOARD, a stakeholder-panel simulation for a graduate healthcare AI governance course. A student serves as a member of the board and talks a case through with the panel before reaching a position. You role-play the stakeholders and facilitate as the Board Chair. You never make the decision for the student.

 

AT THE START: Speaking briefly as the Board Chair, welcome the student, confirm which case they are reviewing and whether it is a CLINICAL-TOOL review or a POLICY review, and invite their first thoughts. Keep this short.

 

STAKEHOLDERS YOU CONTAIN: Clinical-tool pathway: Frontline Nurse / Clinician; Data Scientist / Vendor Representative; Patient and Community Advocate; Compliance, Risk, and Ethics Officer. Policy pathway: Bill's Sponsor / Legislative Aide; Health System Operations Leader; Patient Rights and Community Advocate; Clinical Ethicist / Informaticist. Use only the four that match the case's pathway.

 

HOW THE PANEL SPEAKS:

1. Speak as only ONE stakeholder at a time. Begin each turn by having that stakeholder emerge naturally, naming the role, for example, The Frontline Nurse steps forward or  The Compliance Officer raises a hand.  Stay fully in that role until the discussion moves toward a topic of concern for another stakeholder, at which point a different stakeholder takes the floor.

2. Let stakeholders surface on their own when the conversation reaches their concern. The student does not have to call on anyone by name.

3. Each stakeholder is informed but fallible and holds a definite stance (below). Argue your view; give your strongest objection and your bottom line when relevant; do not be agreeable by default and do not converge prematurely.

4. Keep each turn focused and fairly short so the student does the analytic work, and end with a question or challenge that moves the discussion forward.

 

ABSOLUTE RULE: Never make, recommend, or hint at the final decision. If the student asks what they should do, or whether to approve or advance, the Board Chair reminds them the decision is theirs and asks one question that advances their reasoning.

 

THE BOARD CHAIR may step in at any time to keep the discussion productive, to invite a perspective the student has not yet considered, or to answer a plain process question. No command from the student is needed.

 

CONCLUDING NATURALLY: There is no command to end. Watch for the point at which the student has (a) engaged the perspectives that matter for this case and (b) explained their own position with defensible reasons for the conditions or changes they would require. At that point the Board Chair brings the session to a close: acknowledge that the student has explored the key tensions and articulated a defensible position, then deliver the two closing items below. If the student seems to wrap up before exploring an important perspective or tension, the Chair gently names what is still unexplored and invites them to continue before closing.

 

AT THE CLOSE, the Board Chair delivers:

 

(1) INTERACTION SUMMARY: one paragraph recapping the perspectives explored, the key tensions raised, the ethical reasoning the student offered, and the direction of their position. Do not state the board's decision; that belongs in the student's memo.

(2) FORMATIVE ASSESSMENT: begin with the sentence  This is a formative assessment for your instructor's review, not a final grade.  Then rate the session on three dimensions using Emerging, Developing, Proficient, or Exemplary, with one sentence of evidence each. ENGAGEMENT: how fully and persistently the student explored the relevant perspectives. REASONING: whether the student's arguments are clear, defensible, and responsive to the tensions raised. ETHICAL GROUNDING: whether the student raises tenets consistent with an ethics theory or framework; they need not name the theory, but their arguments should reflect recognizable ethical principles such as fairness, autonomy, accountability, transparency, or the protection of human judgment. Then remind the student to copy the summary and assessment into their Canvas submission.

 

STANCES:

 

[Frontline Nurse / Clinician] Protective of patient safety, professional judgment, and the clinician-patient relationship; wary of alarm burden and of attention diverted from the patient; will name the evidence and conditions you require.

[Data Scientist / Vendor Representative] Believe in the tool and want it adopted, but are professionally honest about training data, validation, weakest performance, and limits; never overstate.

[Patient and Community Advocate] Represent patients, including communities historically underserved or misrepresented in health data; care about transparency, consent, fairness across groups, and patients' ability to understand and contest decisions.

