Generative AI for Self-Adaptive Systems: State of the Art and Research Roadmap

Analysis

Why This Paper Matters

Self-adaptive systems (SASs) are critical for handling dynamic environments, but their traditional feedback loops (MAPE-K) often struggle with complex, unstructured data and reasoning tasks. Generative AI, especially large language models (LLMs), has demonstrated remarkable capabilities in data comprehension and logical reasoning, which align closely with the core functionalities of SASs. This paper is timely because it bridges two rapidly evolving fields, offering a structured overview of how GenAI can enhance SASs and what challenges remain.

The significance lies in its comprehensive survey approach, drawing from four research fields to provide a holistic view. By categorizing benefits into autonomy enhancements and human-on-the-loop improvements, the paper gives practitioners a clear framework for where GenAI can add value. The research roadmap is particularly valuable, as it not only highlights challenges but also proposes mitigation strategies, making it actionable for future work.

Technical Contributions

• Literature synthesis: Gathers and filters literature from four distinct fields (SAS, GenAI, human-computer interaction, and software engineering) to provide a cross-domain perspective.
• Benefit categorization: Organizes potential benefits into two main categories: (i) enhancements to SAS autonomy centered on MAPE-K feedback loop functions (monitoring, analyzing, planning, execution), and (ii) improvements in human-SAS interaction within human-on-the-loop settings.
• Research roadmap: Outlines key research challenges for integrating GenAI into SASs, followed by practical reflection on current shortcomings of GenAI and proposed mitigation strategies.
• Structured analysis: Provides a systematic approach to understanding the alignment between GenAI capabilities and SAS requirements, which is missing in existing literature.

Results

The paper does not present quantitative experimental results but provides a qualitative synthesis of the state of the art. It identifies that GenAI can enhance each MAPE-K function: e.g., LLMs for monitoring (anomaly detection), analyzing (root cause analysis), planning (adaptive strategies), and execution (code generation). For human-on-the-loop, GenAI improves explainability and trust. The roadmap highlights challenges such as reliability, safety, and real-time constraints, with mitigation strategies including hybrid approaches and verification techniques.

Significance

This paper serves as a foundational reference for researchers and practitioners working at the intersection of GenAI and self-adaptive systems. By clarifying the potential benefits and challenges, it accelerates the adoption of GenAI in SASs, which are critical for autonomous systems in domains like robotics, IoT, and cloud computing. The roadmap guides future research, helping to avoid pitfalls and focus efforts on high-impact areas. As GenAI continues to evolve, this work provides a stable framework for evaluating new developments.