As far as technology is concerned, artificial intelligence (AI) & machine learning (ML) are the buzzwords today. However, in certain circumstances where AI models rely completely on machines to do the work, the results are less than perfect, which makes it far more difficult to enhance ML models. One way to overcome the challenges encountered is to engage both computers & humans to work side-by-side to complete tasks.
Amazon Augmented AI (Amazon A2I), a new service from Amazon Web Services (AWS) aims to ease this process with a step-by-step workflow points plan.
Workflow guides human reviewers through a structured framework — the benefits are two-fold:
1st: developers can satisfy their clients by providing accurate assessments, & the other is that any machine employed as part of the process, ‘learns’ with every step, which after all is the objective of Machine Learning.
To give an example of the type of situation where this service would assist developers & businesses, consider an insurance company’s policy approval department. Handwritten, or digitally annotated documents are usually scanned to a format suitable for a machine to review. However, very often poor handwriting renders documents illegible for machines, so these documents require a human reviewer, but if a human reviewer is employed as part of the process, specialized software might be needed in order to make the process uniform & ensure a predefined result. This is exactly the type of situation that will benefit from A2I’s workflow procedure.
Additionally, Amazon has made several other auxiliary services available to enhance its new service. These include Amazon’s TextExtract service, AWS Marketplace Vendor & Amazon Mechanical Turk. A2I supports review processes for documents, images & videos, but if additional services are used as part of the process, consumers will need to pay for each, over & above the costs of the A2I offering.
For new users Amazon has a free tier, which offers to provide services for the first 500 items free of charge. Thereafter, Amazon provides consumers with a clear pricing structure determined by the nature of the document. For example, plain text is the least expensive, while documents that include tables or forms are more expensive respectively.
If consumers use their own internal employees in the review process, the costs are reduced, but the system will work just as well if the workflow points are correctly implemented. Some consumers are advised to use a combination of reviewers, both internal & external to ensure the best results. An example of an instance where this type of dual review might be necessary, is where a consumer allows uploads of images or videos from the general public.
A machine based algorithm might be used to vet the resultant uploads, & internal staff will be used to work on those items which fail to pass the machine’s test. This workload could well benefit from a process where each item is subject to double review before it goes live, ensuring that the consumer is safe to publish the items, especially considering that the volume of uploads might be in the hundreds of thousands daily.
