Evolving Intelligent Complex Systems via Intellicise Networks: Architecture, Technologies, and Pathways

Abstract: Future engineering infrastructures are evolving into large-scale, open, heterogeneous, and wirelessly interconnected complex systems. These systems present significant challenges in optimizing network resource utilization, managing high-dimensional information spaces, and accommodating diverse business requirements. Intellicise networks, characterized by Intent-driven operation, semantic-native capability, and distributed intelligence, offer a promising paradigm for enabling such intelligent complex systems. We provide a systematic exploration of future intelligent complex systems from the perspective of intellicise networks. Specifically, we propose a cross-domain intelligent communication network architecture based on intellicise networks, grounded in information theory, systems theory, game theory, and cybernetics. The architecture comprises a cross-layer organizational framework, multi-functional planes, and novel information flows. The cross-layer framework defines the vertical evolution from perception and cognition to decision, while the control, user, data, computation, intelligence, and security planes deliver horizontal intellicise capabilities. Moreover, data, knowledge, model, and task flows interconnect the various layers and planes, forming a closed-loop process that derives simplicity from high-level intelligene while concurrently pursuing enhanced. Building on this architecture, we review key enabling technologies, tracing their evolution from semantic extraction to intent understanding, from heterogeneous resource integration to self-configuration and self-optimization, from generative AI to agentic AI, and from embodied AI to symbodied AI. Additionally, we present a case study on intellicise networks for embodied agent communications and discuss representative applications and services for intelligent complex systems.

• The paper presents a novel cross-layer intellicise network architecture that integrates semantic communication with agentic AI for optimized task performance.
• It employs deep joint source-channel coding and intent recognition to enhance bandwidth efficiency and reduce latency in complex systems.
• The architecture’s multi-plane framework supports self-adaptive resource pooling and autonomous decision-making in heterogeneous network environments.

Evolving Intelligent Complex Systems via Intellicise Networks: Architecture, Technologies, and Pathways

---

Introduction: Rationale for Intellicise Networks in Complex System Engineering

Modern engineering infrastructures are rapidly transforming into open, large-scale, and heterogeneous systems characterized by extensive wireless interconnectivity, dynamic interaction patterns, and high degrees of autonomy. Traditional layered communication architectures, optimized for bit-level information fidelity, are increasingly inadequate for the nonlinearities, high dimensionality, and adaptive requirements of these complex systems. "Intellicise networks"—networks that are both intelligent and concise—emerge as a conceptual and architectural response to enable intent-driven operation, semantic-focused transmission, and distributed intelligence for future intelligent complex systems (2607.00316).

Figure 1: Overview of the paper’s structure and thematic progression.

---

Foundations of Intellicise Networks in System Science and Complex Systems

Intellicise networks are deeply rooted in system science, leveraging principles from information theory, systems theory, cybernetics, and game theory for both the modeling and management of complex networked infrastructures. This synthesis allows for semantic enhancement, resource organization, cross-layer information propagation, and robust service assurance within dynamic topologies.

Figure 2: System science-based intellicise networks—depicting the integration of complex network modeling, systems principles, and their operationalization in network semantics, resource organization, stability, and assurance.

Key challenges addressed include:

• Nonlinear information dissemination causing uncertainty and scalability bottlenecks
• High-dimensional information spaces leading to the curse of dimensionality
• Coupled, heterogeneous processing demands exceeding traditional packet-switched, best-effort approaches

Intellicise networks thus advance beyond Shannon’s classical framework, focusing on semantic information transmission and task-oriented optimization across distributed nodes.

---

Architectural Framework: Cross-Layer, Multi-Plane Intellicise Networks

A comprehensive cross-domain architecture is developed, grounded in cross-fused theoretical principles and articulated through a vertically integrated, multilayer framework (perception/cognition, access/transmission, interconnection/computing, decision/control) and horizontally differentiated functional planes (control, user, data, computation, intelligence, security).

Figure 3: Cross-domain intelligent network architecture—showcasing cross-fused theory foundations, vertical cross-layer evolution, horizontal multi-plane function, information flow interconnections, and enabling technology linkages to real-world services.

Cross-Layer Organizational Structure:

• Perception and Cognition Layer: Multi-modal physical world sensing, semantic extraction, and knowledge graph creation
• Access and Transmission Layer: Heterogeneous access, semantic-aware transport, dynamic resource allocation
• Interconnection and Computing Layer: Distributed orchestration of computing, communication, and storage resources; support for edge/fog/cloud paradigms
• Decision and Control Layer: Autonomous decision-making and task orchestration via intent interpretation and strategic alignment

Multi-Functional Planes:

• Control Plane: Cross-domain, intent-responsive management
• User Plane: Service, environmental, and AI task data forwarding
• Data and Computation Planes: Unified management of data, training samples, model weights, distributed computing resources
• Intelligence Plane: Full-lifecycle AI/ML deployment, inference, continuous adaptation
• Security Plane: Beyond link-layer, securing models, KBs, cross-plane coordination

Multi-Dimensional Information Flows:

• Data Flow: Raw and contextual observations
• Knowledge Flow: Semantic and relational structures extracted from data
• Model Flow: Distribution and updating of intelligence capabilities, supporting collaborative optimization
• Task Flow: Task-centric execution and adaptive feedback to align strategies with systemic objectives

This architecture enables closed-loop information and control cycles, supporting adaptive autonomous evolution.

