AI Models, Frameworks, Strategies, and Use Cases AAIR Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A healthcare startup is developing an AI system to diagnose skin conditions from photos. The training data is primarily from light-skinned patients. What is the most significant risk this model could exhibit?

   • ABias risk, leading to lower accuracy for patients with darker skin tones
   • BRobustness risk, because the model may be fooled by adversarial examples
   • CExplainability risk, because the model's decisions are hard to interpret
   • DPrivacy risk, because the photos are sensitive health data
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A research team is developing a natural language processing model for legal document review. They need to share the model with a partner organization that uses a different AI framework. What is the most efficient way to ensure interoperability?

   • AConvert the model to ONNX format for cross-framework deployment
   • BProvide the partner with the model's training data so they can rebuild it
   • CDeploy the model as a web service and have the partner access it via API
   • DShare the model's source code and require the partner to retrain it in their framework
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A bank is deploying a machine learning model to approve personal loans. During validation, they discover the model is less accurate for applicants from certain ethnic backgrounds. The bank must address this before deployment. What is the most appropriate governance action?

   • AIncrease the model's complexity to improve overall accuracy, assuming the bias will diminish
   • BRemove the sensitive attributes from the training data and deploy the model as-is
   • CDeploy the model but add a disclaimer that the model may be less accurate for some groups
   • DRetrain the model using a more diverse dataset and implement ongoing bias monitoring
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   An e-commerce company wants to automatically generate product descriptions from images. They have a large dataset of product photos and want the descriptions to be creative and varied. Which AI model category is most appropriate for this task?

   • AA generative adversarial network (GAN) to create new product images
   • BA reinforcement learning agent that learns to write descriptions through trial and error
   • CA transformer-based multimodal model that can process images and generate text
   • DA convolutional neural network (CNN) for image classification
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A healthcare provider wants to use AI to predict patient readmission risk. They have two options: a commercially available model that is a 'black box' but has strong performance, or an internally developed interpretable model with slightly lower accuracy. The provider's risk committee requires that clinicians be able to explain why a patient was flagged. Which approach best balances the competing requirements?

   • ADevelop the interpretable model internally, accepting the lower accuracy to meet the explainability requirement
   • BAdopt the commercial model and use post-hoc explainability tools like LIME to generate explanations
   • CUse the commercial model but only allow it to recommend, not decide, with clinicians making final judgments
   • DDeploy both models and use the commercial one when its confidence is high, otherwise fall back to the interpretable model
   Reveal Solution Discussion Select an answer first