How the accumulation of threats turns our future into a gamble
where the stakes are all of humanity
In the film No Country for Old Men, there is a scene that lasts only a few minutes but contains the entire horror of our time. Anton Chigurh, a man who embodies relentless, random death, stops at a gas station. He flips a coin and offers the unsuspecting store clerk a chance to call heads or tails. The clerk, unaware that his life is on the line, asks in confusion: "What do we stand to win?" And hears the answer: "You stand to win everything." The clerk does not know that he has no choice but to play by the rules of the man holding the coin. He does not know that he already staked everything he had simply by being born into this world.
We are that clerk. We live without realising that every day we are flipping a coin, and that the stake is our lives and the future of the entire species. Heads — we live another day of ordinary life. Tails — and the world as we know it ceases to exist. This coin is not made of metal; it is made of probabilities — six different threats. Each one alone looks harmless, like a small percentage. But together, they amount to nearly a one-in-two chance that, within the next ten years, the bet will be lost.
When scientists moved the Doomsday Clock to 85 seconds to midnight in January 2026, it became the darkest record in the clock's history [1]. The clock, created in 1947 to visualize the nuclear threat, now also accounts for climate, biotechnology, and artificial intelligence. Yet most of us, upon hearing the news, shrugged and within five minutes were thinking about something else. We have grown accustomed to the world constantly teetering on the edge, and we no longer take these warnings seriously. We have been taught to think about threats in isolation and to comfort ourselves with small numbers. "Nuclear war? Only two percent per year. Pandemic? About the same. Artificial intelligence? Well, maybe in thirty years…" — this is how we reason, not noticing that this very way of thinking is a trap that could cost us everything.
Today, at least six mechanisms operate on this planet, each capable not just of disrupting the normal course of things, but of destroying civilization as we know it.
There are nine nuclear states, and no major arms-control treaty is currently in force: the last one, New START between the US and Russia, expired in February 2026 [2]. Nobel laureate in physics David Gross, one of the greatest minds of our time, has stated bluntly that he spends part of his time trying to explain to people that the probability of nuclear catastrophe is about two percent per year [3]. In actuarial terms, that implies an average waiting time of roughly fifty years. Each year we roll a fifty-sided die, and if the wrong number comes up, it is all over.
Pandemics arrive at a similar frequency — epidemiologists estimate two to three percent per year [4]. Bill Gates, who has invested billions in fighting infectious diseases, puts the probability of a new pandemic in the next four years at ten to fifteen percent [5]. But these are only the natural threats. In the age of artificial intelligence, the picture changes dramatically. In 2025, researchers at Los Alamos National Laboratory showed that under a scenario where the number of people capable of synthesising pathogens with the help of large language models increases tenfold, the expected number of deaths from catastrophic outbreaks could rise by fifty thousand over three years [6]. And in 2026, the journal Science published a study demonstrating that DNA screening systems fail to detect more than seventy-five percent of toxins designed by AI [7]. The barriers meant to protect us can already be bypassed.
Climate tipping points are also approaching relentlessly. The World Meteorological Organization gives a ninety-one percent probability that within the next five years we will cross the threshold of 1.5°C of global warming [8]. A major assessment published in Science (Armstrong McKay et al., 2022) found that at 1.5°C of warming, four of the sixteen scientifically established climate tipping elements — among them the Greenland and West Antarctic ice sheets — become likely to be crossed, and five more become possible [9]. This means that before we reach official climate goals, we may lock in irreversible processes — collapsing ice sheets, dying coral reefs — with five more tipping elements entering the danger zone. These changes are not linear — they are cascading, and one tipping point can trigger another. It is important to be precise: crossing 1.5°C is not itself a civilizational collapse. It is a threshold crossing that raises the probability and the violence of everything else.
Equally alarming are the assessments regarding general artificial intelligence. A study by the Massachusetts Institute of Technology, involving 272 international experts, showed that under the current trajectory, eighteen out of twenty-four AI-related risks have a greater than ten percent probability of catastrophic consequences — more than one million deaths or more than one hundred billion dollars in damages — within the next five years [10]. The creator of modern deep learning, Geoffrey Hinton, estimates the probability that highly advanced AI will lead to human extinction within the next three decades at ten to twenty percent [11]. Dario Amodei, CEO of Anthropic, puts the chance of things going "truly badly" in the range of ten to twenty-five percent [12], while some experts give figures as high as ninety-five percent. Ten to twenty percent extinction from a single technology is not a "small risk"; it is an unacceptably high stake in a game where everything is on the line.
