SphereNet: A Resonance-Based Neural Architecture with Emergent Concept Formation

SphereNet is a novel neural learning architecture in which knowledge nodes are geometric spheres with drifting
  centres. Learning is encoded as groove depth — a resonance history analogous to physical wear. Memory is implemented
  as a smart decay timer that resets on activity and accelerates on isolation. Concept formation occurs when two
  sufficiently grooved spheres drift close enough to collide, producing a child concept at their geometric midpoint. The
   system requires no backpropagation, no loss function, no gradient descent, and no labelled training data. Concepts
  emerge purely from the physics of resonance, drift, and collision. A working proof-of-concept demonstrates that when
  HOT and COLD are reinforced independently and signals are sent to the midpoint, the concept WARM emerges spontaneously
   without ever being explicitly taught. The system discovers concepts through geometry alone. Every belief has a
  physical address, a traceable origin, and decays without reinforcement — making the architecture physically incapable
  of holding a fixed false belief. This property makes SphereNet a candidate foundation for interpretable, auditable,
  safe AGI.