Google Professional Machine Learning Engineer Question 456
Select 4Google Cloud PlatformYou are designing an AI solution for a healthcare provider to predict patient readmissions within 30 days of discharge. During the design phase, you identify potential risks to the AI solution. Which risks should you prioritize to ensure the solution is safe, fair, and effective?
- A
Bias in training data leading to unfair predictions for certain demographics
- B
Insufficient model explainability making decisions difficult to justify to healthcare professionals
- C
Overfitting on the training data resulting in poor generalization to new patient data
- D
High compute costs during model inference affecting operational scalability
- E
Changes in healthcare regulations that could make the model non-compliant in the future
Show answer and explanation
Correct answers: A, B, C, E
Explanation
When identifying risks to an AI solution, especially in sensitive domains like healthcare, it is crucial to prioritize risks that impact safety, fairness, and effectiveness. Bias in training data, model explainability, overfitting, and regulatory compliance are all critical factors that could undermine the success of the solution. While compute costs are operationally important, they do not directly affect the ethical or functional aspects of the AI model.
- A. Correct.
Bias in training data is a critical risk, especially in sensitive domains like healthcare, as it can lead to unfair or unethical outcomes.
- B. Correct.
Lack of model explainability is a key risk because healthcare professionals need to understand and trust the AI's decisions to use it effectively.
- C. Correct.
Overfitting is a significant risk as it affects the model's ability to generalize, which is crucial for predicting outcomes on new patient data.
- D. Incorrect.
While high compute costs are a concern for operational efficiency, they do not directly impact the safety, fairness, or effectiveness of the AI solution.
- E. Correct.
Changes in healthcare regulations are a valid risk as non-compliance can result in legal and ethical issues, making it critical to consider during the design phase.