ExplainabilityTransparencyTrustworthinesscs.SE
Explainable AI for Software Engineering
Chakkrit Tantithamthavorn, Jirayus Jiarpakdee, John Grundy
December 3, 2020
International Conference on Automated Software Engineering
85 citations
3 influential
Key Takeaways
-
AI/ML models in software engineering lack transparency: The paper identifies that current AI/ML systems used to improve software quality and developer productivity are "not explainable and not actionable," which creates barriers to their real-world adoption by software teams.
-
Explainability is critical for practical deployment: The research demonstrates through three case studies that making AI models explainable—rather than treating them as "black boxes"—directly addresses adoption challenges and makes these tools more useful for actual software engineering practices.
-
Defect prediction benefits from transparency: The paper specifically highlights software defect prediction as an area where explainable AI techniques improve outcomes, suggesting that when developers understand why an AI system flags potential bugs, they can act on those predictions more effectively.
-
Explainability enables actionable insights: Beyond simply explaining predictions, the paper emphasizes that explainable AI must be "actionable," meaning developers need to understand not just what the AI found, but what they should do about it—a key requirement for responsible AI deployment in professional settings.
Abstract
Artificial Intelligence/Machine Learning techniques have been widely used in software engineering to improve developer productivity, the quality of software systems, and decision-making. However, such AI/ML models for software engineering are still impractical, not explainable, and not actionable. These concerns often hinder the adoption of AI/ML models in software engineering practices. In this article, we first highlight the need for explainable AI in software engineering. Then, we summarize three successful case studies on how explainable AI techniques can be used to address the aforementioned challenges by making software defect prediction models more practical, explainable, and actionable.