Researchers have published a quantitative framework suggesting that human cognitive capabilities could collapse catastrophically once artificial intelligence assumes responsibility for enough mental tasks — a phenomenon they term the 'enrichment paradox.' The preprint, posted to arXiv (https://arxiv.org/abs/2603.24391), presents a two-variable dynamical systems model tracking human capability (H) and delegation to AI (D) over time.

The model rests on three axioms: that learning requires existing capability, that skills require active practice to maintain, and that disuse causes forgetting. Calibrated against four domains — education, medicine, navigation, and aviation — the model identifies a critical threshold designated K*, approximately 0.85, beyond which human capability does not gradually erode but collapses abruptly. Crucially, the authors note that a broader scope of AI involvement lowers this threshold, meaning the danger point arrives sooner when AI touches more aspects of cognitive life.

The researchers validated their model against 15 countries' Programme for International Student Assessment (PISA) data, encompassing 102 data points. The model achieved an R-squared value of 0.946 using only three parameters and registered the lowest Bayesian Information Criterion score among competing models — a standard measure of explanatory efficiency — lending it statistical credibility.

Among the model's more striking predictions: periodic AI system failures actually improve long-term human capability retention by a factor of 2.7, compared to uninterrupted AI use. The simulation baseline assumes a 5% background rate of AI failure. Additionally, mandating 20% of tasks be performed without AI assistance is projected to preserve 92% more capability than the baseline trajectory.

The findings arrive amid intensifying policy debate over AI's role in schools, hospitals, and professional environments. While AI advocates emphasize efficiency gains, this research offers a formal quantitative argument for governance mechanisms — including mandatory practice requirements and deliberate 'friction' in AI systems — to prevent dependency from crossing irreversible thresholds.

The paper, titled 'The enrichment paradox: critical capability thresholds and irreversible dependency in human-AI symbiosis,' has not yet undergone peer review. Its authors argue the model provides 'quantitative foundations for AI capability-threshold governance,' a policy domain that to date has lacked rigorous mathematical grounding. Whether regulators in education or healthcare will incorporate such thresholds into AI deployment standards remains an open question.