Module 1 · AI Foundations
AI Ethics, Bias, and Explainability
Explore common sources of AI bias, ethical risks, and methods for explaining model predictions to technical and business audiences.
Core idea
Bias can enter at every system boundary
Bias is not only a property of a model. It can originate in how a problem is framed, who is represented in data, how labels are assigned, which metric is optimized, how thresholds are chosen, and how people use the output. Ethics asks whether the system should be built and under what conditions; fairness and explainability help teams examine its behavior.
Bias pathway
From social process to real-world outcome
A technically accurate model can still be harmful if the target, data, threshold, or operational use is inappropriate.
Fairness investigation
| Question | Evidence to inspect | Possible response |
|---|---|---|
| Who is represented? | Coverage, missingness, sample size, and historical exclusions. | Improve collection, scope the system, or avoid unsupported groups. |
| Who experiences errors? | False-positive and false-negative rates across relevant groups. | Revisit features, labels, thresholds, objectives, or workflow. |
| Can people understand and contest? | Explanation quality, notice, appeal, and human-review records. | Add explanations, escalation, documentation, and recourse. |
| Does deployment change behavior? | User reliance, automation bias, overrides, complaints, and feedback loops. | Train users, limit automation, monitor, and redesign controls. |
Explainability at three levels
Global
How does the model generally behave, which inputs matter, and where are its limits?
Local
Which factors most influenced this individual prediction or recommendation?
System
How do data, model, thresholds, rules, humans, and interfaces combine into the final outcome?
User-centered
What explanation does this audience need to understand, act, question, or appeal?
Exam lens
Scenario signals
- Historical data can reproduce historical inequity.
- Protected attributes are not the only risk; other features can act as proxies.
- Explainability methods support understanding but do not prove fairness or causality.
- Use subgroup metrics, documentation, monitoring, and human accountability together.
Key takeaways
- Bias can enter through data and design
- Fairness must be measured, not assumed
- Explainability clarifies prediction drivers
- Ethical review continues after deployment
Check your understanding
- Can you explain this concept in two sentences without using jargon?
- Can you identify the input, process, output, and validation step in the example?
- Can you name one suitable use case and one case where another approach is better?
- Which risk or limitation should a responsible implementation address?
Research references
Public sources and further reading
This lesson is original educational writing informed by the public references below. Use the sources to explore definitions, technical details, and current AWS exam objectives.