Artificial Intelligence based on Machine Learning, and in particular Deep Learning, is today the fastest growing trend in software development, and literally used in all other research disciplines, with a very high impact on the modern society. However, a wide use of AI in many systems, in particular dependable systems, is still far away of being widely used. On the one hand there is a shortage of expertise, on the other hand the challenges for managing AI-based complex and dependable systems are enormous, though less known, and in general underestimated. Some aspects of these challenges are based on management of resources, including computational, data storage capacity, performance, and real-time constraints. Introduction of AI-based components, i.e. components that includes AI algorithms, require significant changes in system and software architecture, and its successful deployment is based on many architectural decisions and on changes of the development process.

This talk discusses some of these challenges, illustrate a case of Cyber-physical systems, and gives some ideas for new research in software engineering inducing software architecture, i.e. for AI engineering.