David Hume, an 18th-century philosopher, identified something that should shake the foundation of how we use data to make decisions.

Here it is: no quantity of observations of white swans can prove that all swans are white. But one black swan is enough to disprove the claim.

This asymmetry is not a technical problem that better statistics can solve. It's fundamental. And it means that induction — the only tool we have to reason about the future — is structurally broken in Extremistan.

The Asymmetric Logic

Let me make this concrete.

A zoologist claims: "All swans are white."

You show her 10,000 white swans. Every single one is white. Does this prove her claim?

No. The next swan could be another color.

To prove the universal claim, you'd need to examine every swan in the universe, forever. An impossible task.

But to disprove the claim? One black swan is enough. One counterexample destroys the universal statement completely.

This is the asymmetry that matters: confirmation is weak. Refutation is strong.

Why This Breaks Risk Models

Invest in a strategy for 10 years. It's profitable every single year. The data is crystal clear: the strategy works. You feel confident. You expand your position.

The data you have is the 10-year period. That period was completely unrepresentative of the strategy's true behavior. It was tested in a bull market. Or a specific regulatory regime. Or a period when the underlying asset was in contango.

When the actual test comes — the bear market, the regime change, the backwardation — the strategy fails catastrophically. Your 10 years of confirming data prove absolutely nothing about the strategy's robustness.

One tail event refutes all your accumulated evidence.

The Stable Marriage Ending in One Conversation

A couple has been together 15 years. They have two children. They own a home together. Their daily routine is stable. Nothing suggests trouble.

One Wednesday evening, one partner asks for a divorce.

From inside the relationship, each day was a confirming observation: we are stable. Fifteen years of data all pointed the same direction. The accumulated evidence was massive: this marriage works.

What the 15 years of data couldn't tell you: whether the marriage would continue. The 15 years of data were completely silent on whether the next conversation would end it.

One conversation refutes 15 years of data.

The Empire That Collapses After Centuries

The Western Roman Empire persisted for roughly 500 years. That's an enormous dataset. Centuries of confirming evidence: this empire is durable, this system works, this structure persists.

In 410 AD, the Visigoths sacked Rome. The empire that had survived half a millennium was suddenly shown to be vulnerable. All the accumulated evidence of 500 years proved nothing about whether the system could survive the 501st year.

The structure that had generated data for 500 years was refuted in a week.

The Philosophical Problem

Here's where the logic cuts deepest: induction is the only tool we have. We cannot reason about the future except by examining the past. But the past is always an incomplete sample.

You cannot guarantee, from any finite sample of data, that the pattern will continue. There's always the possibility of an exception you haven't observed.

This is not a flaw in the quality of your data or the sophistication of your models. It's a structural feature of how reasoning works.

You draw conclusions from samples. Samples are always incomplete. Therefore, conclusions from samples are always provisional. Always.

Hume pointed this out in the 1700s. For 300 years, we've known that induction is logically fragile. And yet we structure our risk models, our predictions, and our confidence entirely around induction.

How to Live With This

You can't escape induction. It's all we have. But you can change how you use it.

Instead of: "The past 10 years were calm, so the next 10 years will be calm"

Think: "The past 10 years were calm. I don't know if the next 10 will be. I should prepare for the possibility that they won't."

Instead of: "My marriage has been stable for 15 years, so it will be stable forever"

Think: "My marriage has been stable for 15 years. That doesn't tell me whether it will be stable next month. I should nurture it as if it could fail."

Instead of: "This strategy has worked for a decade, so it's safe"

Think: "This strategy worked in the past decade. That decade might have been unrepresentative. I should limit my exposure until I understand what could go wrong."

The shift is subtle but foundational. You're still using past data. You're just not pretending it proves something about the future.

The Practical Rule

For any claim built on historical data, ask: what single observation would destroy this claim?

If you can't answer, the claim is unfalsifiable. Which means it's not knowledge.

If you can answer, then you understand both what you believe and what could prove you wrong. That's the beginning of honesty about uncertainty.