AI-901 ยท Principles of responsible AI
Reliability and safety
Building AI that performs consistently and safely: testing, edge cases, failure handling, human oversight, harmful content, content filters and groundedness.
Key points
- Reliable and safe AI performs consistently, handles unexpected inputs and does not cause harm.
- Test with edge cases and adversarial prompts, apply content filters, keep generated answers grounded in trusted data, plan failure handling and keep humans able to intervene.
Source: https://learn.microsoft.com/en-us/azure/foundry/responsible-use-of-ai-overview
Practice questions
Which responsible AI principle covers consistent performance, handling of unexpected input and avoiding harm?
- Reliability and safety
- Inclusiveness
- Fairness
- Transparency
Show answer
Answer: A. Reliability and safety
Reliability and safety is about dependable, harm-avoiding behaviour. Inclusiveness is about empowering everyone, fairness about equitable treatment, transparency about explainability.
Which practices improve the reliability and safety of a generative AI application?
- Test with adversarial prompts and unusual edge cases before release
- Apply content filters to model outputs
- Provide a path to a human when the system cannot answer safely
- Turn off logging to reduce cost
- Test only with typical, well-formed inputs
Show answer
Answer: A. Test with adversarial prompts and unusual edge cases before release, B. Apply content filters to model outputs, C. Provide a path to a human when the system cannot answer safely
Adversarial testing, output filters and human escalation reduce risk. Disabling logging removes visibility and testing only happy paths misses failures.
Once an AI model has passed its pre-release tests, ongoing monitoring is unnecessary.
Show answer
Answer: False
False. Real-world inputs and data drift can expose problems that tests did not, so tracing and monitoring should continue in production.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
