Cognitive enterprises and intelligent workflows are poised to take centre stage across business platforms to drive greater business outcomes. A cornerstone of any intelligent workflow will always be AI infused applications and accelerators. The question then arises on the quality engineering and the test rigour that the AI Infused solutions undergo. From the data through the APIs and to the application front end - how is the AI infused solution and workflow being engineered and tested for quality?
Testing of AI infused business solutions tend to see depth on two ends of a typical software development life cycle- Unit testing and business acceptance testing with loosely defined continuity in between. The IBM Automation, Quality Engineering and Test Practices and IBM Research have been working to address this gap. We introduce the Full AI-cycle Test approach. Starting with the data, the FACT approach provides automation solutions to baseline and establish the quality of the training and test data, provide insights on the errors in data labelling and content to identify potential skews in the data that could influence modelling decisions upfront. Ingesting the model and validating the model performance covers the model testing aspects with automated insights. From business coverage perspective, the functional test design uses proven cognitive design algorithms that are enhanced to cover the predictions and influencing data to provide ML test coverage. The solution enables build out of comprehensive test coverage for the business tests and the AI predictions and enable testing of the data, model and application in entirety. Thus providing Full AI-cycle Testing. We share this concept and provide a demo of the component solutions for realising the appraoch.