Implement AI Solutions Responsibly AI-102 Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A large online marketplace uses Azure AI Content Safety to moderate product listings. The marketplace has a policy to block listings that contain hate symbols, but the moderation team is overwhelmed by false positives from the standard image filter. They need to reduce false positives while still catching hate symbols, including those that are subtly embedded in product images. The team also wants to prevent sellers from using text variations to bypass the filter. What should you do?

   • AUse Azure Computer Vision to detect all objects in images and manually review any image that contains a symbol.
   • BUse a custom blocklist for known hate symbols and enable variation matching for text, while keeping the standard image filter at a moderate threshold to reduce false positives.
   • CIncrease the standard image filter threshold to the highest level to block all potentially offensive images.
   • DDisable the standard image filter and rely solely on the custom blocklist for hate symbols.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A multinational corporation is rolling out an AI-powered resume screening tool. The legal team requires that the tool does not discriminate against candidates based on protected attributes, and that the company can prove compliance. What should you include in the responsible AI governance framework?

   • ARemove all demographic data from resumes before screening to ensure fairness.
   • BPublish a public statement that the tool is fair and does not discriminate.
   • CDefine policies for fairness testing, require bias mitigation during model development, and set up ongoing monitoring with responsible AI insights to detect drift.
   • DUse a third-party auditor to review the tool once before deployment.
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A social network is using Azure AI Content Safety to moderate user posts. The network has a policy to block posts that contain hate speech, but the moderation team is concerned about posts that use code words or slang that are not in the standard filter. The team also wants to prevent users from using homoglyphs to bypass the filter. What is the most effective way to handle this?

   • AUse only the standard hate filter, since it is trained to detect all forms of hate speech.
   • BCreate a custom blocklist with the code words and slang, enable variation matching to catch homoglyphs, and use the standard hate filter for general hate speech.
   • CSet the severity threshold to the highest level to block all potentially offensive posts.
   • DDisable the standard filter and rely solely on the custom blocklist.
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A social media company uses Azure AI Content Safety to moderate user-generated images and text. The moderation team wants to block a specific set of known harmful phrases that are not caught by the default severity levels. They also need to ensure that any content containing these phrases is rejected before it reaches the moderation queue. What should they configure?

   • AConfigure Azure Policy to deny access to the Content Safety endpoint for users who post those phrases.
   • BIncrease the severity threshold for all categories in the Content Safety service.
   • CUse Azure AI Language to create a custom text classification model for the phrases.
   • DAdd the phrases to a custom blocklist and set the blocklist action to 'Reject'.
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A government agency is deploying an Azure AI solution that uses Azure OpenAI to draft public responses. The agency must comply with a directive that requires all AI outputs to be reviewed by a human before release, and they must maintain an audit trail of all AI interactions. They also need to ensure that the AI does not generate content that violates specific legal disclaimers. Which approach best satisfies all requirements?

   • AConfigure Azure Policy to require a manager's sign-off for any deployment change, and enable prompt shields.
   • BImplement a human-approval workflow in the application, log all AI interactions to Azure Monitor, and add legal disclaimers to a custom blocklist.
   • CUse content filters to block all outputs that are not approved by the model itself, and store logs in Azure Blob Storage.
   • DSet up a Power Automate flow that emails every AI output to the legal team, and use Azure AI Content Safety with a high severity threshold.
   Reveal Solution Discussion Select an answer first