Title: Cognitive Divergence Theory of AI Adoption
Tags: skill-divide, work-and-productivity
---
The central paper of the trilogy asks what happens to the people who can work productively with AI, and to those who cannot. Its claim is that productivity gains from AI are proportional to the cognitive ability of the user, so adoption produces a categorical split and not a gradual one. The paper draws on literature about cognitive ability, skill-biased technological change and cognitive load, and on reported usage patterns among software developers.

A productivity model with four calibrated parameters is run as a Monte Carlo simulation on 10,000 synthetic agents over 24 months. In the simulation the Gini coefficient rises from 0.17 to 0.55, the top quartile roughly triples its output, the bottom half collapses, and the ratio between the two reaches about 71 to 1. The **dual-effect framework** reconciles this with studies that found AI to be an equalizer: on simple tasks over a few months AI levels performance, while on complex tasks over one to two years it amplifies existing ability, and economic value migrates towards the second kind of task.

The projection for 2027-2028 is a three-tier structure: **Homo Symbioticus** (20-25% of knowledge workers), a precarious middle, and a displaced majority. Institutions change the form of the outcome (unemployment in the EU, wage collapse in the United States, containment by redistribution in the Nordic countries), and Eastern Europe faces an added loss of talent. Twelve numerical predictions are given with thresholds for refutation: four failures falsify the theory.

A simulation of the Cognitive Stimulation Mechanism of the MEG standard reduces the gap to about 14 to 1, but helps only users in a middle band and harms the lowest. An adaptive version, MCS 2.0, which follows the trajectory of the conversation instead of estimating IQ, is outlined as a research agenda.