Q: What is Amazon Augmented AI (Amazon A2I)?
A: Amazon Augmented AI (Amazon A2I) is a service that makes it easy to build the workflows required for human review of ML predictions. Amazon A2I brings human review to all developers, removing the undifferentiated heavy lifting associated with building human review systems or managing large numbers of human reviewers.
Using Amazon A2I
Q: Why should I use Amazon A2I?
A: Many machine learning applications require humans to review low confidence predictions to ensure the results are correct. For example, extracting information from scanned mortgage application forms can require human review in some cases due to low-quality scans or poor handwriting. But building human review systems can be time consuming and expensive because it involves implementing complex processes or “workflows”, writing custom software to manage review tasks and results, and in many cases, managing large groups of reviewers.
Amazon A2I makes it easy to build and manage human reviews for machine learning applications. Amazon A2I provides built-in human review workflows for common machine learning use cases, such as content moderation and text extraction from documents, which allows predictions from Amazon Rekognition and Amazon Textract to be reviewed easily. You can also create your own workflows for ML models built on Amazon SageMaker or any other tools. Using Amazon A2I, you can allow human reviewers to step in when a model is unable to make a high confidence prediction or to audit its predictions on an ongoing basis.
Q: How do I get started with Amazon A2I?
A: Amazon A2I provides a managed experience where you can set up an entire human review workflow in a few easy steps. To get started with Amazon A2I, sign in to your AWS Console, and navigate to the Amazon SageMaker console. From there, select Human review workflows under Augmented AI. First, as a part of the human review workflow, you provide a pointer to the S3 bucket where the review results should be stored. Next, you select the appropriate task type and define conditions when a human review should be triggered. Amazon A2I provides pre-built workflows where you only need to enter a few choices and provide instructions on how your objects should be reviewed by humans. Alternatively, you can create your own custom workflow and use your own custom review templates. Once created, the workflow can be used directly in your applications using a generated unique identifier for this workflow.
Q: How can I decide what objects are sent for human review?
A: With A2I, you can define what is an acceptable prediction confidence for your business problem. You can define business rules for the machine learning predictions, based on which a human review is triggered. For Amazon Rekognition image moderation tasks, you can use the confidence score that Amazon Rekognition provides for each label it outputs to trigger human review. For Amazon Textract tasks, you can trigger a human review when specific form keys are missing or when form key detection confidence is low. You can also trigger a human review if, after evaluating all form keys in the text, confidence is lower than your required threshold for any form key. For your own custom workflow, you can write the code for business conditions in AWS Lambda or directly in your client application.
Q: How do I access a human workforce using Amazon A2I?
A: With Amazon A2I, you can choose from three workforce options: (1) Amazon Mechanical Turk; (2) Third party data labeling service providers available through the AWS Marketplace; and (3) Your own employees. See the Amazon A2I developer guide for more information.