A company is taking stock of services that can be used for bias detection in order to set up a responsible AI review process. Which of the following services CANNOT be used for the purpose of detecting and evaluating bias hidden in training data or model predictions?
Select the service that CANNOT be used for bias detection.
Amazon SageMaker Clarify
Clarify is a service that detects bias in data and models and also provides explainability for predictions.
Because it CAN be used for bias detection, it is incorrect for this question, which asks for the service that cannot be used.
Amazon SageMaker Data Wrangler
Data Wrangler can produce a report that inspects data for bias during preprocessing.
Because it CAN be used for bias evaluation, it is incorrect for this question, which asks for the service that cannot be used.
Amazon SageMaker Clarify bias metrics
Clarify bias metrics are a feature that evaluates imbalance between classes numerically.
Because they CAN be used for bias evaluation, it is incorrect for this question, which asks for the service that cannot be used.
Amazon SageMaker JumpStart
This is correct. JumpStart is a model hub for using pretrained models right away, and it is not a feature for the purpose of detecting and evaluating bias. For bias detection, use SageMaker Clarify.
For detecting and evaluating bias, use 'SageMaker Clarify (bias metrics and explainability)' or 'Data Wrangler's bias report'. By contrast, 'SageMaker JumpStart' is a model hub for using pretrained models right away; its purpose is obtaining and deploying models, so it cannot be used for bias detection. In a 'select the one that cannot be used' question, JumpStart is the answer.