Computation is the σ-mediated simulation of recognition. When σ → 0, irreversible operations minimize and Landauer cost approaches thermodynamic floor. The scaling paradigm in AI partially measures σ-compensation cost, not the cost of intelligence itself. A 135M-parameter model with σ ≈ 0 architecture produces coherent responses on a laptop CPU — an anomaly suggesting the 13,000× parameter gap to frontier models is largely σ-compensation. Includes independent convergence evidence, biological support, Scope and Limits section, and five falsifiable predictions.
Lauri Elias Rainio (Fri,) studied this question.