AWS is 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 at scale. Amazon A2I removes the undifferentiated heavy lifting associated with building and managing expensive and complex human review systems, so you can ensure your ML models produce accurate predictions. Amazon A2I enables humans and machines to do what they do best by easily inserting human judgment into the ML pipeline.
Amazon A2I provides built-in human review workflows for common ML tasks such as content moderation and text extraction from documents, in combination with Amazon Rekognition and Amazon Textract. You can also create your own human review workflows for ML models built with Amazon SageMaker or with any on-premises or cloud tools via its API.
Amazon A2I also gives you the ability to work with your choice of human reviewers. You can use your own reviewers or choose from a workforce of over 500,000 independent contractors who already do ML-related tasks through Amazon Mechanical Turk. If your ML application requires confidentiality or special skills, you can use workforce vendors that are experienced and pre-screened by AWS for quality and security procedures.
Amazon A2I gives you the flexibility to incorporate human reviews based on your specific requirements. You simply set the business rules (how confident an ML model is in its predictions) to decide which predictions to use automatically or route to a human for validation.
When data comes through your ML pipeline, you can decide, based on your requirements, if the prediction meets your minimum confidence threshold. For example, if you want to extract a unique identifier like a Social Security Number (SSN) from numerous documents, they must be absolutely correct for your downstream application to be successful. You might set your threshold high (such as 99%) to achieve desired model accuracy.
Other data in your application might be secondary to the SSN and not require the same level of scrutiny. Therefore, you might set the threshold lower than SSN extraction. Amazon A2I gives you the flexibility to set different threshold levels based on your needs. Additionally, you can choose to randomly sample ML model outputs for human review so you can regularly evaluate if the model is still performing the way you intended.
Amazon A2I also works well with other services. You can use direct integrations with Amazon Textract and Amazon Rekognition, or use a custom workflow in Amazon A2I for human-in-the-loop validation with Amazon Comprehend, Amazon Translate or other AWS AI services. You can also use the Amazon A2I API to add human reviews to any ML application that uses a custom ML model built with Amazon SageMaker or any other on-premises or cloud tool.
How Amazon A2I works with Amazon Textract
The following diagram shows how Amazon A2I integrates with Amazon Textract. Documents go through Amazon Textract and, based on your business rules, Amazon A2I sends low-confidence predictions to humans to review. You can store these results in an Amazon S3 bucket for your client application to use and be confident of their accuracy.
For example, a mortgage application has hundreds of documents associated with the application process. Each page has different information, signatures, and dates, which are scanned and uploaded into your system. At times, these document scans can be low-quality and require humans to review them to make sure the data is accurate and complete. With Amazon A2I, you can process these documents through Amazon Textract and send the low-confidence or hard-to-read documents to a human reviewer.
Other times, the prediction is at an acceptable confidence level for your application. When this occurs, Amazon A2I returns the prediction to the client application immediately, without a human review. You set the thresholds as needed for your use case.
Using Amazon Textract and Amazon A2I
Belle Fleur Technologies, an AWS Partner Network (APN) Advanced Consulting Partner, believes the ML revolution is altering the way we live, work, and relate to one another, and will transform the way every business in every industry operates.
Belle Fleur knew Amazon Textract was the right solution when working with financial institutions. Amazon Textract allowed them to go through vast quantities of documents and extract the relevant data their clients needed. However, they spent significant time reviewing the more nuanced and critical data manually.
Adding Amazon A2I to the equation was a good fit for their customers. Amazon A2I decreased the time spent building human validation and pulled all the relevant extracted data into one place in an easy-to-understand workflow so reviewers could quickly and easily review ML outputs. Tia Dubuisson, President at Belle Fleur, says, “Amazon A2I not only provides us and our customers peace of mind that the more nuanced data extracted is reviewed by humans, but it also helps train and improve our ML models over time through continuous auditing and improvement.”
For more information about other ways our customers are using Amazon A2I in their ML workflows, see the Augmented AI Customer page.
To get started, sign in to the Amazon A2I console or search for Amazon Augmented AI on the AWS Management Console. For more information about creating a human review workflow, see Create a Flow Definition.
About the Authors
Andrea Morton-Youmans is a Product Marketing Manager on the AI Services team at AWS. Over the past 10 years she has worked in the technology and telecommunications industries, focused on developer storytelling and marketing campaigns. In her spare time, she enjoys heading to the lake with her husband and Aussie dog Oakley, tasting wine and enjoying a movie from time to time.
Anuj Gupta is the Product Manager for Amazon Augmented AI. He focusing on delivering products that make it easier for customers to adopt machine learning. In his spare time, he enjoys road trips and watching Formula 1.