Existing theories of intelligence — psychometric g, emotional intelligence, and social intelligence — each capture one dimension of a more complete picture while treating the others as peripheral. We propose KAVI Space, a unified theory in which intelligence is defined as the volume of a three-dimensional space formed by three orthogonal axes: Knowledge (K), Awareness (A), and Vulnerability (V), from which Insight (I) emerges. The governing formula I = K × A × V has the multiplicative property that any dimension reduced to zero collapses intelligence entirely, regardless of the magnitude of the other two. The same error-correlation variable, ρ̂, that governs AI reliability in ensemble architectures governs the growth rate on each KAVI axis. Intelligence growth rate is proportional to (1 − ρ̂K) × (1 − ρ̂A) × n(1 − ρ̂V), where n is the number of independent perspectives accessed. Empirical calibration from a companion programme of 7,590 individually scored AI evaluations across six frontier models and three epistemological zones shows ρ̂ = 0.80 under single-axis compute scaling and ρ̂ = 0.19 under GAAS role-separated multi-axis activation — a four-fold reduction directly demonstrating the V-axis growth mechanism. The three-zone accuracy staircase (93.0% → 79.2% → 66.8%) replicates the KAVI prediction that axis suppression produces predictable, domain-specific failure. The GAAS (Generator–Auditor–Adversary–Synthesizer) architecture is shown to be formally isomorphic to KAVI Space itself.
Pandit et al. (2026) studied this question.