AWS Helps ML Devs Streamline Human Review with Amazon A2I

Amazon Web Services (AWS), now offers a new managed service called Amazon Augmented AI. This allows developers to improve the accuracy and quality of their machine-learning models with human oversight.
Amazon A21 offers developers pre-configured workflows to help them in situations where their machine learning predictions require extra human verification. Some machine learning applications, such as those that scan financial documents or perform facial identification for law enforcement, require extremely high levels of accuracy. Developers are required to bring in human beings, either from a third-party auditor, or from within the company, to validate their machine-learning predictions.
These so-called “human review system” can be costly to build and time-consuming for maintenance.
AWS explained that human reviews can be difficult and costly to build and run at scale. This includes multiple workflow steps, custom software to manage human review tasks, results and recruitment, as well as managing large groups of reviewers. “Developers sometimes spend more time managing the human review process that building the intended app, or they have no choice but to forego human reviews, which can lead to less confidence when deploying applications that use machine learning.
Amazon A2I automates the human review process for AWS services like Textract (character recognition) and Rekognition (image recognition). It comes with 60 pre-built workflows for common tasks such as speech transcription, image analysis, and content moderation.
Developers have the option to initiate a human review for any machine-learning prediction below a certain confidence level. They can choose the number of reviewers needed for each prediction, and how they will be sourced. This includes whether they are in-house, outsourced to AWS partners, or crowd-sourced via Amazon Mechanical Turk. Amazon A2I allows developers to post instructions to reviewers to ensure consistency and allow them to make adjustments as necessary to correct recurring reviewer mistakes.
You can find more information about Amazon A2I here.