AI-901 Responsible AI practice questions

12 sample questions with full explanations. Try each one before you open the answer. New to this area? Read the Responsible AI study guide first.

Question 1

A company trains a résumé-screening model on ten years of its own hiring decisions. Testing shows the model ranks equally qualified women lower than men. Which action best addresses the responsible AI principle at risk?
- A Encrypt the résumé data at rest and in transit, and limit who can access it
- B Check the training data for bias and compare results across groups
- C Tell every applicant that an AI system screens their résumé and explain how it works
- D Add screen-reader support to the application portal
Show answer and explanation
Answer: B

Equally qualified people getting different outcomes is a fairness problem. Past hiring decisions can carry unconscious bias into the training data, so the fix is to look for bias in that data and test whether the model treats different groups the same way.
- A: Encryption protects the data (privacy and security). It does nothing about biased rankings.
- C: Disclosing that AI is used supports transparency, but the model would still rank women lower.
- D: Screen-reader support is about inclusiveness: letting more people use the portal. It doesn't change how the model ranks applicants.
Source: Responsible AI (Introduction to AI concepts module) (Microsoft Learn)

Question 2

A healthcare provider is checking whether its appointment-priority model is fair. For each statement, choose Yes if it is true or No if it is false.

The model could produce unfair results even if nobody intended it to, because its training data might reflect unconscious bias.
Comparing how often the model is wrong for different age groups is one way to assess its fairness.
Removing patient names from the input guarantees that the model is fair.

Show answer and explanation

The model could produce unfair results even if nobody intended it to, because its training data might reflect unconscious bias.	Yes
Comparing how often the model is wrong for different age groups is one way to assess its fairness.	Yes
Removing patient names from the input guarantees that the model is fair.	No

Bias usually comes from the training data or how it was selected, not from anyone's intent, so a model can be unfair by accident. Fairness is assessed by comparing results across groups such as age or gender. Removing names doesn't guarantee anything, because other inputs such as postcode or age can still lead to unequal outcomes.

Source: Responsible AI (Introduction to AI concepts module) (Microsoft Learn)

Question 3

A hospital is building a chatbot that answers patients' questions about their medications. Which two actions most directly support the reliability and safety principle? Choose two.

Choose two.
- A Test how the chatbot responds to unexpected, misleading and adversarial questions before release
- B Translate the chatbot into the main languages spoken by patients
- C Encrypt the stored chat history
- D Make the chatbot decline to answer when it can't find the answer in approved medical content
- E Publish a page describing the data the chatbot was built on
Show answer and explanation
Answer: A and D

Reliability and safety is about the system behaving consistently and safely, including under unexpected conditions and attempts to manipulate it. Testing with unusual and adversarial input checks that behavior, and declining when there's no approved answer keeps the chatbot from giving unsafe guesses.
- B: Supporting more languages is inclusiveness. It lets more people use the chatbot but doesn't make its answers safer.
- C: Encrypting chat history protects personal data, which is privacy and security.
- E: Describing the data the system was built on helps people understand it, which is transparency.
Source: What is Responsible AI? (Microsoft Learn)
Question 4

An airport uses facial identification to let travelers through a gate. Each traveler's photo is deleted as soon as they have passed through, and only staff who need to see the photos can view them. Which responsible AI principle do these practices support?
- A Transparency
- B Privacy and security
- C Accountability
- D Fairness
Show answer and explanation
Answer: B

Deleting personal images once they're no longer needed and limiting who can see them both protect people's personal data. That is the privacy and security principle.
- A: Transparency would mean explaining to travelers how the system works. These practices protect data rather than explain anything.
- C: Accountability is about people being answerable for the system, such as an owner who approves its use.
- D: Fairness is about the system treating similar travelers the same way, for example recognizing all faces equally well.
Source: Responsible AI (Introduction to AI concepts module) (Microsoft Learn)

Question 5

An insurance company is building an AI claims assistant. For each statement, choose Yes if it is true or No if it is false.

Restricting access to the training data to specific users and groups supports the privacy and security principle.
Encrypting claims data in transit and at rest supports the privacy and security principle.
Showing customers which factors led to a claim decision is mainly a privacy and security measure.

Show answer and explanation

Restricting access to the training data to specific users and groups supports the privacy and security principle.	Yes
Encrypting claims data in transit and at rest supports the privacy and security principle.	Yes
Showing customers which factors led to a claim decision is mainly a privacy and security measure.	No

Limiting who can access data and encrypting it are standard privacy and security controls. Explaining which factors led to a decision helps people understand the system, which is transparency, not privacy and security.

Source: What is Responsible AI? (Microsoft Learn)

Question 6

A bank uses an AI system to help decide loan applications. It tells every applicant that AI is involved and describes the kinds of data the system was trained on, without revealing anything confidential. Which responsible AI principle is the bank applying?
- A Fairness
- B Accountability
- C Transparency
- D Privacy and security
Show answer and explanation
Answer: C

Transparency means making people aware of how an AI system works and what its limits are. Disclosing that AI is used and describing its training data are both transparency practices.
- A: Fairness is about applicants with similar circumstances getting the same result. Disclosure doesn't change the decisions themselves.
- B: Accountability is about people inside the bank being answerable for the system, for example through governance and human review.
- D: Privacy and security is about protecting applicants' data. Keeping confidential details back is a nice touch, but the main purpose of the disclosure is transparency.
Source: Responsible AI (Introduction to AI concepts module) (Microsoft Learn)

Question 7

A recruiting firm uses an AI model to shortlist candidates. For each statement, choose Yes if it is true or No if it is false.

