Applied AI matters: AI4Code

The marriage of Artificial Intelligence (AI) techniques to problems surrounding the generation, maintenance, and use of source code has come to the fore in recent years as an important AI application area1. A large chunk of this recent attention can be attributed to contemporaneous advancements in Natural Language Processing (NLP) techniques and sub-fields. The naturalness hypothesis, which states that "software is a form of human communication" and that code exhibits patterns that are similar to (human) natural languages (Devanbu, 2015; Hindle, Barr, Gabel, Su, & Devanbu, 2016), has allowed for the application of many of these NLP advances to code-centric usecases. This development has contributed to a spate of work in the community --- much of it captured in a survey by Allamanis, Barr, Devanbu, and Sutton (2018) that focuses on classifying these approaches by the type of probabilistic model applied to source code. This increase in the variety of AI techniques applied to source code has found various manifestations in the industry at large. Code and software form the backbone that underpins almost all modern technical advancements: it is thus natural that breakthroughs in this area should reflect in the emergence of real world deployments.