Most of the AI breakthroughs in the past 4 years have resulted from deep learning. Can deep learning help software developers? Most work in this field has been devoted to metrics and estimates. Metrics are shadows of snowflakes, and thus not particularly informative for understanding the purpose of a system or the behavior of a component. Improving estimation, while an interesting application and a laudable goal, likewise does not directly help developers create better software faster or manage the growing complexity of real-world systems. Can we apply machine learning (ML) in ways that help guide development, inform software archaeology, and reduce effort for developers? If so, can it help us take on the systems-of-systems-of-systems of the future, where million-line code bases will be considered quaint? After a brief survey of existing work in the area, we'll investigate the question "what does software look like?" and examine the more interesting question of "what does good software look like?" Then we’ll use these insights to explore how ML can enable developers to create and manage ever-larger software systems. No ML experience required. Lots of pretty pictures.
