I definitely fall into the camp of thinking of AI as augmenting human capability and capacity. – Satya Nadella
Our latest QA Webinar on ‘How AI is changing Defect Detection?’ almost had a similar theme. In case, you missed it live, here are the learnings from our AI webinar on QA.
The 1-hour live session had 90+ attendees registered, and 25% of them watching it live. The presenter, Vasudevan Swaminathan walked the attendees through the need for defect detection and how Artificial Intelligence is changing that.
It all started on 9 Sept 1947, the day when the world’s first bug was found and the defect evolution, over the years costing a fortune for companies, which called for the need for defect detection.
As the aphorism goes, ‘Software Testing proves the existence of bugs, not their absence”. The existence of defects and its serious impacts altered the ways of software testing methods. From Waterfall methodology to the current DevOps era, we have seen it all. The latest and what’s considered to be the future of software testing is AI. The transition to the adoption of AI in testing is no more a buzzword today.
AI relies on Data.
The more and diverse the data in forms of past defects, defect trends, etc. fed to the Machine Learning model, the better is the outcome.
Working together with people, AI can raise the data literacy of the entire workforce. While some AI applications rely only on machine automation, most complex business problems like Defect Detection require human interaction and perspective. And that is why we prefer to call it Augmented Intelligence rather calling it Artificial Intelligence.
Towards the end of the webinar, there were a few interesting questions from the attendees.
Q1: For supervised learning, you need to have labeled data. I think you’re suggesting that the QA results from previous sprints will be the source of the labels on the data. Can that help identify defects in future sprints, where new defects might be introduced?
A1: Labelling is definitely something we do. But we do that for the test cases (for now) and not for the results (because we have made a conservative assumption on mapping a test-case to a test-result). Since the test-cases are already mapped to existing issues (regression if any?), churn-rate of the codebase, we do sometimes gain access to forecasting future (we pick only the outliers for now). Yes, it helps, but we are working on using the available models to forecast the future and right now we have mixed results.
Q2: Can you give a real time example how AI help automate a basic login use case
A2: Login test automation – The primary benefit of AI or learning based solutions that Zuci focuses on is to generate test-cases, not to automate them. These are the use cases that Zuci focuses on,
- Finding a better way to promote one document (a test case) over another document
- Or grouping one document along with another,
- Or creating a document (generating a test case)
Q3: Which company has built Spider AI tool?
A3: Spider AI tool – ‘Spider AI tools’ are common crawler tools that you find in Search Engine Optimization and other areas. In the context of identifying test cases, Zuci uses a similar solution to recognize the optimal ones.