Before AI, There Was the Network: The Internet as a Pre-Intelligent System

Artificial intelligence is commonly described as a property of models: larger architectures, more data, and increasingly capable procedures for prediction, generation, and control. This paper begins from a different object: the network. Before contemporary generative AI existed, the Internet had already assembled several functions associated with intelligent behavior, including persistent memory, large-scale connectivity, retrieval, selection, feedback, and temporal continuity. This paper introduces the concept of a pre-intelligent system and represents network function through an eight-dimensional profile: N = (M, C, R, S, I, A, F, T) corresponding to memory, connectivity, retrieval, selection, interpretation, action, feedback, and temporal continuity. The framework distinguishes the early Web, search engines, algorithmic social systems, stateless language models, retrieval-augmented models, and tool-using agents as different functional configurations rather than points on a single intelligence scale. The paper further proposes that system intelligence is not equivalent to model capability (I_sys ≠ I_model). Effective capability depends on how models are coupled to memory, retrieval, tools, feedback, and time. This work extends the network-dynamic perspective introduced in the Kosuke Protocol, shifting the unit of analysis from isolated models toward the infrastructures and couplings in which intelligent behavior occurs.