System analysis approaches for the development of advanced architectures

Abstract

Artificial Intelligence (AI) has established a new technological and organizational landscape across all industries, yet the process of creating AI architecture remains disjointed, intuitive, and domain-oriented. The proposed study introduces a System-Analytical Framework of Artificial Intelligence Architecture Design (SA-AI Framework), a collection of principles from systems analysis and multi-criteria decision-making intended to harmonize architectural design processes across heterogeneous AI domains. The paper is based on twenty-five current studies in explainable AI, neural-symbolic architectures, human-centered robotics, agentic systems, and domain-specific applications, such as healthcare, finance, energy, and education, that inform a comparative synthesis of architectural design principles. A Research Design Science Research (DSR) approach is used to model and test the SA-AI Framework conceptually across four stages: requirements analysis, identification of architectural alternatives, multi-criteria evaluation, and model validation. Results show that the framework offers a systematic, defensible method for constructing AI architectures, enabling greater exposure, flexibility, and decision-making effectiveness. It has cohesively unified the different models of AI, including the agentic AI, Explainable AI (XAI), Human-Centered AI (HCAI), and safety-driven architectures, providing both theory and practice to AI engineering and policy design. This paper concludes that system analysis helps design AI systems that integrate both technical and socio-technical systems to enhance the governance, reliability, and sustainability of intelligent architecture in the next generation.