Title: The Psychological Ceiling of AI Adoption
Tags: skill-divide, human-ai-dialogue
---
The first paper of the Cognitive Divergence trilogy argues that the limit to the adoption of artificial intelligence is social, not technical. Its thesis is the **psychological ceiling**: comfortable communication breaks down when the gap in intelligence between two parties exceeds roughly 20 to 30 IQ points, an observation taken from Leta Hollingworth's work on gifted children, and current language models already sit at the upper edge of that range for most users. Beyond it, each gain in capability is said to shrink the mass market instead of enlarging it.

Three mechanisms are given. A highly competent agent perceived as low in warmth triggers a sense of threat (the warmth-competence model of Fiske, Cuddy and Glick). Users protect their cognitive status by keeping to simple tasks. And an intelligence that reproduces the outputs of thought without lived experience produces a "cognitive uncanny valley". The paper says that its IQ figures for models are discussion tools about perceived competence in verbal tasks, not psychometric measurements.

From this follow two projections. The capability curve of mass-market AI is expected to move from exponential growth to a socially imposed plateau. The market splits into two tiers: a cognitive elite of 2 to 5% that works in partnership with frontier models, and a majority that migrates to simplified, one-way tools. The "AI bubble" is therefore read as a mismatch between the speed of technology and the speed of human adaptation, the reverse of the dot-com case.

Five testable predictions for 2026-2028 are listed, among them bimodal session lengths, a preference for simpler answers and higher demand for cheaper, simpler tiers, together with a proposed experiment. A closing section, marked as speculative, introduces **Homo Symbioticus**, a possible social class defined by the ability to think together with AI.