Towards domain-specific automated testing via behavioural cloning

Abstract

When setting out a research roadmap for software testing, Bertolino [1] presented four dreams, one of which was 100% automatic testing. Fifteen years later, the dream has not been realised but the promise of artificial intelligence techniques brings us closer than ever before. In this paper, we propose that one way to achieve this goal is to leverage the commonalities that exist amongst domain-specific applications. That is to say that whilst every application within a particular domain is arguably unique, they all share a considerable overlap in terms of features. We propose an approach based on Behavioural Cloning, an AI technique whereby an agent observes traces by an expert and attempts to carry out domain-specific tasks in previously unseen contexts based on those traces. Using online stores as a case study, we discuss initial investigations into this idea, present results and identify a roadmap going forward.