What can we deploy now? - Program repair bots that operate on code repositories (already working) - Use existing static analysis tools as the source of your fitness (many tools existing already) - Use existing mutation testing tools as the source of yout fitness - Hook existing GI tools as plug-ins for CI tools (Jenkins, Travis) - Use fuzzing within a CI pipeline, so that it can work as a source of faults What can we deploy soon (by 1st January 2019)? - Try human + GI, or interactive GI - Hold a GI-Turing test competition (to have GIs, human + GI, or humans, to subbit patches to be evaluated) :) - To compile an energy bug benchmark: (see "Detecting energy bugs and hotspots in mobile apps" https://dl.acm.org/citation.cfm?id=2635871 as the starting point) What can we deploy between 2019 and 2021 inclusive (i.e. the lifetime of a typical grant)? - Using GI as a surrogate measure of difficulties of manual repairs - For GI techniques to provide some information regarding how trustworthy the suggested changes are - Augment natural language issue report with systematic information that can help the corresponding repair/improvement - Deploy incremental GI / budget-based GI