AI-900 Question 64
Select 3A healthcare organization is developing an AI-powered diagnostic tool for detecting skin conditions. During the development process, the team notices that their training dataset contains significantly more images of lighter skin tones compared to darker skin tones. Which steps should they take to ensure their AI solution is inclusive?
- A
Expand the training dataset to include a diverse range of skin tones.
- B
Test the AI model on a representative sample of all skin tones and evaluate its performance.
- C
Rely on the existing dataset and assume the model will generalize effectively.
- D
Engage with domain experts and affected communities to identify potential biases and address them.
- E
Focus only on the accuracy of the model, as inclusiveness does not impact AI performance.
Show answer and explanation
Correct answers: A, B, D
Explanation
To ensure inclusiveness in an AI solution, it is essential to address dataset diversity, test the model on representative samples, and engage with relevant stakeholders to identify and mitigate potential biases. Ignoring these steps can lead to biased outcomes and reduce the effectiveness and fairness of the AI solution.
- A. Correct.
Expanding the training dataset to include diverse skin tones is necessary to ensure the AI model performs well across all demographic groups.
- B. Correct.
Testing the model on a representative sample helps identify performance gaps and biases, ensuring the solution is inclusive and fair.
- C. Incorrect.
Assuming that the model will generalize effectively without addressing dataset imbalance is not a valid approach. This can lead to biased outcomes.
- D. Correct.
Engaging with domain experts and affected communities helps identify areas of bias and ensures the solution meets the needs of all users.
- E. Incorrect.
Focusing only on accuracy ignores the broader ethical implications of inclusiveness and fairness in AI solutions, which are critical considerations.