Lesson 11 Quiz — Responsible AI: Bias, Fairness, Guardrails & Explainability (10 easy + 10 hard)

Which AWS feature enforces denied topics, content filters, and PII redaction on Bedrock model input/output?

Which service detects potential bias in data/models and provides feature-importance explainability?

What does Amazon A2I stand for?

What does Amazon A2I primarily provide?

What is a 'black box' model?

Which document standardizes a model's intended use, training data, and evaluation results?

Evaluating model performance separately per demographic/segment rather than only in aggregate is called:

Which dataset characteristic means no single group/class disproportionately dominates the data?

Which is a genuine legal/business risk of GenAI covered in this lesson?

Which consideration in responsible model selection concerns the energy consumed to train/run a model?

A model has strong aggregate accuracy but performs notably worse for an underrepresented demographic subgroup. Which technique would have caught this?

Which of the following are core responsible-AI features named in this lesson?

A highly regulated financial use case must be auditable by regulators, even at the cost of some raw model performance. What tradeoff does this describe?

Order these from most 'black box' (least explainable) to most transparent/explainable.

Match each responsible-AI tool to its primary role.

A model was trained on data where 95% of examples come from one demographic. What is the most likely consequence for the underrepresented 5%?

A company wants a low-confidence loan-approval prediction to be reviewed by a human before being finalized. Which AWS service is purpose-built for this?

Which of these are human-centered design elements that support responsible AI, per this lesson?

A company chooses an open-source model partly for easier auditability of its architecture and training data. What should they still verify?

Which best distinguishes 'bias' from 'fairness' as used in this lesson?