Teaching Requirements Engineering for AI: A Goal-Oriented Approach in Software Engineering Courses

TL;DR

This study evaluates the use of Goal-Oriented Requirements Engineering, specifically the KAOS method, in teaching requirements elicitation for AI-based systems, highlighting its strengths and challenges in an educational setting.

Contribution

It demonstrates the effectiveness and usability of GORE, especially KAOS, for teaching requirements engineering for AI systems in software engineering courses.

Findings

GORE effectively captures high-level requirements like user expectations.

Students successfully applied the KAOS methodology with minor issues.

GORE's structured approach aids managing complex requirements but faces challenges in goal refinement and diagram complexity.

Abstract

Context: Requirements Engineering for AI-based systems (RE4AI) presents unique challenges due to the inherent volatility and complexity of AI technologies, necessitating the development of specialized methodologies. It is crucial to prepare upcoming software engineers with the abilities to specify high-quality requirements for AI-based systems. Goal: This research aims to evaluate the effectiveness and applicability of Goal-Oriented Requirements Engineering (GORE), specifically the KAOS method, in facilitating requirements elicitation for AI-based systems within an educational context. Method: We conducted an empirical study in an introductory software engineering class, combining presentations, practical exercises, and a survey to assess students' experience using GORE. Results: The analysis revealed that GORE is particularly effective in capturing high-level requirements, such as user expectations and system necessity. However, it is less effective for detailed planning, such as ensuring privacy and handling errors. The majority of students were able to apply the KAOS methodology correctly or with minor inadequacies, indicating its usability and effectiveness in educational settings. Students identified several benefits of GORE, including its goal-oriented nature and structured approach, which facilitated the management of complex requirements. However, challenges such as determining goal refinement stopping criteria and managing diagram complexity were also noted. Conclusion: GORE shows significant potential for enhancing requirements elicitation in AI-based systems. While generally effective, the approach could benefit from additional support and resources to address identified challenges. These findings suggest that GORE can be a valuable tool in both educational and practical contexts, provided that enhancements are made to facilitate its application.