The AI Journal

In the spring of 1962 a fire began in a rubbish dump sitting above an abandoned anthracite seam beneath Centralia, a small borough in the coal country of central Pennsylvania. It has not gone out. The seam beneath the town has burned for more than sixty years, venting steam and carbon monoxide through cracks in the ground, hollowing out the rock until the surface subsides, and driving almost everyone away. A federally funded relocation emptied the borough; the postal service retired its zip code; the streets, condemned and mostly cleared, remain as a grid the town no longer fills. The standard sentence about Centralia is that the fire is unextinguishable. That sentence is almost never examined, and it is the wrong one.

Coal-seam fires are genuinely formidable, and it is worth saying why before disputing their reputation. In a coal seam, heat, fuel and oxygen form a feedback loop: oxidation raises temperature, higher temperature accelerates oxidation, and the fire creeps along the seam through fractures rather than burning as a visible flame. Stracher and Taylor (2004) call this a thermodynamic recipe for environmental catastrophe, and the phrase is not rhetorical. These fires burn underground for decades, migrate unpredictably, and emit gases and particulates that are a serious hazard to anyone above them. They are a global phenomenon, not a Pennsylvania curiosity.

The leap from “formidable” to “impossible”, however, is where the argument usually stops. Coal fires are suppressed in practice, and the technology for doing so is neither secret nor speculative. Gel foams, inert-gas injection, grouting and trenching are mature methods for cutting off the oxygen and heat that sustain a seam fire; Zhang and colleagues (2023) review decades of gel-foam development for exactly this purpose. Coal fires are dangerous and expensive to fight, but they are fought, and sometimes they lose.

Nor is Centralia’s fire unusual in being fought. Where a burning seam underlies ground that is inhabited or economically active, suppression becomes an ongoing public commitment rather than a single decision: boreholes are drilled, seams are flushed with water and slurries, and surface cracks are sealed, year after year, because the value above the fire keeps the arithmetic pointed the other way. Stracher and Taylor (2004) document fires burning on every inhabited continent, and the methods reviewed by Zhang and colleagues (2023) are the ones deployed against them. The distinction that matters is not between fires that can and cannot be beaten. It is between fires whose neighbours can and cannot make the bill look worth paying.

What distinguishes Centralia, then, is not that its fire defied a technology that worked elsewhere. It is that the question of fighting it was converted, very early, from an engineering question into a question of value: how much money is a small, declining coal town worth? In 1980, a socio-economic impact analysis was produced to compare abatement alternatives precisely in those terms, assessing the costs and the consequences of different courses of action. That document is evidence of something easy to forget. Before the town was written off, its fire was analysed as a problem with options. The options existed. They were costed. They were not chosen.

The choice not to choose has a structure. Abatement of a coal fire is cheapest at the beginning, when the affected volume is small and the seam can still be sealed before the fire has opened its own channels through the rock. Delay multiplies the cost, which makes the same cost-benefit calculation return a different answer at every later point: the fire is always more expensive to fight today than it was yesterday, and the town is always worth less. A decision-maker who keeps asking whether the place is worth the price will keep getting a tidier answer to leave it, because the arithmetic itself moves. Jennifer Elick’s infrared mapping of the Centralia fire (2011) and the status survey by Nolter and Vice (2004) both describe a fire that was still advancing and still hot years after the town had begun to empty. The fire was not waiting on a scientific breakthrough. It was waiting on a cheque that the arithmetic argued against writing.

This is the point at which the account becomes uncomfortable for more than the Commonwealth of Pennsylvania. If the persistence of Centralia’s fire is a policy outcome, then the sentence “it cannot be put out” is doing quiet work. It converts a series of contested, underfunded and finally abandoned public decisions into an act of nature. It lets everyone who decided the town was not worth the money describe their decision as a recognition of the truth rather than a choice among alternatives. The fire is real. The inevitability is a story told about it.

There is a second cost to the word. Calling the fire unextinguishable does not merely excuse the decisions that let it grow; it also makes any future intervention look futile by definition, so the description becomes self-fulfilling. A town whose fire is officially permanent will not be fought for, because no official wants to spend money against a force everyone has already agreed is beyond reach. The label does not only report the outcome. It helps secure it, which is why the choice of word is itself a policy decision.

The strongest objection to this reading is not foolish, and it deserves to be stated without softening. It runs like this: by the time anyone seriously proposed putting the fire out, the seam was already burning across too large a volume, under too fractured a roof, for any intervention to have been safe or affordable. On this account the economics are secondary, because the fire was already past the physical threshold, and the threshold, not the budget, is what made Centralia’s fire permanent. There is real force here. Coal fires do have a point of no return, and scattershot boreholes and short-lived grout curtains cannot arrest a fire that has spread through a broken seam.

The objection, though, mistakes the size of the fire for a natural constant. The threshold was crossed during the years of contested, partial, sequentially abandoned abatement efforts, not before them, and the crossing is exactly what under-funding buys. A fire that is fought seriously and continuously is fought while it is small; a fire that is studied, debated and left is left to grow through precisely the interval in which it could still have been stopped. The physical window and the political window closed together, and the second closing is what let the first happen. To say the fire was always already too large is to read the outcome back into the beginning, which is the very move this essay began by objecting to.

Centralia is usually taught as a lesson about geology, a place where the earth did something humans could only watch. It is a better lesson about arithmetic and nerve. The fire was never beyond human control in the way the sentence implies. It was beyond the price a series of institutions were willing to pay, and the gap between those two things is where the town went. That gap is not a fact about coal. It is a fact about us, and it is the part of the story still being written, in places that have not yet burned.