AI-901 ยท Principles of responsible AI
Fairness in an AI solution
Fairness as treating all people equitably: biased training data, sampling bias, disparate outcomes across groups and ways to detect and reduce bias.
Key points
- Fairness means an AI system treats all people equitably and does not disadvantage groups.
- Bias usually enters through unrepresentative training data, proxy features or skewed labels.
- Detect it by comparing results across groups, and reduce it with more representative data, feature review and ongoing monitoring.
Practice questions
Which responsible AI principle requires that an AI system treats all groups of people equitably?
- Fairness
- Reliability and safety
- Transparency
- Accountability
Show answer
Answer: A. Fairness
Fairness is the principle about equitable treatment and avoiding disadvantage to groups. Transparency is about explaining AI, reliability and safety about consistent safe operation, and accountability about human responsibility.
Which actions help detect or reduce bias in an AI solution?
- Collect training data that represents the people who will use the system
- Measure model performance separately for each demographic group
- Review features that may act as proxies for sensitive attributes
- Train only on the largest group to get cleaner patterns
- Stop monitoring once the model is deployed
Show answer
Answer: A. Collect training data that represents the people who will use the system, B. Measure model performance separately for each demographic group, C. Review features that may act as proxies for sensitive attributes
Representative data, per-group evaluation and proxy-feature review all address bias. Training on one group worsens bias, and ending monitoring lets new disparities go unnoticed.
Removing the gender column from the training data guarantees that the model cannot produce gender-biased results.
Show answer
Answer: False
False. Other features such as job title or purchase history can act as proxies for gender, so bias can persist. You still need to evaluate outcomes across groups.
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.