Telling candidates that an AI system helps shortlist them supports the transparency principle.
Documenting the model's known limitations supports the transparency principle.
If the model is biased against a group, explaining its decisions clearly makes it fair.

Show answer and explanation

Telling candidates that an AI system helps shortlist them supports the transparency principle.	Yes
Documenting the model's known limitations supports the transparency principle.	Yes
If the model is biased against a group, explaining its decisions clearly makes it fair.	No

Making people aware that AI is involved, and what its limits are, is transparency. But explaining a biased decision doesn't remove the bias: transparency helps people spot fairness problems, and fixing them is a separate job.

Source: Responsible AI (Introduction to AI concepts module) (Microsoft Learn)

Question 8

A lender's data science team has trained a credit-limit model in Azure Machine Learning. Before release, they must check whether the model's results differ across groups such as gender, ethnicity and age. Which component of the Responsible AI dashboard is built for this?
- A Model interpretability
- B Counterfactual what-if
- C Fairness assessment
- D The Responsible AI scorecard
Show answer and explanation
Answer: C

The fairness assessment component compares how a model behaves across sensitive groups, such as gender, ethnicity and age. That shows whether similar people in different groups are treated differently, which is the core fairness question.
- A: Model interpretability explains which features drive the model's predictions. It doesn't compare outcomes between groups.
- B: Counterfactual what-if shows how one prediction changes when you change its inputs. It isn't a group-by-group fairness comparison.
- D: The scorecard is a PDF report for sharing results with stakeholders. It summarizes findings; it doesn't run the fairness analysis itself.
Source: What is Responsible AI? (Azure Machine Learning) (Microsoft Learn)
Question 9

A bank builds a voice assistant for its mobile app. Some customers have limited vision, and others prefer to type quietly in public places. Which design choice best supports the inclusiveness principle?
- A Log every conversation so reviewers can audit the assistant later
- B Publish a document that explains how the assistant was trained
- C Let customers use voice or keyboard, as they prefer
- D Encrypt every conversation before it is stored
Show answer and explanation
Answer: C

Inclusiveness means designing for people with a wide range of abilities and situations. Letting users choose voice or keyboard input lets each customer use the assistant the way that suits them.
- A: Audit logs support accountability, but they don't make the assistant easier for more people to use.
- B: Explaining how the system was trained supports transparency. It doesn't change who can use the assistant.
- D: Encryption protects customer data (privacy and security). It doesn't help people with different needs use the assistant.
Source: Accessibility and inclusion for agent design (Microsoft Learn)
Question 10

A company's help assistant suggests starter prompts full of specialized industry jargon. Customers in many countries say the suggestions are confusing. What should the team do to improve inclusiveness?
- A Add a notice that customers are talking to an AI system
- B Record each customer's prompts so errors can be traced later
- C Block the assistant from answering questions it finds sensitive
- D Rewrite the suggestions in simple, clear language
Show answer and explanation
Answer: D

Jargon excludes people who don't know it, including many people who use the assistant in a second language. Plain, clear prompt suggestions make the assistant usable by a wider, global audience.
- A: Telling people they're talking to an AI is a transparency step. The suggestions would still be confusing.
- B: Logging helps with accountability and troubleshooting. It doesn't make prompts easier to understand.
- C: Blocking sensitive answers is a safety measure. The problem here is unclear wording.
Source: Accessibility and inclusion for agent design (Microsoft Learn)
Question 11

A bank's loan-approval model turns down applicants from one postal region far more often than equally qualified applicants from other regions. Which Microsoft responsible AI principle is most directly at risk?
- A Fairness
- B Reliability and safety
- C Privacy and security
- D Transparency
Show answer and explanation
Answer: A

Fairness means an AI system should treat similar people the same way. A model that rejects equally qualified applicants because of where they live produces an unfair outcome, even if location was never meant to matter.
- B: Reliability and safety is about the system behaving consistently and safely, including in unexpected conditions.
- C: Privacy and security is about protecting personal data and the system itself.
- D: Transparency is about people understanding how the system works. It would help uncover the problem, but it isn't the principle being violated.
Source: What is Responsible AI? (Microsoft Learn)

Question 12

A hospital is reviewing an AI system that flags patient scans for urgent follow-up. For each statement, choose Yes if it is true or No if it is false.

Testing how the system behaves on blurry or unusual scans supports the reliability and safety principle.
Giving doctors a clear explanation of why a scan was flagged supports the transparency principle.
Encrypting the stored scans is the main way to meet the fairness principle.

Show answer and explanation

Testing how the system behaves on blurry or unusual scans supports the reliability and safety principle.	Yes
Giving doctors a clear explanation of why a scan was flagged supports the transparency principle.	Yes
Encrypting the stored scans is the main way to meet the fairness principle.	No

Reliability and safety means the system works consistently and safely, including in conditions it wasn't mainly designed for, so testing unusual inputs supports it. Transparency is about people understanding the system's behavior, so explaining why a scan was flagged supports it. Encryption protects data, which is privacy and security, not fairness.

Source: What is Responsible AI? (Microsoft Learn)

Want more practice on Responsible AI?

Practice the full question bank for this area, in random order, with your progress saved in your browser.

Practice Responsible AI