Why do humans consistently overestimate dramatic risks and underestimate slow ones?

After a shark attack receives extensive media coverage, beach attendance drops significantly in areas where sharks are extremely rare and statistically negligible threats. After a dramatic plane crash, people switch to driving, a form of transport that kills orders of magnitude more people per kilometre. After a terrorist attack, governments spend billions on security measures for scenarios that will almost certainly never recur, while chronic diseases that kill millions every year receive proportionally less attention and urgency.

This pattern is so consistent and so well-documented that it is hard to treat as a series of mistakes. It is a feature of human cognition. Understanding why it exists tells you something important about both how we think and why the modern world is so strange for us to navigate.

The most influential explanation comes from the work of Daniel Kahneman and Amos Tversky, who identified what they called the availability heuristic. When estimating the probability or frequency of something, humans rely heavily on how easily examples come to mind. If you can quickly recall several instances of something happening, you judge it to be common. If you cannot, you judge it to be rare. This is a reasonable shortcut in many situations. In ancestral environments where your data set was limited to personal experience and what the people around you had witnessed, availability was a decent proxy for frequency.

The problem is that availability is now heavily mediated by information systems that are not designed to give you an accurate picture of what is actually dangerous. News organisations select for the dramatic, the surprising, and the rare. A plane crash appears on every front page; the road accidents that happened simultaneously appear nowhere. A terrorist attack generates weeks of coverage; the hundred people who died of preventable causes in the same period are not a story. Over time, this creates a vivid catalogue of rare dramatic events in your memory and almost no trace of the slow, diffuse, statistical harms that actually kill most people.

There is a related mechanism called the affect heuristic. We use emotional response as a guide to risk. Things that produce strong fear feel dangerous. Things that produce mild anxiety or no emotion feel less dangerous, even if the numbers tell a different story. Dying in a fire feels more terrifying than dying from a sedentary lifestyle, so we install smoke detectors obsessively and rarely change our habits around exercise. The vividness of the imagined experience drives the risk assessment, not the probability.

There is also a deep asymmetry in how we process threats depending on whether they are acute or chronic. Our threat-detection systems evolved for acute dangers, a predator, a falling rock, a hostile stranger. These produce immediate, identifiable signals and call for rapid action. Chronic risks, the accumulation of small unhealthy choices, the slow degradation of an ecosystem, a financial system gradually becoming fragile, produce no single alarming signal. They are easy to ignore because there is never a moment that clearly demands attention.

This asymmetry has significant consequences for policy and collective decision-making. Democratic governments respond to salience. An acute crisis, a flood, a terrorist attack, a pandemic in its first weeks, generates political pressure for action. A slow crisis, rising rates of chronic disease, gradual infrastructure decay, incremental institutional weakening, generates much less political urgency, even when the expected harm is far greater.

Climate change is the most consequential current example. The harms are statistical, diffuse, and distributed across time and space in ways that make them very difficult to make vivid to people whose attention is calibrated for sharp immediate threats. The people who will suffer most severely are not yet born. The damage accumulates through processes that are individually invisible. Against this, the costs of action are immediate, concrete, and politically concentrated. The mismatch between our evolved risk-perception systems and the structure of the problem is part, not all, but part, of why collective action has been so difficult.

What can be done about this? The honest answer is: not as much as we would like, and what can be done requires effort.

At the individual level, the corrective is deliberate, effortful statistical thinking. When you feel strongly that something is dangerous, check the numbers. When something does not produce anxiety but the data suggests it should, notice the gap. This is cognitively expensive and most people will not do it consistently, but those who do tend to make better risk decisions over time.

At the systemic level, the corrective is institutions and processes designed to counteract the biases of unmediated human intuition. Regulatory bodies that set standards on the basis of evidence rather than salience. Long-term cost-benefit analyses that give future harms appropriate weight. Journalism norms that try to represent statistical reality alongside individual stories. None of these work perfectly. All of them work better than relying on unaided intuition.

The uncomfortable conclusion is that the world we live in, saturated with vivid, dramatic, algorithmically amplified information, is precisely calibrated to make our risk-perception biases worse. The solution requires swimming against a very strong current. That is not impossible. But it needs to start with understanding why the current is running the way it is.