Modeling and AI: Friends or Foes? (Keynote)

摘要

AI is infiltrating all industries and the software industry is no exception. In fact, as of today, there are already several initiatives claiming the (prospective) applications of AI in the different phases of the software development lifecycle [1, 4, 5], from the requirement analysis and design to the development, testing, deployment and maintenance. But, is this a fad or really the future of software development? And, if so, what role could modeling play in this future? In this talk, we will review promising applications of AI techniques in conceptual modeling, business process modeling, systems modeling and, in general, any branch of model-driven engineering; and discuss the many challenges that remain to be solved before we see the first real and usable AI-enhanced modeling technique or IDE. Among other examples, we will discuss how neural networks can kill model transformation languages, how virtual modelers could become our ideal 'pair designer' or the use of graph kernels to cluster modeling artefacts [2]. The recently created Modelia initiative aims to support research activities around these topics. But make no mistake, AI may need us more than we need AI. The future of AI is model-based. During the talk, we will also cover how modeling can help to bring AI to the masses and simplify the fragmented landscape of AI libraries, platforms and tools. As an example, we will see how DSLs (Domain-Specific Languages) can facilitate the definition and generation of AI pipelines and components, such as chatbots [3], by enabling their specification at a higher abstraction level.