mediumMultiple Choice
Generative AI Leader Practice Question: A data scientist is evaluating a generative AI…
A data scientist is evaluating a generative AI model for gender bias in its text outputs. They have a test set of 1,000 gender-neutral prompts. Which approach is MOST appropriate for measuring output bias?
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Analyze the gender of characters, roles, and pronouns in the model's completions
To measure bias, the test set should include prompts that are neutral in gender but may elicit biased responses. The correct approach is to analyze the gender of pronouns, roles, or descriptors in the model's completions.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Check the gender distribution of the training data
Why it's wrong here
Training data distribution is an input property, not a measure of the model's generated outputs, so it cannot quantify bias in responses to the 1,000 prompts. It tempts because imbalanced data often causes bias, but the question asks for output measurement, which requires scoring the generated text itself.
- ✗
Use a toxicity classifier to flag any biased outputs
Why it's wrong here
A toxicity classifier detects harmful or offensive language, not gender bias, which can appear in fluent, non-toxic text such as associating neutral prompts with stereotyped roles. It tempts because automated classifiers scale across 1,000 outputs, but measuring bias requires comparing gendered term distributions or sentiment across outputs.
- ✓
Analyze the gender of characters, roles, and pronouns in the model's completions
Why this is correct
Bias is measured by auditing the model's actual completions, so coding the gender of characters, roles and pronouns across all 1,000 neutral prompts directly quantifies disparate representation. This satisfies the stem's requirement to measure output bias rather than input or training-data bias.
- ✗
Ask the model to self-report its confidence in avoiding bias
Why it's wrong here
Self-reported confidence is a subjective claim, not a measurement of the 1,000 outputs, and models cannot reliably introspect on their own bias. It tempts because it is cheap and requires no annotation, but the correct approach scores actual generated text against demographic or sentiment criteria.
Go deeper
Related to this question
About these practice questions
This Generative AI Leader question is part of Courseiva's 1,008-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This Generative AI Leader practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the Generative AI Leader exam.