From Artificial General Intelligence to Superhuman General Intelligence: Formal Models, Metaphysical Constraints, and a Prognosis for Society and Science

We develop a mathematically explicit and metaphysically disciplined framework for analyzing the possible evolution of AI toward AGI and SGI, treating progress not as a smooth inevitability but as a sequence of threshold crossings shaped by compute, autonomy, verification, institutions, and risk. It extends standard timeline forecasting with a punctuated-equilibrium model, using transformers, agentic AI, multimodal/world-simulation systems, advanced neuroanatomical architectures, reasoning-RL, test-time scaling, scientific foundation models, and AI-R&D automation as examples of discontinuities that may compress or redirect AGI/SGI timelines. The central conclusion is cautiously forward-looking: future breakthroughs could substantially accelerate scientific discovery and institutional transformation, but only systems that combine robust autonomy, verifiable reasoning, recursive research capability, and controllable deployment would justify strong claims about AGI or SGI.