From traditional to model-based requirements specification with AI

Abstract

When people think of AI, they often think of "strong AI" - the AI of the future that is generally capable of what we humans can do. Today's "weak AI" is a collective term for technologies that help to transfer individual skills from humans to machines. This can help wherever very specific but simple, limited activities have to be repeated thousands of times by humans.
The topic of "from traditional to model-based requirements management" offers some potential where "weak AI" can provide practical support with little effort. One example: On the traditional side, there are really many individual textual requirements that contain implicit information from a modelling perspective, e.g. "If the airbag is triggered, then the crash warning lights should be triggered". On the other hand, there is the model-based requirements specification, in which precisely these dependencies are to be explicitly represented in models, e.g. by means of active chains. Here, people would have to sift through hundreds of thousands of requirements to extract such links. This is an ideal field of application for NLP (Natural Language Processing), a sub-area of "weak AI". In my presentation, I will show you how we use NLP to extract links from specifications in order to initialise models.