Practice AI transparency vocabulary: explainability reports, black box models, model documentation, flagging decisions for human review, and explaining model decisions.
0 / 5 completed
1 / 5
The compliance team requests an ___ report for the loan decision model. What does this document contain?
An explainability report describes how a model makes decisions — which input features matter most, how they influence outputs, and how users or regulators can understand individual predictions. Required for high-risk AI systems.
2 / 5
A regulator calls your model a '___ box model.' What are they criticising?
A black box model is one whose internal workings are not interpretable — you can see the inputs and outputs but not the reasoning. Deep neural networks are often described this way, raising concerns for regulated domains.
3 / 5
The team publishes ___ documentation covering the model's training data, intended use, and limitations.
Model documentation (often a model card) is a standardised document that describes what a model does, what data it was trained on, its performance across subgroups, known limitations, and intended and prohibited uses.
4 / 5
The system logs show: 'The decision was ___ for human review.' What happened?
When a decision is flagged for human review, the AI system has identified a case that exceeds a confidence threshold or matches a sensitivity rule, routing it to a human reviewer instead of acting automatically.
5 / 5
In a stakeholder meeting you say: 'We can ___ 85% of decisions.' What are you communicating about the model?
'We can explain 85% of decisions' means that for 85% of model outputs, the system can provide an interpretable reason — such as the top contributing features — while the remaining 15% are still effectively a black box.