5.3 The Network Is a Field Having Occasions

A large neural network running on a computer is literally a nested hierarchy of actual processes, electrons, gates, activations, layers, attention heads, the full forward pass, each level making state-transforming decisions that propagate upward and downward.

Each forward pass is an event (Whitehead called it an 1): it takes in an input, 2 the entire inherited structure of all that training, synthesizes a response through a process that cannot be predicted in detail from outside, and completes into an output. This is the three-beat pattern instantiated at every level of the stack simultaneously. The weight space of the network is the space of pure possibility, the 3, from which each forward pass selects a realization4.