As a pre-release review of a recruiting support model, a company wants to detect whether the training data or the model's predictions are biased toward specific demographic groups. A mechanism for monitoring drift after deployment is already in place separately. Which AWS service is the MOST suitable?
Select the AWS service that detects bias.
Amazon SageMaker Model Monitor
SageMaker Model Monitor is a service that monitors data/quality drift of a model after deployment. Drift refers to the phenomenon where the tendency of production input data shifts from the time of training as operations continue, gradually degrading the model's prediction accuracy.
It is not a service whose main purpose is bias detection, so it is incorrect for this question.
Amazon Augmented AI (A2I)
Amazon A2I is a mechanism for routing low-confidence predictions and the like to human review.
It is not a service that automatically detects and measures bias, so it is incorrect.
Amazon SageMaker Clarify
This is correct. SageMaker Clarify is a service that detects bias contained in training data and model predictions and also visualizes the contribution of each feature (explainability). It is used for fairness evaluation.
Amazon SageMaker JumpStart
SageMaker JumpStart is a model hub that provides pretrained models and ready-made solutions.
It is not a service that detects bias, so it is incorrect.
Remember responsible AI tools by their roles.
- SageMaker Clarify: bias detection for data/predictions + explainability (SHAP).
- Bedrock Guardrails: suppression of harmful content.
- Amazon A2I: human review.
'Bias detection' is almost always Clarify. It differs in purpose from Model Monitor (drift monitoring) and JumpStart (model hub).