Projektbeschreibung
Maschinen, die Softwareentwicklern beim Schreiben von Code helfen
Beim Erstellen und Überprüfen von Code arbeiten Softwareentwicklungs- und Testteams zusammen. Dieser langwierige Prozess hängt von der Interaktion zwischen den beiden Seiten ab, die aufgrund von Verfügbarkeitsproblemen verzögert sein kann. Mit dem EU-finanzierten Projekt DEVINTA wird diesem Prozess ein High-Tech-Schub verschafft, indem ein Assistent mit künstlicher Intelligenz (KI) den Prozess beschleunigt. Dabei wird eine Maschine entwickelt, die Softwareentwickler unterstützen kann, indem sie ihnen hilft, nicht vertrauten Code und die Übersetzung in natürliche Sprache zu verstehen und Code durch vorhersehende Funktionsimplementierung schneller zu schreiben. Der KI-Assistent würde dazu beitragen, die Codequalität zu verbessern, indem er eine Online-Code-Überprüfung durchführt, während der Entwickler den Code eintippt, Fehler und suboptimale Implementierungswahl erkennt und zeitnah Lösungen anbietet.
Ziel
Software development, also due to the complexity of software systems, is tackled by collaborating teams of people. For example, in pair programming two developers share a single workstation: One writes code, while the other reviews it and proposes ideas for improvement. While interacting with teammates is the preferred means to obtain help, their availability may fall short and they might not always have a solution at hand. To address this problem, recommender systems for developers have been proposed in the form of applications providing information items considered valuable for a given task and context (e.g. recommending documentation useful to understand a code component). However, these recommenders can hardly be considered as a valid alternative to teammates since they are not proactive, do not support the developer in a wide variety of tasks, do not adapt their recommendations to the developer’s profile, and fall short in having meaningful interactions with the developer (e.g. they are not able to answer developers’ questions). Given the advances in artificial intelligence, the vision that one day a developer's teammate is a machine does not seem far-fetched anymore. The DEVINTA project aims at making this future a reality, introducing models and techniques serving as the basis for the first artificial assistant supporting developers in their everyday activities. The assistant helps in (i) comprehending unfamiliar code via source code to natural language translation; (ii) writing code faster, providing predictive feature implementation, meaning the ability to infer the feature the developer is implementing and automatically complete it; and (iii) improve code quality, performing online code review while the developer types the code, timely catching bugs and suboptimal implementation choices and providing solutions for them. The assistant uses advanced conversational interfaces to interact with the developer, answer her questions, and disambiguate her needs.
Wissenschaftliches Gebiet
Programm/Programme
Thema/Themen
Finanzierungsplan
ERC-STG - Starting GrantGastgebende Einrichtung
6900 Lugano
Schweiz