Abstract
Artificial intelligence (AI) and machine learning (ML) is becoming commonplace in numerous fields. As they are often embedded in the context of larger software systems, issues that are faced with software systems in general are also applicable to AI/ML. In this paper, we address ML systems and their characteristics in the light of software maintenance and its attributes, modularity, testability, reusability, analysability, and modifiability. To achieve this, we pinpoint similarities and differences between ML software and software as we traditionally understand it, and draw parallels as well as provide a programmer’s view to ML at a general level, using the established software design principles as the starting point.