Finally, we are facing the degradation of global institutions that were supposed to coordinate our efforts against all these threats. In 2025, the UN published its World Social Report, documenting a sharp decline in institutional trust worldwide [13]. The researchers emphasize: trust is the glue that binds cooperation and solidarity between groups and between citizens and states. When trust falls, the ability to cooperate falls as well. The World Economic Forum's 2026 report notes that the world has entered an "era of competition," and ninety percent of the 1,300 global leaders surveyed consider the risk of global catastrophes over the next two years to be at least moderate, with forty-two percent rating it elevated, and eight percent as imminent [14]. The institutions that should protect us are paralyzed: they cannot agree, they cannot act, they cannot prevent disaster. One clarification is needed: unlike the other mechanisms, institutional degradation is not an event with an annual probability. It is a background condition — one that raises the likelihood and the deadliness of everything else.
Each of these threats alone does not look too frightening when viewed in a vacuum. Two percent, five percent, ten percent — they seem small because our brains are not wired to evaluate accumulated probabilities. But we make a fundamental mistake when we consider them in isolation. We behave as if these threats exist separately, as if they do not intersect and amplify each other. In reality, they are all linked. Climate crisis drives mass migration, migration intensifies political conflicts, conflicts increase the risk of nuclear strikes, and nuclear war can destroy healthcare systems already struggling with pandemics, while also disrupting supply chains essential for AI governance. This is not a sum; it is a cascade, where each new factor multiplies the danger of the previous one.
To estimate the combined probability, we must turn to simple mathematics. If we have several independent risks with annual probabilities
then the probability that none of them occurs over years is the product:
Therefore, the probability that at least one event occurs is:
For example, take three risks, each with an annual probability of two percent — say, nuclear war, a natural pandemic, and one more factor of the same magnitude. Over five years, the probability that none of them materializes is (0.98)15≈0.74(0.98)15≈0.74. Hence, the probability of at least one event is 1−0.74=0.261−0.74=0.26, or twenty-six percent. More than one chance in four.
Now scale the same calculation to a decade — and be deliberately conservative. Take only three event-type risks, each at a modest two percent per year. Ignore the climate cascade. Ignore institutional degradation. Ignore every estimate above the lowest published one. Even then: (0.98)30≈0.55(0.98)30≈0.55, meaning a forty-five percent chance of at least one catastrophe within ten years. Add the remaining risks under the conservative assumption of independence, and the probability of the coin landing on tails this decade comfortably exceeds forty percent.
Is the independence assumption realistic? Partly. In one direction, dependencies slightly reduce the probability of "at least one" event for fixed estimates; in the other, they sharply increase the damage when several risks land together. What is not debatable is the conclusion: we are not dealing with remote possibilities. We are dealing with a probability that, in any other domain — aviation, construction, medicine — would ground the entire fleet.
But why do we continue to cling to small numbers and ignore this reality? There are several deep reasons, all rooted in the workings of our brain and our social systems.
There is a deeper, mathematical explanation for why we misjudge these risks. It is called the theory of fat tails, and it overturns our usual view of the world. Most of us were taught in school that random events follow a bell-shaped curve — the normal distribution — where extreme values are nearly impossible. Human height follows such a curve: you will never meet a person ten meters tall, because the probability is so small it can be ignored.
But there are phenomena that behave differently. The distribution of wealth, city sizes, earthquake frequencies, and, crucially, the scale of pandemics — all these have so-called "fat tails." Extreme events occur far more frequently than the standard curve predicts. Researchers studying the spread of infectious diseases have found that the death toll of epidemics follows exactly this pattern: large catastrophes occur with alarming regularity, and we are surprised every time.
Fat tails mean that average risk estimates are a trap. We cannot say: "The average probability of a pandemic is such-and-such," because the tail harbours events that are hundreds or thousands of times worse than the average. That is why we should not be comforted by hearing "only two percent." That two percent belongs to a distribution with an unknown tail, and that tail could contain anything — from a local outbreak to species extinction.
