Generative AI Leader Responsible AI and Data Governance Practice Test 8 — 15 Questions

Question 1

A healthcare company is developing a generative AI system to assist radiologists by highlighting potential abnormalities in X-ray images. They want to ensure the system's outputs are explainable and can be verified by medical professionals. Which combination of features should they use?

Accepted Answer

Grounding (citing sources) and chain-of-thought reasoning. Grounding (citing sources) and chain-of-thought reasoning provide explainability by showing the reasoning steps and source evidence. Confidence indicators help radiologists assess reliability. The other options either lack explainability or are less suitable for medical decision support.

Answer

AutoML Tables and Vertex Explainable AI

Answer

Model Cards and Datasheets for Datasets

Answer

Vertex AI Model Registry and Feature Store

Question 2

An e-commerce company uses a generative AI model to generate product descriptions. They observe that descriptions for high-end products use more sophisticated language compared to budget products, potentially reinforcing class stereotypes. What is the most likely cause, and what should they do to mitigate it?

Accepted Answer

The training data reflects real-world associations; fine-tune with a balanced dataset that includes diverse product descriptions across price ranges. The bias stems from training data correlations. Fine-tuning on balanced, diverse data can reduce stereotypical associations. The other options either do not address the root cause or are less effective.

Answer

The safety filters are too aggressive; relax them

Answer

The temperature parameter is set too low; increase it to introduce more randomness

Answer

The model architecture is biased; switch to a different base model

Question 3

A company wants to ensure that its generative AI application complies with the GDPR right to erasure (right to be forgotten) for user data used in model fine-tuning. What is the best approach?

Accepted Answer

Maintain a mapping of user identities to training data, and upon request, remove the specific data points and retrain the model. Only option B fully addresses GDPR compliance by identifying and removing specific user data from the training set, then retraining. The other options do not effectively erase the user's influence from the model.

Answer

Store data with expiration dates and automatically delete after a set period

Answer

Use a broad data deletion request on all training data

Answer

Implement differential privacy during training to prevent memorization

Question 4

A financial institution is deploying a generative AI chatbot for investment advice. According to Google's AI Principles and responsible AI practices, what is a mandatory requirement before this chatbot can be used with customers?

Accepted Answer

All responses must be reviewed by a human financial advisor before being shown to the customer. Option C is correct because Google's AI Principles emphasize that high-risk AI applications, such as financial investment advice, must include meaningful human oversight to prevent harm. In this context, a human financial advisor must review and approve each chatbot response before it reaches the customer, ensuring compliance with responsible AI practices and regulatory requirements for fiduciary duty.

Answer

The chatbot must use SynthID to watermark its outputs

Answer

The training data must be publicly available

Answer

The chatbot must be deployed on-premises to ensure data residency

Question 5

A company is building a generative AI application that will generate marketing copy. They want to ensure that the content does not contain hate speech, violence, sexual content, or dangerous instructions. Which Google Cloud feature should they enable?

Accepted Answer

Configure safety filters and custom content controls in Vertex AI. Google's safety filters are designed to block harmful content categories. Vertex AI provides custom content controls to adjust thresholds. The other options are for different purposes.

Answer

Use the People + AI Guidebook to design prompts

Answer

Enable Model Card generation in Vertex AI

Answer

Implement a human review loop for all outputs

Question 6

A machine learning engineer is evaluating a generative AI model for bias. They have a diverse test set covering gender, race, and age groups. Which metric would best indicate if the model's performance is systematically worse for certain demographic groups?

Accepted Answer

Equalized odds across demographic groups. Equalized odds measures whether a model's predictions have equal false positive/negative rates across groups. The other options either measure different aspects or are not specific to fairness.

Answer

Model perplexity on held-out data

Answer

Overall accuracy on the test set

Answer

Area under the ROC curve (AUC)

Question 7

Which Google DeepMind technology can be used to embed an invisible watermark into AI-generated images to help identify their origin?

Accepted Answer

SynthID. SynthID is Google DeepMind's tool for watermarking AI-generated content. The other options are unrelated.

Answer

TensorFlow Privacy

Answer

PAIR Explorables

Answer

What-If Tool

Question 8

A company is using a generative AI model to screen job applications. They want to ensure compliance with Google's AI Principle of avoiding unfair bias. Which practice is most effective in mitigating bias during the screening process?

