Amazon Augmented AI A2I
Human review of ML results
Brings humans into the loop for checking ML predictions (human-in-the-loop): when confidence is low or by random sampling, results are routed to human reviewers (internal staff, vendors, or Mechanical Turk).
- Cost$ Low
- Difficulty●●●●●
- Costs money when idle?No
- Serverless
Think of it asLike the chief accountant signing off on invoices the machine couldn't read with confidence.
When to use it
✓ Use it when
Results from Textract, Rekognition, or your own models need human confirmation before use.
✗ Avoid it when
You need to label data to train models → SageMaker Ground Truth.
Common uses
- Re-review low-confidence invoice extractions
- Moderate sensitive content
- Random-sample checks of model quality
What does it cost?
- Pricing model: Charged by the number of objects reviewed by humans, plus fees paid to the review workforce if using Mechanical Turk or vendors.
