Amazon Augmented AI (A2I)
A service for building human review workflows for ML predictions.
Exam Tip: A2I = human-in-the-loop for ML. If the question mentions human review of ML predictions or "human oversight" for AI decisions, the answer is A2I. It's a key service for responsible AI on the exam.
Human Review Workflows
- Create workflows where humans review ML predictions that fall below a confidence threshold
- Pre-built integration with Amazon Textract, Rekognition, and custom ML models
- Uses Amazon Mechanical Turk, private workforces, or third-party vendors for human reviewers
- Configurable review conditions (e.g., "review when confidence < 95%")
Integration with ML Services
- Textract: Human review of document extraction results
- Rekognition: Human review of content moderation decisions
- Custom Models: Human review of any ML model's predictions via custom task templates
- SageMaker: Integrate A2I with SageMaker endpoints
Common Use Cases
- Content moderation with human oversight for edge cases
- Medical image review where ML confidence is low
- Financial document processing requiring human verification
- Legal document review with AI-assisted pre-screening
- Any high-stakes AI decision requiring human approval