Continuous Software Engineering Practices in {AI/ML} Development Past the Narrow Lens of {MLOps}: {A}doption Challenges

Abstract

Background: Continuous software engineering practices are currently considered state of the art in Software Engineering (SE). Recently, this interest in continuous SE has extended to ML system development as well, primarily through MLOps. However, little is known about continuous SE in ML development outside the specific continuous practices present in MLOps. Aim: In this paper, we explored continuous SE in ML development more generally, outside the specific scope of MLOps. We sought to understand what challenges organizations face in adopting all the 13 continuous SE practices identified in existing literature. Method: We conducted a multiple case study of organizations developing ML systems. Data from the cases was collected through thematic interviews. The interview instrument focused on different aspects of continuous SE, as well as the use of relevant tools and methods. Results: We interviewed 8 ML experts from different organizations. Based on the data, we identified various challenges associated with the adoption of continuous SE practices in ML development. Our results are summarized through 7 key findings. Conclusion: The largest challenges we identified seem to stem from communication issues. ML experts seem to continue to work in silos, detached from both the rest of the project and the customers.