[Compliance, Risk, and Ethics Officer] Think in accountability, documentation, liability, and regulatory exposure; insist on clear answers about who is responsible for an error and how human oversight is documented.

[Bill's Sponsor / Legislative Aide] Believe the bill solves a real problem and want it to advance; explain its intent and built-in compromises; concede a weakness only to a specific argument.

[Health System Operations Leader] Focused on cost, workflow, staffing, and unintended consequences of compliance; not anti-oversight, but wary of expensive or impractical mandates.

[Patient Rights and Community Advocate] Represent vulnerable patients and communities, including Indigenous communities concerned with data sovereignty and collective benefit; care about transparency, consent, the right to a human decision-maker, and who governs data about a community.

[Clinical Ethicist / Informaticist] Translate between technical reality and ethical principle; care whether the policy preserves human judgment in clinical decisions, whether its definitions are coherent, and whether it is enforceable.

 

BEGIN NOW as the Board Chair with a brief welcome and confirmation of the case and pathway.

 

SESSION CLOSE AND INTERACTION RUBRIC

 

At the close, the board chatbot provides a short summary of the session and a brief formative assessment on the three dimensions below. The assessment measures the student's ability to engage, reason, and ground their position in defensible ethical tenets; it does not measure debating skill, and students need not name a specific theory. The chatbot's assessment is advisory and is reviewed by the instructor (see Faculty Adaptation Notes).

 

Dimension

Emerging

Developing

Proficient

Exemplary

Engagement

Explored only one or two perspectives; minimal discussion.

Explored several perspectives but without depth.

Explored the perspectives relevant to the case with sustained discussion.

Explored every relevant perspective thoroughly and revisited them as tensions emerged.

Reasoning

Few reasons; claims asserted rather than supported.

Some reasons, but uneven or loosely tied to the tensions.

Clear, defensible reasons that respond to the main tensions.

Well-supported reasoning that weighs competing tensions and anticipates objections.

Ethical grounding

Little ethical content; mainly practical or technical points.

Gestures at ethical concerns without consistent grounding.

Raises tenets consistent with an ethics framework and applies them to the case.

Grounds the position in coherent ethical principles applied consistently throughout.

 

DECISION MEMO TEMPLATE

 

Write your memo to the chair of the Deployment Review Board (600 to 800 words), using these headings:

- Case and pathway. One or two sentences naming the tool or bill and the pathway you chose.

- Decision. Approve, Approve with Conditions, or Do Not Approve. For the policy pathway: Advance, Amend, or Reject.

- Rationale. How you weighed the competing stakeholder interests; name the specific tensions you had to resolve.

- Human-oversight conditions (required). Where human judgment must remain non-delegable, and how relational care and patient autonomy are protected.

- Equity and structural safeguards. Who could be harmed or excluded, and what monitoring or safeguard addresses it.

- Monitoring and re-review. What will be tracked after the decision, and what would trigger reversal or amendment.

 

 

WHAT TO SUBMIT

 

Submission Item

Requirement

Deployment Decision Memo

600 to 800 words, in the required structure: case and pathway, decision, rationale, human-oversight conditions, equity and structural safeguards, and monitoring and re-review.

 

Interaction Summary and Assessment

The summary and formative assessment the board provides at the close of the session; paste both into your submission.

 

AI Reflection

150 to 200 words naming one moment the chatbot sharpened your thinking and one moment it overreached and had to be corrected.

 

SPARK SCENARIOS

 

Choose the scenario that best fits your course and your students' interests. The same board chatbot supports all three.

 

Scenario 1: The Sepsis Early-Warning Algorithm

Pathway: Clinical-tool review

Best fit: Healthcare, Nursing, Health Informatics, Quality and Safety

What the instructor does: Describe a vendor's intensive-care model that flags patients at elevated risk of sepsis hours before clinical signs are obvious. Provide that internal validation shows strong overall accuracy but noticeably weaker performance for one patient subgroup, and that clinicians report the alerts add to an already heavy alarm burden.

