mediumMultiple Choice
Generative AI Leader Practice Question: A data scientist observes that their text…
A data scientist observes that their text generation model frequently uses gender stereotypes when generating job descriptions. Which responsible AI practice should be applied FIRST to address this issue?
⚠ Common exam trap
Google often tests the principle that evaluation must precede mitigation, tempting candidates to jump to a corrective action (like fine-tuning or filtering) without first diagnosing the problem.
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
✓
Evaluate the model's outputs for bias using a diverse test set that includes gender-neutral prompts
The first step in addressing model bias is to evaluate and measure the bias using a diverse, representative test set. This diagnostic step identifies the specific types and severity of gender stereotypes before any mitigation technique is applied, ensuring that subsequent interventions are targeted and effective. Without this evaluation, any corrective action risks being misapplied or introducing new biases.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Fine-tune the model on a dataset with more male-dominated job descriptions
Why it's wrong here
Fine-tuning on more male-dominated descriptions amplifies the very bias observed, worsening stereotype output rather than diagnosing its source. Fine-tuning is tempting because it steers model behaviour, but it is the right move for adapting style or domain vocabulary, not for first investigating and mitigating representational harm.
- ✗
Apply SynthID watermarking to the generated content
Why it's wrong here
SynthID watermarking marks AI-generated content for provenance detection; it neither measures nor removes the stereotypical associations in the model's outputs. Watermarking is tempting because it is a visible responsible-AI control, but it is the correct choice when the goal is identifying synthetic media, not auditing bias.
- ✗
Increase the number of safety filters to block all gender-related content
Why it's wrong here
Blocking all gender-related content suppresses legitimate output and leaves the underlying stereotype in the model's weights unexamined, so it fails to address the root cause. Safety filters are tempting as an immediate guardrail, but they are the right tool for blocking prohibited categories, not for diagnosing and mitigating bias.
- ✓
Evaluate the model's outputs for bias using a diverse test set that includes gender-neutral prompts
Why this is correct
Measuring bias first establishes whether and where stereotypes appear, using gender-neutral prompts across a diverse test set. This diagnostic step precedes mitigation, since debiasing, prompt engineering or fine-tuning cannot be validated without baseline evidence identifying the specific stereotyped outputs to correct.
Go deeper
Related to this question
About these practice questions
One of 1,008 original Generative AI Leader practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. 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.