The Model-Governance Review Board

Case study · AI Tools & Strategy · Learning Design

A model that helps some groups and harms others goes before a NIST AI Risk Management Framework (AI RMF) board. You quantify the harm, map it to the framework, choose a mitigation that names its cost, cast a vote, and defend it on camera.

Module 7 · Ethics and governance · summative simulation

A bar chart where most subgroups score high but one subgroup collapses well below the dashed aggregate line

Run the simulation

The board scenario is fixed; the answer order varies, and is seeded per learner when opened with a roster identifier. Runs in your browser and stores no personal data; the recorded rehearsal stays on your device. Start in the panel below, or open it full screen →.

What this simulation is

This is the summative simulation for the data and AI ethics and governance module. The learner brings forward the very model they audited in Module 5, whose subgroup-performance table shows real disparate impact, convenes a NIST AI Risk Management Framework (AI RMF) review, assembles a governance dossier (a Model Card and a Datasheet for the Dataset), maps the harms to the four RMF functions, and takes a position: approve, reject, or condition.

The interactive board tool walks the learner through the decision, then hands off to a recorded defense with randomized questions. The live defense before an instructor-and-peer board is still the graded event.

What it assesses

Quantifying disparate impact with subgroup metrics, ratios, and honest treatment of small subgroups.
Completing a Model Card with intended and out-of-scope use and candid limitations.
Completing a Datasheet for the Dataset covering provenance and consent.
Mapping harms onto Govern, Map, Measure, and Manage with concrete, non-generic controls.
Proposing mitigations that each name the metric or capability they degrade.
Defending an approve, reject, or condition vote under live questioning.

Why it resists AI substitution

The dossier is grounded in the student’s own audited model and its specific subgroup numbers, the mapping uses the actual RMF functions rather than generic principles, and the vote must be defended live under questions that react to the individual’s answers. A generic AI-generated ethics essay cannot supply the real subgroup table or answer the board’s follow-ups in real time.