A Bayesian Doomsday Clock
I don’t need to introduce you to the Doomsday Clock. It’s the metaphorical device that informs us what a collection of esteemed thinkers consider to be the risk of complete civilization collapse. The adjustments are based on events since the last adjustment and seem to represent an informal Bayesian inference—events that suggest more danger adjust the time closer to midnight and the reverse for encouraging signs. Until 2007, nuclear warfare was the only catastrophe considered, but now irreversible climate change is also included.
Looking at the changes and the reasons given, they look reasonable to me. However, there’s a very ad-hoc feel to the actual adjustments. Which is not surprising given the original setting was profoundly ad-hoc:
As a prior estimate, that’s probably as good as anyone could have done just 2 years after the first atomic weapon was tested. Only, it’s not clear what those minutes mean. Certainly not (as my much younger self assumed) that it was how much warning we could expect before the end of the world. Instead, it seems to be a symbol designed to draw out emotions of appropriate dread and concern. The exact value of X in “X minutes to midnight” is less important than whether we are closer or further from total destruction.
Now that most people living have faced the threat of annihilation all their lives, I think we can come up with a better estimate. To start, I’m going to assume the 24 hours of the clock represent the span of civilization from the earliest cities to whatever end we succumb to. If we count the proto-city of Jericho, we’ve already experienced 11,000 years of civilization. Assuming we are very close to midnight (the furthest the clock got was 23:43 in 1991) each minute represents a little less than 8 years. That means the initial setting (23:53) implied about 54 years left on civilization’s clock. At 2 minutes, we’re looking at more like 15 years.
Again, the actual Doomsday Clock isn’t making any such claim, but I think many people feel that level of anxiety about the future right now. But should we feel that worried? How confident can we be about living out our lives without seeing mushroom clouds or experiencing out-of-control warming?
As it happens, there is a statistical way to come up with a prior estimate for at least the nuclear threat: the rule of three. Since we’ve experienced ~11,000 years of civilization without destroying ourselves, we can estimate there’s a 0.027% (3/11,000) change of that event happening next year. That doesn’t seem so bad, does it?
One problem: the risk profile changed recently. Historically, it’s not uncommon for a single city to get wiped out by flood, fire, earthquake, famine, disease or invading armies. But those events are very local and won’t harm cities hundreds of miles away. So the odds that all cities would be destroyed were indeed very low. Once it became possible to destroy entire cities in an instant and simultaneously inaugurate a new ice age, the calculus changed.
If we calculate our prior based on the last 73 years of not initiating an apocalypse, the odds of one occurring next year is about 4%. That works out to a median time to doomsday of 16 ½ years. Converting that to time on the clock and I estimate we’re at 2 minutes to midnight, which is what the official clock now reads. And the only variable I looked at was how many years it’s been since the Trinity test.
Now, my estimate continues to get more optimistic the more years we go without the end of the world. Clearly, there have been events over the years that increase and decrease the odds. For instance, there was no real risk of nuclear winter in 1945 because there were only 4 bombs in existence and 3 of them were actually detonated above ground. As more and better devices were built and more countries built them, the risk increased. Similarly, treaties reducing the size of the arsenal and preventing proliferation reduce the risk.
But how much ought the minute hand change with these sorts of events? Take for instance the increased tensions between North Korea and, well, the rest of the world. A nuclear power with an unstable dictator intent on picking fights seems pretty worrisome. However, they have at most 60 nuclear weapons (and likely far fewer) which are closer in destructive power to the bombs the US dropped on Japan than what other countries possess. And we don’t really know if those devices have been made small enough to be installed in North Korea’s ballistic missile systems. Successful attacks on major cities within their reach would, of course, be tragic but probably not world-ending. Even the environmental impact might be less than you might expect.
On the other hand, there’s got to be some risk that an unprovoked attack by one of the small nuclear powers would draw Russia or the US into conflict. How big a risk is hard to estimate without creating a model informed with a detailed understanding of global conflict theory. This isn’t exactly my speciality, so I trust the Bulletin of the Atomic Scientists experts’ assessment—it’s not looking good right now.
