Technical Debt in AI-Enabled Systems: On the Prevalence, Severity, Impact, and Management Strategies for Code and Architecture

Abstract

Context: Artificial Intelligence (AI) is pervasive in several application domains, promising to be even more diffused in the next decades. Developing high-quality AI-enabled systems is urgently needed to avoid performance drifts, undesirable socially-relevant consequences, or a slow reaction to evolutionary changes. Recent work proposed AI technical debt, a potential liability for developing AI-enabled systems whose impact is limited to system qualities. While the problem of AI technical debt is gaining the attention of the software engineering community, scientific knowledge on the matter is still limited. Objective: In this paper, we leverage the expertise of practitioners to offer useful insights to the research community, aiming to enhance researchers' awareness about the detection and mitigation of AI technical debt. Our ultimate goal is to empower practitioners by providing them with tools and methods. Method: We developed a survey study featuring 53 AI developers to collect information on the practical prevalence, severity, impact, and the strategies practitioners apply to identify and mitigate AI-specific architectural and code debt. Results: The study's key findings reveal the multiple impacts that code and architectural debt may have on the quality of AI-enabled systems.Conclusion: We conclude the article by distilling lessons learned and insights for researchers.

Keywords: AI Technical Debt, software quality, Survey Studies, Software Engineering for Artificial Intelligence, Empirical Software Engineering