Will Artificial Intelligence Surpass Humanity? Dating Human Parity, Dimension by Dimension, with RAYOBETH Theory

Whether artificial intelligence (AI) will surpass humanity is among the most widely asked questions of the decade, yet it is rarely posed in a form that evidence can answer. This article reformulates it within RAYOBETH Theory as a set of dated, dimension-specific parity questions and answers them with public data. It proves a horizon parity law: if AI failures occur at a hazard per minute of human work that declines exponentially, the date at which AI matches human success on a task family rises with the logarithm of task length, and parity on reliability lags parity on the majority of tasks by a gap that depends only on the human success rate and the domain’s doubling time. It further shows that general parity across all dimensions is governed by the slowest dimension, cannot be dated while any dimension is unmeasured, and may never be certified because benchmarks are retired before parity. Applying the theory to 14 benchmarks with documented human baselines and to task horizon data, the article finds that AI crossed human performance on knowledge tests in 2021 and 2022, on expert science questions in 2024 and on abstract reasoning puzzles in 2025 and 2026; that calendar parity dates rise with task length (Spearman ρ = 0.83, p = 0.011); that the horizon of reliable work doubles every 3.5 months in software but every 9.5 months in real freelance projects; and that majority parity on 11.5 hour freelance projects is projected for 2027, with reliability parity between 2028 and 2031. Physical, social and governance dimensions remain unmeasured, so whether AI surpasses humanity as a whole cannot yet be dated.