Information Molecule based Internal Communication Architecture for Brain-inspired AI Systems

Existing large-scale AI systems and multi-agent architectures usually adopt rigid communication mechanisms, such as fixed interfaces, message queues, remote procedure calls, and global state sharing. Although such mechanisms are efficient in single-task scenarios, they often exhibit high coupling, poor scalability, and insufficient adaptability in complex modular systems, especially in brain-inspired general artificial intelligence systems. Inspired by biological information transmission mechanisms (e.g., neurotransmitters, hormones, and cytokines), this paper proposes an information-molecule-based internal communication architecture for brain-inspired AI systems. In this framework, instructions, requests, emotional signals, state feedback, and regulatory commands are abstracted as information molecules that carry attributes such as molecule type, concentration, diffusion range, lifespan, and target receptor. Modules communicate by releasing, diffusing, binding, and automatically degrading molecules, which enables distributed, loosely coupled, and highly robust interaction without direct calls. Theoretical analysis shows that this architecture can effectively reduce inter-module coupling, improve dynamic adaptability, and provide a novel bio-inspired paradigm for constructing general AI systems, multi-role agent systems, and anthropomorphic AI architectures.