Confirmation Bias: How You Prove Yourself Right

Here's a fact about how your mind works: You do not search for evidence impartially.

You search for evidence that confirms what you already believe. When you find confirming evidence, you feel satisfied. When you encounter disconfirming evidence, you find reasons to exclude it.

This is not a weakness of untrained minds. It is how trained and untrained minds alike process evidence. Taleb calls this confirmation bias, and it is the core error that keeps people imprisoned in false beliefs — sometimes with catastrophic consequences.

The Asymmetric Logic

Understanding confirmation bias requires understanding a principle of logic that sounds simple but is almost universally misunderstood:

No number of confirming observations can prove a universal claim.

I could show you a million white swans. Each one is consistent with the claim "all swans are white." But the claim is not proven. The next swan could still be another color.

One disconfirming observation destroys a universal claim.

A single black swan is sufficient to falsify "all swans are white." The asymmetry is mathematical. It applies to every universal claim about "all investors," "all economies," "all technologies," "all people." The asymmetry is absolute.

This means: intellectually honest inquiry should hunt disconfirmations, not confirmations. The hard work of evidence-gathering is finding the thing that could prove you wrong, then testing whether you are.

But we do the opposite. We seek evidence that validates what we already think. When we find it, we congratulate ourselves for rigor.

This is confirmation bias in its purest form.

The Investor Who Only Reads Good News

I've watched this pattern in action many times, and Taleb describes it perfectly:

A retail investor buys a stock. From that moment on, they read analyst upgrades and ignore downgrades. Positive earnings surprises become evidence the thesis is working; negative ones become "one-time items" or "the market is being shortsighted."

Supportive news is shared with friends and family. Skeptical news is dismissed as uninformed or bearish for its own sake.

Over time, the investor accumulates mountains of confirmation that the position is sound. The position, meanwhile, may be deteriorating by every objective measure.

This happens to sophisticated investors as well as retail ones. The asymmetric filtering of confirming vs. disconfirming evidence is a universal human tendency. It is the single most expensive bias in financial markets.

The corrective move is obvious in theory and excruciating in practice: weight the disconfirming observation more than the confirming one. If your thesis is strong, it should be able to survive hard challenges. If it cannot, the thesis is not actually strong — it is just appealing to you.

The Researcher and the Subgroup Analysis

In medical research, confirmation bias appears as a particular pathology:

A researcher proposes that a drug lowers cholesterol. The researcher designs a trial. Results come back mixed. Some patients improved; others didn't. The hypothesis is on the edge of statistical significance.

So the researcher performs subgroup analyses. Does the drug work better for patients over sixty? For women? For patients with a specific genetic marker? For patients with high baseline cholesterol?

Eventually, the researcher finds a subgroup where the drug's effect is strong and statistically significant. A paper is published celebrating the finding.

This is confirmation in action. The researcher went hunting for where the hypothesis would be supported and stopped searching when it was found.

Real science requires the opposite: specify the conditions under which the hypothesis would be rejected before running the test, then run the test. Stop when the pre-specified conditions are met. Do not keep searching until you find something promising.

Most published medical research does not do this. Which is why so much of it does not replicate.

The All-Swans-Are-White Asymmetry

Here's how the asymmetry works in real belief systems:

A researcher claims that a particular phenomenon exists or is true. You show them a hundred cases where the claim holds. Consistent with the claim. The researcher smiles.

You show them another hundred cases where it holds. Now they are confident.

You show them a thousand confirming cases. The researcher is very confident now. They may even publish a paper based on this accumulated evidence.

But then someone finds a single case where the claim does not hold — one counterexample, one disconfirmation.

That one case should destroy the claim of universality. It does, mathematically. But in practice, the researcher finds reasons to exclude the counterexample: it was an unusual case, an exception, a special circumstance that doesn't reflect the true pattern.

The thousand confirmations mean very little. The one disconfirmation should mean everything. But we treat the weight backwards.

The Round-Trip Fallacy

Confirmation bias often works in tandem with another error: the round-trip fallacy. This is where conditional probabilities get inverted, sometimes with catastrophic policy consequences.

Here's an example: "Most smokers do not get lung cancer."

