NCA-GENL Question 203
Select 3A team is developing a generative AI model that integrates text, image, and audio data. To ensure the system is designed and applied in a transparent, fair, and verifiable manner, what practices should the team prioritize?
- A
Document the data sources and preprocessing steps for each data type.
- B
Ensure the model’s predictions are only tested on text data to simplify evaluation.
- C
Perform regular audits to check for bias in the integrated dataset.
- D
Use explainable AI (XAI) techniques to provide insights into the model’s decision-making process.
- E
Avoid using synthetic datasets to prevent introducing potential inaccuracies.
Show answer and explanation
Correct answers: A, C, D
Explanation
Building a generative AI system that integrates multiple data types requires prioritizing transparency, fairness, and verifiability. This involves documenting data handling processes, auditing for bias, and employing techniques like explainable AI to provide insights into the model's behavior. These practices ensure that the model's design and application align with ethical and technical standards, fostering trust and accountability.
- A. Correct.
Documenting data sources and preprocessing steps ensures transparency, which is a key aspect of fair and verifiable AI system design.
- B. Incorrect.
Testing the model only on text data limits the evaluation and does not account for the integration of other data types like images and audio, which is essential in this scenario.
- C. Correct.
Regular audits help identify and mitigate biases in the integrated dataset, promoting fairness and reliability.
- D. Correct.
Explainable AI techniques allow stakeholders to understand how the model makes decisions, ensuring transparency and trustworthiness.
- E. Incorrect.
Synthetic datasets can be valuable for training or augmenting models if properly validated, and outright avoiding them may limit the model's capabilities or robustness.