Accepted Answer

Audit the training data for demographic representativeness and evaluate the model using fairness metrics. Option C is correct because auditing training data for demographic representativeness and evaluating the model using fairness metrics directly addresses the root causes of bias in generative AI systems. This practice aligns with Google's AI Principle of avoiding unfair bias by proactively identifying and mitigating imbalances in the data and measuring the model's performance across demographic groups using metrics like demographic parity or equal opportunity.

Answer

Use a pre-trained model without modification

Answer

Remove all demographic information from the resumes before processing

Answer

Only allow human reviewers to see the top 10% of candidates

Question 9

A startup is building a generative AI legal document assistant for small law firms. They want to ensure that the model's outputs are accurate and can be traced back to specific legal statutes. Which approach best supports this requirement?

Accepted Answer

Use a RAG architecture that retrieves relevant statutes and includes them as citations in the model's response. Option D is correct because Retrieval-Augmented Generation (RAG) architecture retrieves specific legal statutes from a trusted external knowledge base and includes them as citations in the model's response. This ensures both accuracy (by grounding outputs in verifiable sources) and traceability (by providing direct references to the statutes used). Fine-tuning alone cannot guarantee that the model will cite specific statutes correctly, as it may hallucinate or misremember legal references.

Answer

Fine-tune the model on a large corpus of legal documents

Answer

Apply a high temperature setting to encourage diverse outputs

Answer

Use a model larger than 70B parameters

Question 10

A company wants to use a generative AI model to create marketing content that will be used commercially. They are concerned about copyright infringement from the model's training data. Which step should they take to mitigate this risk?

Accepted Answer

Review the model's training data provenance and ensure the model is trained on properly licensed or public domain data. Reviewing the training data provenance and using models with appropriate licenses is critical. The other options either do not address the root cause or are insufficient.

Answer

Use SynthID to watermark the content

Answer

Add a disclaimer that content is AI-generated

Answer

Use only open-source models

Question 11

A company is deploying a generative AI system for medical diagnosis support. To comply with Google's AI Principles and regulatory requirements, which TWO actions are essential? (Select 2)

Accepted Answer

Implement a human-in-the-loop review for all diagnostic suggestions. High-stakes medical decisions require human oversight (Principle: be accountable to people) and data privacy (Principle: incorporate privacy design principles). The other options are beneficial but not essential for compliance.

Answer

Publish a Model Card for the model

Answer

Use SynthID to watermark all output

Answer

Use a larger model to improve accuracy

Question 12

A company is developing a generative AI application that will be used by customers in the EU. To comply with the GDPR and the upcoming EU AI Act, which THREE measures should they implement? (Select 3)

Accepted Answer

Store and process all user data within the EU or in regions with adequacy decisions. GDPR requires data residency/local processing (measures 1), explanation of AI decisions (right to explanation) (measure 3), and transparency about AI interaction (measure 5). The other options are not mandated by these regulations.

Answer

Use only open-source models to ensure transparency

Answer

Automatically delete all user logs after 30 days

Question 13

According to Google's AI Principles, which TWO of the following are core commitments? (Select 2)

Accepted Answer

Be built and tested for safety. Option D is correct because Google's AI Principles explicitly commit to building and testing AI systems for safety, including conducting rigorous safety testing and implementing appropriate safeguards to mitigate risks such as bias, misuse, and unintended harm. This commitment ensures that AI systems are designed to avoid creating or reinforcing unfair outcomes and to operate reliably under foreseeable conditions.

Answer

Maximize shareholder value

Answer

Always use the largest model available

Answer

Achieve state-of-the-art performance on all benchmarks

Question 14

A data scientist is fine-tuning a generative AI model for customer sentiment analysis. To ensure the fine-tuned model does not inadvertently memorize and reproduce personally identifiable information (PII) from the training data, which THREE practices should they follow? (Select 3)

Accepted Answer

Apply differential privacy during fine-tuning. Differential privacy limits memorization; PII removal prevents it from being present; data minimization reduces risk. Quantization and k-fold cross-validation do not directly address PII memorization.

Answer

Apply model quantization to reduce model size

Answer

Use k-fold cross-validation to evaluate the model

Question 15

A company is deploying a generative AI chatbot for customer support. They want to ensure that the chatbot does not generate harmful content and that they can customize the safety thresholds. Which TWO features in Vertex AI should they use? (Select 2)

Accepted Answer

Custom safety settings (adjustable thresholds). Safety filters block harmful categories; custom safety settings allow adjustment of thresholds. Model Cards and the People + AI Guidebook are not operational safety controls.

Answer

AutoML Tables

Answer

Model Cards

Answer

People + AI Guidebook