This is statistically true. Most smokers do not develop lung cancer. The statement is factually correct.

But someone uses this statement to argue: "Therefore, smoking isn't dangerous."

Wait. That's not what the original statement says. The original is P(no cancer | smoker) — the probability of no cancer given that someone smokes. That is indeed high, maybe 90%.

But the claim "smoking isn't dangerous" would require P(cancer | no smoking) — the probability of cancer for non-smokers — to be approximately equal to P(cancer | smoker). And it is not.

The two conditional probabilities go in opposite directions and are wildly different numbers. Smoking overwhelmingly causes lung cancer even though most smokers don't get it.

Confusing them is the round-trip fallacy, and it operates everywhere.

"Most terrorists are Muslim" is reversed into "most Muslims are terrorists" — a complete inversion, and a dangerous one.

"Most successful entrepreneurs are disciplined" gets reversed into "most disciplined people become entrepreneurs." Not true at all.

The directional confusion produces some of the worst policy errors of our age.

The Medical Test Paradox

Here's a specific case where the round-trip fallacy kills people:

A disease affects 1 in 10,000 people. A test for it is 99% accurate — meaning it correctly identifies sick people 99% of the time and correctly identifies healthy people 99% of the time.

You test positive. What's the probability you have the disease?

Most people say 99%. It feels obvious.

The actual answer, using Bayes' rule, is about 1%.

Here's why: Among 10,000 people, 1 has the disease and correctly tests positive. Among the remaining 9,999, 1% of them (99 people) falsely test positive.

So among everyone who tests positive, there are about 100 people total (1 true positive + 99 false positives). Only 1 of them actually has the disease.

You are 99 times more likely to be a false positive than a true positive.

Almost everyone — including doctors — gets this wrong. They confuse P(positive | disease) with P(disease | positive). Enormous medical and legal errors trace back to this one confusion.

The Practical Rule

How do you protect yourself from confirmation bias?

Taleb offers a method that sounds simple but is nearly impossible to execute:

For any hypothesis you favor, name the single observation that would make you abandon it.

If you cannot name one, the hypothesis is unfalsifiable. If it is unfalsifiable, it is not knowledge. It is a faith position.

Try this with your own beliefs. Pick something you are confident about. Then ask yourself: what would have to be true for me to change my mind?

If you cannot answer, you are not actually reasoning about this — you are defending a position.

For investing, this means naming the exit conditions before you enter. Not vaguely ("I'll sell if things get bad"), but specifically. "I will sell if the company misses guidance three quarters running," or "I will sell if the patent is denied," or "I will sell when the stock rises 40%."

If you cannot name the condition in advance, you don't actually have an investment thesis. You have a hope.

Institutional Applications

Confirmation bias is not just an individual cognitive error. It is baked into institutions.

In corporate culture, the confirmation bias works like this: the executive's belief in a strategy, once established, becomes the organizing principle of the entire organization. Every team finds evidence that the strategy is working. Data that contradicts the strategy are dismissed as anomalies or short-term noise.

The organization becomes efficient at confirming the leadership's hypothesis, which means it becomes blind to disconfirmations.

By the time the market has moved decisively against the strategy, the entire institution is still in confirmation mode.

This is why some of the largest companies have been blindsided by technological shifts. They were too good at confirming their existing strategy and too poor at hunting for disconfirmations.

The Path Forward

The asymmetric logic is available to you. It is not secret. One disconfirmation destroys a universal claim; confirmations prove nothing.

Yet you will still seek confirmations. You will still find reasons to exclude disconfirmations. You will still be surprised when your confident beliefs turn out to be wrong.

The task is not to eliminate the bias — you cannot. The task is to structure your decisions and your institutions so that the bias does not ruin you.

Ask yourself regularly: what would prove me wrong? What am I not seeing? If you can construct a belief system where disconfirmations are actively sought rather than excluded, you will be better-calibrated to the world.

Most people and most institutions move in the opposite direction. They become more confident in their views over time, not less. They gather more evidence that confirms their position, which makes them more confident.

Confidence and accuracy are not the same thing. A belief that is unfalsifiable is not confident — it is trapped.