Architecture of autonomous systems calls for symbolic AI integration and reveals a reliability gap

A new research paper by researcher Joseph Sifakis, deposited in the Arkhive platform under the Computer Science and Artificial Intelligence category, defines autonomous systems as the final stage and ultimate goal in the trajectory of AI development. The paper begins with an analysis of the core technical challenges that require combining two main schools: connectionist AI based on neural networks and symbolic AI based on rules and logic, with the integration of traditional systems-engineering concepts into AI architecture to ensure design and implementation rigor.

The study presents a comprehensive framework for designing and evaluating autonomous systems, based on a general agent architecture that defines system behavior through the integration of a set of cognitive functions. These functions focus onLong-term memory contains advanced knowledgeThe agent continuously consumes new experiences and information. The paper delineates the engineering challenges that impede the implementation of the core features of this architecture, foremost among them building the bridge between sensory data from the external environment and the structured data stored in memory, as well as handling the planning and decision-making processes related to achieving the agent’s specified goals.

The framework does not stop at the agent’s individual performance; it also extends to studying the coordination of collective work among multiple agents to combine individual intelligence with shared collective intelligence. In this context, the paper opens a highly sensitive file concerning the reliability of autonomous systems, emphasizing that the trust criteria used in traditional systems engineering are no longer sufficient here. While the industry has traditionally measured reliability through direct behavioral attributes by testing inputs and outputs, autonomous systems require the addition of a fundamental dimension linked to cognitive properties, upon which they depend.The system's validity and reliability depend on how the agent uses its advanced knowledge.While making decisions, which prompts them to propose new methodological pathways for evaluating and testing these cognitive properties.

The paper concludes with an explicit critical assessment that revealsA tangible, broad gap between the ambitious visionFor independent multi-agent systems and the actual state that technology has reached today, the article underscores that excessive optimism about the capabilities of independent agents overlooks the architectural complexities related to perception reliability and the difficulty of tightly linking sensory perception with organized knowledge, making the attainment of safe and dependable collective autonomy a matter that requires radical engineering solutions before large-scale field deployment.

This proposal outlines a practical roadmap for organizations and tech teams in the Gulf, Egypt, and the Levant, especially as regional investments accelerate toward process automation and the adoption of AI agents in sectors such as facilities management, financial services, logistics, and government platforms. The prevailing local approach often focuses on invoking language models without building layers of symbolic intelligence or scrutinizing the cognitive properties of memory architectures, which places systems at risk of unexpected errors in sensitive environments. Real change here requires redefining the target skills in the region’s labor markets, shifting from merely integrating models to attracting expertise in systems engineering and testing the reliability of cognitive decision-making, while reassessing timelines for full-autonomy projects to account for the true technology gap.