Explain model behaviour, inspect group differences and make probabilities usable for decisions.
This is one focused step in the 60-lesson course. Use the retrieval check before moving on.
Lesson goal
Explain model behaviour, inspect group differences and make probabilities usable for decisions.
The core idea
Interpretability is not one thing. A global model summary, a local explanation, a fairness audit and a calibrated probability answer different questions.
Mental model
Picture it this way. Interpretability is not one thing. A global model summary, a local explanation, a fairness audit and a calibrated probability answer different questions. The important question is what assumption this picture makes, and whether that assumption fits the data.
Mathematical core
Group metrics compare error rates or calibration across defined groups. Calibration maps scores to observed frequencies. Explanations can be model-specific or model-agnostic, but all depend on the data and background distribution.
Worked example
A lending model may have good overall AUC but different false-negative rates across groups. A calibrated probability can still be unfair if the decision policy or data reflects unequal treatment.
When to use it
It earns a place when
Use explanations to debug and communicate, fairness metrics to reveal disparities, and calibration when downstream decisions use risk probabilities.
You can evaluate it against a credible baseline.
Its output fits the decision and data constraints.
Do not make it the default when
Do not reduce fairness to one metric, or treat SHAP or feature importance as causal proof. Consider the decision, affected people and recourse.
A simpler model has not been tested.
The data or target definition is still unclear.
Failure modes
Watch for this. Metrics can conflict. Proxy variables, missing groups, selective labels and feedback loops can make an audit incomplete.
When a result looks surprisingly good, inspect the split, target timing, error slices and data-generating process before celebrating.
Practice
Use this as a small experiment rather than a recipe to copy blindly. Change one thing, record the result and explain the change.
Do this. Choose one use case and write a model card: purpose, data, metrics, group slices, known limits, threshold policy and escalation path.
Retrieval check
Answer from memory first. The buttons reveal feedback, but the durable step is explaining why.
1. What does calibration ask?
2. Why can fairness metrics conflict?
3. What is a model card useful for?
Transfer prompt. Describe one real problem where this model or idea would be a sensible candidate. Name the target, the main risk and the metric you would inspect.
Primary source
NIST AI Risk Management Framework. Use the source for the deeper treatment after you can explain the lesson's core idea without looking.