NCA-GENL Question 204
Select 3When integrating various forms of data into a generative AI Large Language Model (LLM), which practices help ensure the process is transparent, fair, and verifiable?
- A
Documenting the data sources, preprocessing steps, and any transformations applied to the data.
- B
Ensuring that only publicly available data is used, regardless of its relevance or biases.
- C
Applying bias detection techniques and correcting imbalances in the dataset before training the model.
- D
Providing clear model performance metrics specific to different demographic and data segments.
- E
Using only structured data to simplify the integration process and reduce risks of unfairness.
Show answer and explanation
Correct answers: A, C, D
Explanation
To ensure transparency, fairness, and verifiability in integrating data into generative AI LLMs, it is important to document the data preparation process, identify and address biases, and provide clear performance metrics for different use cases or populations. These practices help foster trust and accountability in AI systems, while simply limiting data types or sources may not address underlying fairness concerns.
- A. Correct.
Documenting the data sources, preprocessing steps, and transformations ensures transparency and helps verify how the data was prepared, which is critical for ensuring fairness and accountability.
- B. Incorrect.
Using only publicly available data does not inherently ensure fairness or transparency, as publicly available data can still contain biases and may not be representative of diverse populations.
- C. Correct.
Bias detection and correction are essential for ensuring fairness in the dataset, as they help address potential imbalances that could lead to discriminatory outcomes.
- D. Correct.
Providing performance metrics across different demographic and data segments is critical for verifying that the model performs fairly and does not disproportionately impact certain groups.
- E. Incorrect.
Using only structured data is not a requirement for fairness or transparency; unstructured data can also be integrated fairly and transparently if proper preprocessing and documentation steps are followed.