AI service
Amazon Augmented AI (A2I)
Adds human review workflows for ML predictions.
Key points
- Adds human review steps to machine learning prediction workflows.
- Can route low-confidence or sensitive predictions to human reviewers.
- Integrates with services such as Textract and Rekognition and can support custom ML workflows.
- Helps balance automation with human oversight for accuracy, compliance, or business policy.
- Uses human review workflows rather than training a model itself.
When to use it
- Choose A2I when ML outputs need human review before final decisions.
- Use it for document extraction review, content moderation review, or sensitive approval workflows.
- Use it when the scenario mentions low confidence predictions or human-in-the-loop review.
Exam tips
- A2I reviews predictions; Ground Truth labels training data.
- A2I does not replace model monitoring, bias detection, or access controls.
- Human review can improve trust but increases latency and operational cost.
- Use A2I with Textract when extracted document fields need human verification.