Posted On: Apr 24, 2020
Today we are excited to announce the general availability of Amazon Augmented AI (Amazon A2I), a new service that makes it easy to implement human reviews of machine learning (ML) predictions. Amazon A2I brings human review capabilities to all developers by removing the undifferentiated heavy lifting associated with building and managing human review systems.
Many machine learning use cases require human judgement to get high confidence predictions, or to audit predictions on an on-going basis. For example, extracting information from scanned mortgage application forms might require human review due to low-quality scans or poor handwriting. Building human review systems to do this work 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 a large workforce of reviewers.
Amazon A2I provides built-in human review workflows for common ML use cases like content moderation (with Amazon Rekognition) and text extraction (with Amazon Textract). There is also an option to create human review workflows for custom ML models including those built on Amazon SageMaker. You can use a pool of reviewers from within your own organization, or you can access the workforce of over 500,000 independent contractors who are already performing machine learning tasks through Amazon Mechanical Turk. Alternatively, if your data requires confidentiality or special skills, there is a workforce of vendors who are experienced with review projects and pre-screened by AWS for quality and security procedures, including iVision, CapeStart Inc., Cogito, and iMerit
Below are a few examples of how Amazon A2I works with Amazon Textract, Amazon Rekognition and custom models in SageMaker:
Amazon A2I with Amazon Textract: Amazon Textract is a service that automatically extracts text and data from scanned documents. Amazon Textract goes beyond simple optical character recognition (OCR) to also identify the contents of fields in forms and information stored in tables. This service allows you to define the conditions in which a human reviewer is needed by using the Amazon Textract form data extraction API and Amazon A2I. The conditions can be adjusted at any time to achieve the right balance between accuracy and cost-effectiveness. With this you can specify which form fields are important in the documents and send those to human review. You can also choose to send a random sample of Amazon Textract predictions to human reviewers, using those results to learn more about the model’s performance and to audit model inferences. Learn more about Amazon Textract and A2I.
Amazon A2I with Amazon Rekognition: Amazon Rekognition helps you identify potentially unsafe or inappropriate content across both image and video assets and provides you with detailed labels that allow you to accurately control what you want to allow based on your needs. With Amazon Rekognition, you can automate content moderation workflows by setting business rules for the prediction confidence. For sensitive or nuanced content, customers often need trained human experts to verify machine predictions that are lower confidence. With Amazon A2I, you can easily add human reviews to moderation workflows, for example, flagging all images that return a prediction confidence for explicit nudity between 50% to 80% for additional human review. Learn more about how to set up Amazon A2I with Amazon Rekognition moderation.
Amazon A2I with Custom ML models: Amazon A2I makes it easy to integrate human judgement and AI into any ML application, regardless of whether it's run on AWS. You can use the Amazon A2I APIs to add human reviews to any machine learning application that uses a custom ML model built with Amazon SageMaker or other solutions. Learn more about how to use Amazon A2I with any machine learning model.