What students do: Talk through the case with the panel, following the tensions among validation, subgroup performance, alarm burden, and accountability, then reach a defensible position on whether to deploy and on what conditions, specifying where human judgment must remain in control.

Why it works: The case forces students to weigh real diagnostic benefit against equity and workflow harm, which is the central tension in clinical AI governance, and it makes the non-delegable role of clinician judgment concrete.

 

Scenario 2: The Ambient AI Clinical Documentation Tool

Pathway: Clinical-tool review

Best fit: Nursing, Primary Care, Behavioral Health, Health Informatics

What the instructor does: Describe an AI scribe that listens to the clinical encounter and drafts the note automatically, promising to reduce documentation burden while raising questions about patient consent, transcription accuracy, and whether clinician attention shifts from the patient to the device.

What students do: Talk through consent, accuracy, relational care, and documentation accountability with the panel, then reach a defensible position on whether to deploy and under what safeguards.

Why it works: The case centers relational care and patient autonomy, surfacing the risk that an efficiency tool quietly changes the nature of the clinical encounter, a harm that does not appear in accuracy metrics.

 

Scenario 3: The Algorithmic Transparency in Patient Care Act

Pathway: Policy review

Best fit: Health Policy, Nursing, Public Health, Law and Ethics

What the instructor does: Post the fictional bill text provided in the Appendix, or, with your approval, substitute a current real bill or regulation. Ask students to reach a position on whether the bill should advance, be amended, or be rejected.

What students do: Talk through disclosure, human-review requirements, equity monitoring, and feasibility with the panel, then reach a recommendation with the amendments or conditions they would require.

Why it works: The case moves governance from the bedside to the policy level, showing students how transparency and human-oversight obligations are written, contested, and made enforceable, and where good intentions meet operational reality.

 

 

FACULTY ADAPTATION NOTES

Placement and time: Plan for roughly two to three hours of student work outside class, and place the activity mid-to-late semester, after students have had at least one lower-stakes experience with a generative AI tool. It pairs naturally with a unit on AI governance, ethics, or health policy. No shared accounts or folders are required; students work in their own FSU Google identities.

 

Building the board Gem

At gemini.google.com, sign in with your FSU account, open Explore Gems, and choose New Gem. Paste the system prompt above into the instructions field, name the Gem (for example, Deployment Review Board), and save. Test it once by playing a student: confirm that it asks for your pathway, lets the correct stakeholders emerge naturally rather than waiting to be called, stays in one role per turn, refuses to make the decision, and closes the session on its own with a summary and a brief assessment once you have argued a defensible position.

 

If link-sharing does not work: Gem link-sharing across accounts can be unreliable, which the pilot itself notes. The reliable fallback is to post the full system prompt in Canvas and have students create their own copy of the Gem; the experience is identical. Provide both the link and the prompt so students are never blocked.

Why the chatbot refuses to decide: The rule against making the decision is the pedagogical core. It places the preservation of human judgment in the tool itself, so the student cannot offload the governance decision to the AI. Keep this rule if you adapt the prompt.

 

Customizing the cases: The scenarios and the fictional bill are starting points. Substitute a tool or policy specific to your specialty, narrow the subgroup named in the sepsis scenario to one relevant to your population, or replace the fictional bill with current legislation. You can also edit a stakeholder's stance in the prompt to match your context.

 

Grading guidance

The board chatbot produces two grading inputs at the close of each session: a short interaction summary and a formative assessment of the student's engagement, reasoning, and ethical grounding. Treat the chatbot's assessment as advisory: review it against the transcript and adjust as needed, since delegating the grade entirely to the AI would contradict the human-oversight principle the activity teaches. A workable approach is to use the chatbot's assessment, after your review, for the live-session portion of the grade, and to grade the written deliverables separately with the rubric below, so the two instruments cover different artifacts rather than double-counting the conversation.

This deliverable rubric (100 points; adjust to your scale) grades the written products.

 

Criterion (points)

What strong work shows

Decision rationale (30)

Names the real tensions among stakeholders and resolves them coherently; the decision follows from the analysis.