And yet . . . I’m skeptical that North Korea is really a threat to instigate a global nuclear conflict. It seems to me that we might be victims of generations of survivorship bias. Looking at a list of close calls on Wikipedia, about half were regional conflicts that might have escalated into armageddon. The rest were mistakes in which one super power suspected the other had already put a preemptive strike plan into operation. It’s terrifying to think how close we have come to World War III, but those near misses demonstrate that nuclear powers actually have robust safeguards to prevent it.
Given 11 such incidents (there might be more we don’t know about), the rule of three suggests the next one has a 27% of initiating a nuclear strike. It’s difficult to guess how likely a limited strike would be to starting Global Thermonuclear War. (I’m not updating my odds since I believe this evidence is baked into my prior. The worst hasn’t happened yet no matter how you break it down.) As nerve wracking as these situations are to experience, surviving them gives us more confidence that they are survivable. Which isn’t to say we shouldn’t work to reduce those risks by improving our controls over these weapons and reducing the stockpile.
The one scenario in which nuclear weapons have been used in aggression remains the most likely, in my estimate. If a nuclear power is unable to finish a conventional war conventionally, it might be tempted to use overwhelming force. Either a nation that is struggling to get the ball across the goal line (as with the Pacific theater during WWII) or a nation facing defeat might see their nuclear deterrence as a way out of the situation. So a huge risk indicator is the size of global nuclear stockpile and the number of nations who control them.
by Wm. Robert Johnston
Things have gotten better since the mid-80s. Most of the gains came from just the Russian and US weapons. Thankfully, it’s not just a case of fewer, but also less powerful warheads. Both superpowers have dramatically reduced the number of high yield (> 4.5 megatons) weapons. Unfortunately, non-signatory states (Israel, Pakistan and India) have steadily increased their capabilities. Even worse, two of those nations are in constant diplomatic tension with each other and the third located in perhaps the most unstable region on Earth.
Since these countries seem to have sufficient power to cause nuclear winter, we should probably adjust the minute hand of my Doomsday clock. India has had a nuclear capability since 1980 and Pakistan since 1990. One of their four wars occured in 1999 and there have been three standoffs or skirmishes since. In addition, Wikipedia lists four ongoing armed conflicts. So there have been 8 opportunities for each side to use extreme measures and neither side has taken that option. To go back to the rule of three, that means there’s a 37.5% chance the next conflict will result in a nuclear strike. Assuming the number of conflicts since 1990 is representative, that works out to about a 29% chance of a new conflict occurring each year. So that’s about 11% odds of a nuclear showdown in the Indian subcontinent next year.
Would that result in the total destruction of civilization? Maybe. It would be tragic, for sure. Crops would fail and we’d probably have huge rates of radiation sickness wherever the wind blew. The next season of The Walking Dead would probably be canceled. But the human race has proven to be very resilient. The Black Death killed 30–60% of Europe and people found a way to keep going. If I give a Indo-Pakistani nuclear exchange a 1 in 3 chance to end civilization, that works out to less than 4% chance of Doomsday in a year. Is that an additional risk to the prior I already calculated? It depends on whether you think, as I do, that nuclear powers really are effective at preventing this scenario.
The crux of the problem is that estimating events that haven’t occurred yet requires a robust model. I kinda assumed that a nuclear conflict between the US and Russia would end the world. Certainly many cities would be in ruins. Unlike Hiroshima (which also suffered major damage from a typhoon in 1945) and Nagasaki, there probably won’t be much help from outside to rebuild. Between radiation and particulate matter tossed into the atmosphere, the environment would be immediately and dramatically changed. This would cause yet more death and suffering far from the targeted cities. If you want to call it Doomsday, I would not argue with you. But it might not be the end of civilization.
And this brings us to climate change. There’s no doubt that global temperatures have been rising and that human activity has been the primary contributor. We also know that adding more carbon dioxide will accelerate the greenhouse process with dire consequences. As we learn more and more about our planet, climate models are getting more and more accurate. The only trouble is that they are difficult to validate. There’s just not enough historical data and climate cycles take so long. Both nuclear winter and runaway greenhouse effect have solid theoretical grounding, but the exact parameters needed to cause each can’t be determined with certainty.
So overall, I think a Bayesian Doomsday clock would show only a couple of minutes to some awful outcomes: nuclear strikes and catastrophic global warming to name two. But an actual end to civilization? Well, I’m not so sure.
