Adverts
Microsoft researchers have created an Artificial Intelligence (AI) solution that they believe will help programmers debug applications faster and more accurately.
Microsoft creates AI tools that help programmers. BugLab is an AI that uses a “hide and seek” game model to create Generative Adversarial Networks (GAN).
However, Miltos Allamanis (principal researcher) and Marc Brockschmidt (senior principal research manager) described the research in a blog; explaining how they created two networks and pitted them against each other; in a similar way to hide and seek.
Adverts
However a network is; designed to introduce bugs into existing code, large and small; while the other is; designed to detect them. The AI improves to the point that it can identify hidden bugs in the real code as the game progresses and both “participants” improve.
The two models were; trained in a self-supervised way on “millions of code snippets” without labeled data; according to the researchers.
Adverts
However despite the fact that the goal was to create a program that could detect arbitrarily complex bugs; the researchers claim that these are still “beyond the reach of modern AI methods”.
Instead, they focused on bugs which are; commonly found; such as incorrect comparisons, incorrect Boolean operators, incorrect variable usages, and other issues.
Microsoft creates AI tool that helps programmers
However the test was; carried out in Python and once the application was; trained, it was time to put it to the test in the real world.
“To demonstrate performance, we typically annotate a small set of crash bugs in the Python Pаckаge Index with these bugs and show that models plotted with our “hide and seek” method are up to 30% better than other models; like other detectors traced with randomly inserted bugs,” according to the blog.
However, the results were “promising”, according to the duo; with about a quarter of bugs (26%) being found and fixed automatically. Additionally, 19 previously unknown bugs were discovered; among the bugs found.
Ultimately, the researchers concluded that much more training is needed before such a model can be; used in practice.
