Google Professional Machine Learning Engineer Question 455
Select 4Google Cloud PlatformYou are tasked with designing an AI solution to automate loan approvals for a financial institution. During the risk assessment phase, you identify potential risks to the solution. Which of the following should you consider as part of your risk mitigation strategy?
- A
Bias in training data leading to discriminatory loan approvals
- B
Model performance degrading over time due to changes in borrower behavior
- C
High compute costs during model training and inference
- D
Data breaches exposing sensitive customer information
- E
The inability of the model to explain its predictions to regulators
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Identifying risks to AI solutions involves addressing ethical, legal, and functional challenges, especially in regulated industries like finance. Risks such as bias, concept drift, security vulnerabilities, and explainability gaps can have severe consequences if not mitigated. Operational concerns like compute costs are important but not typically categorized as risks to the AI solution's viability or compliance.
- A. Correct.
Bias in training data can lead to unfair or discriminatory decisions, which is a significant risk in AI solutions, especially in sensitive domains like finance.
- B. Correct.
Model performance degrading over time is a risk due to concept drift, where the data distribution changes over time, impacting prediction accuracy.
- C. Incorrect.
While high compute costs are a concern, they are more of an operational challenge rather than a risk specific to the AI solution's ethical, legal, or functional outcomes.
- D. Correct.
Data breaches are a critical risk in AI solutions that handle sensitive customer information, as they can lead to legal and reputational consequences.
- E. Correct.
The inability of the model to explain its predictions (lack of explainability) can result in non-compliance with regulations and loss of trust from users and regulators.