And these estimates are not static. Nuclear risk grows as treaty frameworks erode. Pandemic risk grows because AI puts pathogen design within reach of non-state actors. Climate risk grows because emissions are falling too slowly. Each trend accelerates the others. Therefore, if anything, the figures above err on the side of optimism — not because the inputs are understated, but because the system itself is accelerating, and we cannot predict its behaviour when several of its subsystems approach critical states simultaneously. In the scientific literature, there are works, such as those by Kelly and colleagues, showing that global risks are cumulative and can lead to synchronous collapses [15]. Ignoring the combination of threats is suicidal.
Now imagine flipping a coin every day, knowing that in one out of two scenarios over a ten-year horizon, the world you know may collapse. How can you continue living as you did? How can you build a business without accounting for this risk? How can you invest in yachts, villas, and stocks, hoping that everything will be fine? If it is not fine, all those assets lose their value the moment the system that underpins them disappears.
Of course, the rational response also includes reducing these probabilities directly — arms control, pandemic preparedness, AI safety. But mitigation and insurance answer different questions. Mitigation lowers the chance that the coin lands badly. Insurance limits the damage when it does. A civilization with no backup has no second line of defense at all.
The only sensible insurance today is a reserve circuit of civilization — one that can survive a catastrophe rather than merely record it: nuclear war, superpandemic, climate collapse. This sounds like science fiction, but technically it is already possible. We know how to build closed-loop life-support systems. We know how to land probes on other planets. We know how to preserve knowledge in digital form. The question is not technology. The question is whether we can stop soothing ourselves with small percentages and admit that the coin has already been tossed.
"The only sensible insurance today is a reserve circuit of civilization."
We face a choice. Continue to close our eyes and hope for a miracle — or start building an ark before it is too late. History does not forgive those who refuse to face the truth. Civilizations perished not from a lack of resources or ignorance — they perished because they refused to recognize the threat until it became too large.
We have a chance. There is still time. But there is less of it than we think. And if we want our children and grandchildren to see tomorrow, we must start acting today. Because flipping a coin with a fifty percent chance of catastrophe is not a game. It is a challenge we must meet. And the only right response is to do everything possible to make the coin fall in our favour. That is exactly why the Martian Ark project was created — as a practical engineering plan to build an independent reserve circuit of civilization, one that does not depend on Earth's degrading institutions and can preserve our species in any future scenario. All the necessary information, calculations, and details are gathered in an open white paper available at the link below.
And here I must make an important clarification, because honesty demands transparency. The figure of forty to fifty percent presented above is not an objective scientific fact, not a UN statistic, and not a peer-reviewed finding. It is an author's calculation built on explicit assumptions. We took the most studied threats, used conservative published estimates for each (for example, ~2% per year for nuclear war — David Gross's estimate; 2–3% per year for pandemics — epidemiological estimates), assumed independence, and extrapolated over a ten-year horizon using the formula (1−p)N(1−p)N. Two non-event factors — the climate tipping cascade and institutional degradation — were deliberately excluded from the arithmetic and treated instead as amplifiers. Independence is a simplification, and it cuts both ways: interdependencies slightly reduce the headline probability of "at least one" event, while sharply increasing the severity of outcomes when several risks strike together. With different assumptions — different expert estimates, a different time horizon, a different treatment of dependencies — the final number changes. It might be thirty percent. It might be sixty. But even under the most optimistic adjustments, it remains at a level that, in any other domain — aviation, construction, medicine — would be treated as catastrophic and acted on immediately. The point is not the exact number. The point is that the accumulation of threats changes the nature of risk: rare events cease to be rare when there are many of them, and they begin to overlap. That is what we are trying to show. Not to scare, but to warn. Not to claim that everything will collapse tomorrow, but to demonstrate that we cannot afford to pretend that nothing is happening.
The time to decide is running out with each passing day. And if not us, who? If not now, when?
Disclosure: I originally write in Russian; English versions are produced with AI translation and editing. All ideas, calculations, and claims are mine. I will defend any of them in comments in either language.