---

Technological Pathways: Enabling Evolution of Complex Intelligent Systems

The realization of this architecture demands a portfolio of enabling technologies that progressively elevate intelligence, adaptivity, and robustness.

From Semantic Extraction to Intent Understanding

• Semantic Extraction: Employing deep joint source-channel coding (JSCC) and generative prior-based models for efficient semantic embedding of source data, dramatically improving bandwidth utilization and semantic relevance [bourtsoulatze2019deep] [yang2024swinjscc].

  Figure 4: SwinJSCC framework—Swin Transformer-based JSCC for image semantic encoding and transmission.

• Intent Understanding: Incorporating intent recognition, critical information prioritization, and intent-guided policy adaptation using advanced inference and neuro-symbolic AI [thomas2022neuro] [zhang2026towards]. Differentiated resource allocation aligns network operation to user/service demands, maximizing task utility.

From Heterogeneous Resource Integration to Self-Configuration and Optimization

• Heterogeneous resource pooling: Dynamic allocation across communication, computation, sensing, and storage resources, supported by optimization and adaptive multi-agent DRL methods [xu2025heterogeneous].

  Figure 5: CTDE-MADDPG-based resource allocation with satellite agents for joint offloading and resource orchestration.

• Autonomous configuration: Edge SLMs and agentic frameworks synthesize and maintain optimal network topologies and operational parameters, reducing centralized bottlenecks and supporting cognitive self-optimization loops [lwin2026performance].

  Figure 6: Intent-based self-configuration framework, where LLM and SLM agents automate network orchestration.

From Generative AI to Agentic AI

• Generative AI: Underpins semantic encoding, digital twin synthesis, and predictive augmentation in resource-constrained or partially observable environments [zhou2026digtwinai] [gao2026semstediff].

  Figure 7: SemSteDiff—Diffusion model-based coverless semantic steganography for covert, semantically valid communications.

• Agentic AI: Advances closed-loop orchestration, intent-to-action translation, and multi-agent collaboration with operational tool-use capabilities, supporting autonomy at scale [zhang2026agenticsurvey] [meng2026secure].

  Figure 8: Agentic AI-enabled framework for secure, coverless steganographic semantic communication with coordinated agentic generation and recovery.

From Embodied AI to Symbodied AI

• Embodied AI: Closes the gap between intelligent control and environment, merging vision, language, and action within physical feedback loops to enable agentic action [zitkovich2023rt].

  Figure 9: RT-2 framework—Vision-language-action co-finetuning for end-to-end robot control.

• Symbodied AI: Human-machine cognitive alignment, leveraging collaborative reasoning and real-time human feedback for secure, aligned, and adaptive system evolution [sun2025collabvla].

  Figure 10: CollabVLA—Human guidance and self-reflective reasoning for collaborative physical task execution.

---

Case Study: Embodied Agent Communication via JSCCC

A Joint Semantic Cognition-Communication-Control (JSCCC) scheme is implemented for embodied agent scenarios. It enables direct mapping from sensed environment to semantic representation, transmission in the semantic domain, and translation to actuator control—bypassing bit-wise image transmission, object detection, and other classical inference modules.

Figure 11: JSCCC architecture for unified perception, communication, and control in embodied agent systems.

Comprehensive simulations demonstrate that JSCCC achieves marked improvements in task success rate and latency under constrained bandwidth, outperforming both traditional and cascaded semantic communication pipelines.

Figure 12: JSCCC surpasses classic and cascaded semantic pipelines in task success rate and end-to-end latency.

---

Implications, Numerical Results, and Theoretical Outcomes

Strong empirical claims: The JSCCC scheme achieves superior task performance and significantly reduced end-to-end latency relative to cascaded paradigms under bandwidth-limited conditions, establishing the architectural merit of end-to-end semantic processes for embodied intelligence.

Theoretical implications: The cross-fused architectural principles, multi-plane organization, and multidimensional information flows provide a unified model for reconciling the requirements of dynamic, large-scale, and heterogeneous complex systems. By grounding control and communication primitives in intent and semantics, system efficiency, adaptability, and self-optimization are fundamentally improved.

Future-oriented insights: The systematic roadmap outlined enables scenarios ranging from integrated SAGSIN and IIoT to digital twins and collaborative robotics. This will catalyze new research trajectories in cross-layer compatibility, model interpretability, unified security/trust architectures, and universal scenario generalization.

---

Conclusion

This work establishes a rigorous, multi-theoretic framework for evolving intelligent complex systems via intellicise networks. By synthesizing advances in semantic communication, agentic AI, and multi-plane architectural design, the proposed approach provides a blueprint for scalable, task-adaptive, and secure future engineering infrastructures. Fundamental challenges remain—including protocol compatibility, interpretable intelligence, security/privacy, and generalizability across diverse domains—but the architecture and methodologies defined herein offer a coherent pathway for both theoretical and practical advancement in next-generation networked intelligent systems (2607.00316).