AI Concerns, Ethical Principles, and Responsible AI CRAGE Practice Questions (Page 7)

Questions 31–35

1. 31expert · hard

   A multinational retailer is deploying an AI system to personalize marketing offers. The system uses customer purchase history, location, and demographic data. The legal team warns that the system must comply with data protection regulations in multiple countries, including the GDPR. The marketing team wants to maximize personalization. The AI ethics board insists on fairness and transparency. Which approach best balances these competing requirements?

   • AConduct a data protection impact assessment, implement a fairness audit, and provide customers with a clear explanation of how their data is used and how to opt out.
   • BDeploy the system in all countries with the same settings to ensure consistency.
   • CUse only anonymized data and deploy the system without further oversight.
   • DLimit personalization to only customers who explicitly consent, and do not provide explanations.
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A social media platform uses an AI content moderation system. Users report that the system frequently removes posts from minority-language speakers but misses hate speech in majority languages. The platform wants to improve the system's fairness. Which approach is most appropriate?

   • AAsk human moderators to review all posts flagged by the AI, regardless of language.
   • BIncrease the sensitivity threshold for all languages to reduce false removals.
   • CSwitch to a rule-based keyword filter that uses a universal list of offensive terms.
   • DCollect more training data from minority-language communities and re-evaluate the model's error rates across languages.
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A company deploys a chatbot for customer service. The chatbot sometimes generates offensive responses. The company's responsible AI team must decide how to handle this issue. Which action is the most appropriate first step?

   • AInvestigate the root cause by reviewing the training data and model behavior, then implement filters and retrain the model.
   • BBlame the users for triggering the offensive responses.
   • CAdd a disclaimer that the chatbot may occasionally produce offensive content.
   • DImmediately take the chatbot offline and stop all AI customer service initiatives.
   Reveal Solution Discussion Select an answer first
5. 34application · medium

   A university is developing an AI system to screen student applications. The admissions office wants to ensure the system does not discriminate against applicants from underrepresented regions. Which approach is the most direct way to evaluate fairness before deployment?

   • AAsk the model developers to confirm that they did not use region as a feature.
   • BCompare the model's overall accuracy to human admissions officers' accuracy.
   • CMeasure the model's selection rates and error rates across different applicant groups and compare them for statistically significant disparities.
   • DPublish the model's code on GitHub for public review.
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A social media platform's AI recommendation system is found to amplify divisive content. The platform's leadership is concerned about societal harm but also relies on engagement for revenue. The responsible AI team must propose a solution. Which option best addresses the societal concern while minimizing revenue loss?

   • AAdjust the recommendation algorithm to reduce the spread of divisive content while monitoring engagement metrics.
   • BAdd warning labels to divisive content but keep the current algorithm.
   • CConduct a public study on the impact of divisive content before making any changes.
   • DRemove the recommendation system entirely and show content in chronological order.
   Reveal Solution Discussion Select an answer first