Self-Assessment Questions

AAIR self-assessment questions support the content in this manual and provide an understanding of the type and structure of questions that typically appear on the exam. Often a question will require the candidate to choose the MOST likely or BEST answer among the options provided. Please note that these questions are not actual or retired exam items. Please see About This Manual for more guidance regarding practice questions.

1. An enterprise is acquiring datasets needed for a new AI model. The data engineer is currently overseeing the cleaning and preprocessing of the data to ensure it meets the model’s requirements. Which of these stages of the AI life cycle BEST categorizes the current state of this project?
   1. Design
   2. Development
   3. Evaluation
   4. Deployment
2. What is the PRIMARY data privacy risk when using AI for large-scale data analysis?
   1. Inaccurate prediction from AI models
   2. High processing power requirements
   3. Data collection beyond the necessary scope
   4. Limited integration with existing security tools
3. The MOST important reason to implement model output monitoring for AI systems is:
   1. to ensure newly implemented AI models are functioning as expected.
   2. to automatically identify and correct any errors in processing.
   3. to reduce the need for human oversight in AI processes.
   4. to identify and address potential biases that may not be flagged by standard error reports.

Chapter 2 Answer Key — Self-Assessment Questions

1. 1. The design phase typically occurs prior to data collection and preprocessing, as this stage informs the collection and preprocessing activities.

   2. Data preprocessing occurs during the development stage of an AI solution, although data acquisition can begin during the design phase and extend into development.

   3. The evaluation phase typically involves testing the model to ensure it performs correctly after data collection and processing.

   4. Deployment occurs after the model has been fully developed and tested and is ready for production use.

2. 1. Inaccurate predictions are an operational issue, not a privacy risk.

   2. High processing power is a technical challenge and is not directly related to privacy.

   3. AI systems often collect excessive data to improve accuracy, which can lead to privacy concerns, especially if a system goes beyond the necessary scope for the intended analysis.

   4. Integration with security tools is important but is not the primary privacy risk.

3. 1. Model output monitoring is essential for all AI systems, not just newly implemented ones, as biases and issues can emerge over time due to changing data or evolving system behaviors.

   2. Model output monitoring cannot automatically correct all errors and does not replace the need for periodic reviews.

   3. Model output monitoring is not intended to reduce human oversight but to complement it, ensuring that AI processes are transparent, accountable, and free from errors or biases.

   4. Model output monitoring helps identify and address potential biases that may go unnoticed in standard error reports.