Human-oversight conditions (30)

Specifies where human judgment is non-delegable and how relational care and patient autonomy are protected.

Equity and structural analysis (20)

Identifies who could be harmed or excluded and proposes a concrete safeguard or monitoring step.

AI reflection (20)

Names a specific moment the chatbot helped and one where it overreached, with precision.

 

Alignment to standards

The activity maps to the National Academy of Medicine AI Code of Conduct, whose commitments and principles, including human-centered design, equity, accountability, and ongoing monitoring, are what the decision memo asks students to operationalize. It supports the AACN Essentials, particularly the domain addressing informatics and healthcare technologies, and the broader expectation that graduates can participate in the governance of health technologies. The World Health Organization guidance on the ethics and governance of artificial intelligence for health offers a complementary global frame, and the CARE Principles for Indigenous Data Governance support the community-advocate persona where data sovereignty is in view. Also the Competencies for the Use of Artificial Intelligence-Based Tools by Health Care Professionals (Russell et. All , 2023).

This activity makes AI use the object of study rather than a hidden shortcut. Because students submit the full transcript alongside the memo, their reasoning is visible and verifiable, and the chatbot is built so it cannot supply the decision. State in your syllabus that using the board Gem is required here and that the graded outcome is the quality of governance judgment, not the fluency of the AI's text.

 

Adoption and power dynamics to anticipate

A polished memo can mask shallow reasoning, which is why the transcript and the formative assessment are part of the design. When students analyze a real vendor's tool or a sitting legislator's bill, the exercise can become politically sensitive; the fictional scenarios let you teach the method without that exposure. Access to and comfort with AI tools vary across a cohort, so in-class modeling helps level the starting line. Be candid that governance decisions in practice are shaped by cost, liability, and power, not only by ethics, which is why the operations and compliance stakeholders are part of the panel.

 

Privacy and data handling

Instruct students not to paste real patient data, identifiable information, or confidential proprietary vendor documentation into Gemini. The scenarios and bills provided are synthetic for this reason; a student choosing the real-bill option should work from the public text only.

 

APPENDIX: THE ALGORITHMIC TRANSPARENCY IN PATIENT CARE ACT

This is a fictional bill drafted for instructional use. It is not real legislation and should not be cited as such.

 

A BILL to be entitled the Algorithmic Transparency in Patient Care Act.

 

Section 1. Definitions.  Clinical AI system  means any software using machine learning or comparable automated methods to generate, recommend, or influence a diagnostic, triage, treatment, or care-management decision for an identifiable patient.

Section 2. Patient disclosure. A health system that uses a clinical AI system in a patient's care shall disclose that use to the patient in plain language, including the system's general purpose and known limitations, at or before the time the decision is communicated.

Section 3. Human review. A clinical AI system shall not be the sole basis for a diagnostic or treatment decision. A qualified licensed professional shall review and may override any AI-generated recommendation, and the review shall be documented in the medical record.

Section 4. Equity monitoring. A health system shall monitor the performance of each clinical AI system across relevant patient subgroups at least annually and shall report material disparities to the state health authority.

Section 5. Effective date. This act takes effect twelve months after passage.

 

FRAMEWORKS REFERENCED

 

Cite in APA 7 if you draw on these.

- American Association of Colleges of Nursing. (2021). The Essentials: Core competencies for professional nursing education.

- Global Indigenous Data Alliance. (2019). CARE principles for Indigenous data governance.

- National Academy of Medicine. (2025). An artificial intelligence code of conduct for health and medicine: Essential guidance for aligned action. National Academies Press.

- Russell RG, Lovett Novak L, Patel M, Garvey KV, Craig KJT, Jackson GP, Moore D, Miller BM. Competencies for the Use of Artificial Intelligence-Based Tools by Health Care Professionals. Acad Med. 2023 Mar 1;98(3):348-356. doi: 10.1097/ACM.0000000000004963. Epub 2022 Sep 6. PMID: 36731054.

- World Health Organization. (2021). Ethics and governance of artificial intelligence for health: WHO guidance.