Intelligence Flywheel: A Conjecture on Recursive Self-Improvement and the Dynamics of AI Progress

We introduce the Intelligence Flywheel hypothesis: a self-reinforcing cycle where AI systems increasingly contribute to their own development, fundamentally altering the dynamics of technological progress. We formalize four key observations: (1) recursive contribution-AI now generates ∼30% of novel intelligence artifacts; (2) pre-singularity acceleration-while total AI capability remains below human capacity, growth rates are shifting toward exponential; (3) temporal asymmetry-the knowledge production rate in AI exceeds human absorption capacity by orders of magnitude; and (4) civilizational phase transition-the societal impact may parallel the industrial-agricultural transformation. If this hypothesis holds, current linear projections of AI development drastically underestimate near-term disruption, necessitating urgent recalibration of research priorities, governance frameworks, and educational paradigms.