NCA-GENL Question 223
Select 3A company is developing a generative AI model for summarizing medical research papers. During testing, the team identifies that the model generates biased summaries favoring specific treatments over others. Which actions should the team take to minimize bias in the AI system?
- A
Increase the training dataset size by incorporating diverse and balanced medical research papers.
- B
Fine-tune the model using only papers that align with the company's preferred treatments.
- C
Implement bias detection techniques to analyze outputs and refine the training process.
- D
Remove all mentions of treatments from the dataset to avoid introducing bias.
- E
Incorporate human-in-the-loop validation to review and adjust model outputs.
Show answer and explanation
Correct answers: A, C, E
Explanation
Minimizing bias in AI systems requires a combination of strategies, such as using diverse and balanced datasets, applying bias detection techniques, and involving humans in the review process. These approaches collectively help identify and mitigate bias while maintaining the model's utility and relevance.
- A. Correct.
Increasing the dataset size with diverse and balanced data helps to reduce bias by exposing the model to a wide range of perspectives and reducing overrepresentation of specific treatments.
- B. Incorrect.
Fine-tuning the model only on data aligning with the company's preferences would reinforce bias rather than minimize it.
- C. Correct.
Bias detection techniques allow the team to identify and quantify biases in the model's outputs, enabling targeted improvements in the training process.
- D. Incorrect.
Removing all mentions of treatments from the dataset is not a viable solution, as it will degrade the model's ability to generate meaningful and relevant summaries.
- E. Correct.
Incorporating human-in-the-loop validation ensures that model outputs are reviewed and adjusted by experts, reducing the risk of biased or incorrect summaries.