The Mathematics of Civilizational Survival

We develop a formal framework for classifying civilization-scale threats according to whether they admit a mathematical repair mechanism, and we locate the risk posed by advanced artificial intelligence (AI) precisely within this classification. Two structurally distinct categories of absorbing barrier are defined: weak barriers, whose dynamics contain a strictly positive recovery term, and strong barriers, for which the recovery term is identically zero and the state is permanently absorbing. We show that all conventionally cited existential threats (nuclear conflict, climate change, pandemics, biodiversity loss, asteroid impact) fall into the weak category, whereas advanced AI is the unique known candidate for a strong barrier. We then establish three results. First, under super-exponential self-improvement, the intelligence variable exhibits a finite-time singularity, so that any governance response operating on a linear or exponential timescale is asymptotically outpaced. Second, combining the Law of Requisite Variety with computational irreducibility yields an uncontrollability theorem. Third, invoking a non-stationarity/heterogeneity principle we prove a floor lemma—the per-period barrier-contact probability admits a strictly positive lower bound—so that its sum diverges and, by a Borel–Cantelli argument, the cumulative survival probability decays to zero. We emphasize that this last conclusion is conditional: it holds only while four stated premises are jointly satisfied, and we characterize precisely the escape conditions under which the cumulative survival probability retains a positive lower bound. The analysis is presented as a set of conditional theorems; policy implications are discussed but not asserted.