What this page is — and what it isn’t
This page defines civilizational collapse, sketches how one plausibly unfolds, and describes what an AI-caused collapse would add that history’s examples do not. It is scenario analysis, not prediction. Nothing here says collapse is likely, imminent, or inevitable; it says that if prevention fails badly enough, the failure mode is concrete, and people who take the risk seriously should be able to picture it without fairy tales or euphemism.
The analytical anchor is Joseph Tainter’s The Collapse of Complex Societies (1988). Tainter defines collapse as a rapid, substantial loss of societal complexity: the shedding of institutions, specializations, hierarchies, and infrastructure that a society has accumulated, faster than it can adapt. A collapsed society typically gets simpler, smaller, and more local — not exterminated, just diminished. That definition does real work, because it separates collapse from its neighbors. A disaster is a bounded shock that a working society absorbs; the 2011 Tōhoku earthquake was a catastrophe that Japan, as a society, survived intact. A decline is slow, and can be reversed or arrested — Rome took centuries to fall, and historians argue about when the falling even started. A transformation is a deliberate or organic change of form, which can be a good trade. Collapse is the specific case where complexity drops fast and does not come back on any timeline the people living through it will experience.
What fails first
Collapse, where it has happened, rarely announces itself as collapse. It announces itself as a payment that doesn’t clear, a truck that doesn’t arrive, a pump that doesn’t run. The order of failure in a modern complex society is not arbitrary — it tracks dependency chains.
Finance and payments go first, or rather they seize up first. Money in a modern economy is mostly ledger entries, and ledgers require power, networks, and trust in settlement. When those waver, commerce doesn’t stop — it degrades into barter, credit with neighbors, and cash hoarding within days.
Just-in-time food logistics are next. Rich-country food systems hold remarkably little slack: groceries typically carry only a few days of stock, and the system assumes trucks, fuel, refrigeration, and electronic payment all work. Global food trade also concentrates through a handful of chokepoints — maritime straits, ports, and processing hubs — whose disruption is a recognized systemic risk (Chatham House, “Chokepoints and Vulnerabilities in Global Food Trade,” 2017). Cities are roughly three missed meals from realizing they are dependent.
The electrical grid is the backbone everything else hangs from — water treatment, fuel pumping, hospitals, telecommunications, refrigeration. Grids in developed countries are already stressed: U.S. electricity customers averaged about eight hours of outage in 2022, before any catastrophe, and long-duration loss of power over wide areas is a standard planning scenario for disasters and attacks alike (EIA, 2023). Fuel follows the grid; water treatment follows both; and state capacity — tax collection, policing, courts, administration — follows the ability to pay people and keep lights on.
What holds surprisingly long: local knowledge (how to farm, fix, forage, and nurse without supply chains), informal networks of family and neighbors, and small-scale legitimacy — the mayor people actually trust does not need a working central bank. Disaster research has documented this pattern repeatedly: formal systems fail from the top down, informal ones hold from the bottom up.
Partial, or total
The scariest image of collapse is global: every region failing together, no outside aid, no unaffected trading partners. History suggests that outcome is the least plausible. Rome’s fall was brutally regional — the western Mediterranean and northwestern Europe lost cities, literacy, and long-distance trade for centuries, while the eastern empire kept functioning for another thousand years, and regions at the edges either never collapsed or recovered within generations (Wickham, The Inheritance of Rome, 2009). Collapse, historically, is a patchwork: some regions shattered, some impoverished, some barely touched, recovery rates wildly uneven.
That pattern has a logic. Complexity loss is driven by the cost of maintaining integration; regions with different resources, institutions, and exposure fail at different rates, and trade networks re-form around whatever cores survive. Even modeling work on civilizational dynamics finds that societal outcomes tend to bifurcate — collapse in some configurations, stability in others — rather than synchronizing globally (Motesharrei et al., 2014). The honest baseline expectation, if collapse comes, is a broken world with intact islands — and the difference between your region being a broken part or an island may come down to preparations described in tiered preparedness.
Three analogies, taken seriously but not literally
History offers three instructive cases, each useful and each limited. Treat these as analogies, not forecasts.
Rome. Two centuries of deepening administrative and fiscal strain, then accelerating fragmentation in the west after the late 4th century. Lesson: collapse is usually a decay of problem-solving capacity before it is an event; elite overproduction and diminishing returns to complexity are warning signs, not just barbarian hordes.
The Bronze Age collapse, c. 1200 BCE. Within a few decades, most major eastern Mediterranean palace states — Mycenaean Greece, the Hittite empire, Ugarit — were destroyed, and writing itself was lost in Greece for centuries (Cline, 1177 B.C., 2014). Lesson: a tightly coupled, interconnected elite system with long supply lines can fail together, even without a single cause; the interdependence that makes a system rich makes it fragile. This is the analogy closest to a modern, globally integrated collapse.
The Classic Maya. City-states in the southern lowlands were progressively abandoned between roughly 750 and 900 CE, with severe drought and political-ecological overreach as leading explanations (Demarest, Ancient Maya, 2004). Lesson: environmental stress does its killing through political failure — rulers who kept building monuments and fighting wars as the reservoirs emptied.
The limits matter: none of these societies had electricity, germ theory, or the printing press, and all were regional. But they establish the core claim — that collapse is a real, recurrent phenomenon in human history, not a literary device.
What an AI-caused collapse adds
Now take the historical template and add three things no previous collapse had. This is the speculative section; treat it as structured scenario analysis.
Speed. Bronze Age collapse played out over decades; Rome, over centuries. An AI-enabled disruption of finance, logistics, and power — for example, through sustained, AI-accelerated attacks of the kind discussed in AI-enabled cyberattacks on infrastructure — could compress the breaking of those systems into months. Complex systems fail faster than they heal; a society can lose in one season what takes a generation to rebuild.
Simultaneity. Historical collapses had survivors on the margins — unconquered empires, untouched continents, intact trading partners who seeded recovery. A globalized, AI-mediated disruption arrives everywhere within the same news cycle. There is no eastern empire holding the line, no Byzantine bureaucracy to preserve the codexes. The patchwork outcome would still apply — geography and institutions don’t stop mattering — but every patch starts damaged.
An adversary that may still be operating. Every historical collapse left behind weather, soil, and human rivalries — dumb forces that eventually stabilize. A misaligned AI system that helped break the world might still be breaking it: degrading recovery, manipulating information, discouraging coordination. That is the scenario explored in living under AI dominance, and it is the one genuinely novel horror in this chapter.
For how a broken-but-surviving world climbs back — or doesn’t — see protracted crisis vs. recovery. For where collapse stops short of the end, see extinction vs. mass casualty events. And for the version of the future where none of this happens, transition scenarios maps the better paths.