From Immensity to Infinite Intelligence? Extending RAYOBETH Theory to the Past, Present and Future of Artificial Intelligence

Artificial intelligence (AI) is the most demanding test case for any theory of generative possibility: its systems increasingly contribute to the research that produces their successors. This article extends RAYOBETH Theory (Ndambiri, 2026) to the history and prospects of AI. In the extension, AI research capacity enters its own generator, so the recursive generative elasticity β becomes endogenous. Four results follow. An automation threshold theorem shows that β rises with the share of research performed by AI and crosses the explosion boundary at a computable share. A compute bound theorem shows that any explosive phase is transient when compute grows exogenously, after which capability growth inherits the growth of compute. A law of the weakest generative loop shows that sustained explosion requires every complementary input, including compute, energy, data and capital, to be generated by AI itself. A conditional non-finality theorem shows that no observed level of AI capability is a final bound, yet no finite record can prove that progress is unbounded. Using Epoch AI data on 3,654 models, 183 accelerators and 59 benchmarks, the article finds three compute eras with doubling times of 21, 10 and 5 months; estimates of β that rise from about 0.7 before 2010 to about 1.5 after 2019; expected benchmark lifetimes that halve every 2.8 years; demand for compute that outruns hardware efficiency by a factor of more than three each year; and a critical mass threshold in national AI capability that no African country has yet crossed. Twelve postulates and eight falsifiable hypotheses are proposed.