The Broken Mirror Architecture: A Non-Binary Framework for Affective Artificial Intelligence Using Irregular Reflective Layers

Conventional artificial intelligence and deep neural paradigms inherently treat error as an objective discrepancy metric to be iteratively eliminated via gradient descent and backpropagation. Under this traditional formulation, contradictory, noisy, or unreciprocated affective signals induce representational drift and parameter degradation within continuous vector spaces. This paper proposes the Broken Mirror Architecture (BMA), an affective, neuromorphic computing framework rooted in physical reservoir computing and non-Euclidean tactile topologies. Inspired by the optical physics of a fractured mirror—wherein discontinuous shards generate multi-angle phase interference rather than a singular projection—BMA replaces regular sensor lattices with an irregular spatial distribution of sensing nodes interconnected by heterogeneous waveguides of non-uniform path lengths. The architecture processes signals across an odd number of alternating reflective strata (7 or 13 layers), oscillating between spatial coherence (order) and non-linear phase scattering (disorder). Rather than decaying or acting as destructive white noise, intermediate disorder layers feed downstream chaos into subsequent dispersion stages; non-resonant noise modes undergo destructive phase interference (the principle of dynamic phase cancellation), allowing the terminal 7th stratum to operate as a self-organizing attractor basin. In this terminating layer, competing coherent reflections are evaluated ('good vs. better'), and the optimal affective output is resolved via Winner-Take-All phase resonance without external supervised labeling. We redefine error not as algorithmic failure, but as an unreciprocated transmission (a broadcast lacking a positive reflective pulse return), elevating error to an intrinsic topological primitive that quantifies affective distance. We formalize dynamic affective proximity metrics, an iterative resonance test framework, an autonomous attention-switching mechanism triggered by error accumulation thresholds, and provide numerical simulation validation.