Bayesian Probability and How to Live With Uncertainty

We like to divide judgments into right and wrong. Bayesian probability begins with a less comfortable admission: most of the time, we possess only part of the evidence.

Something is true or false. A person is trustworthy or untrustworthy. A choice will succeed or fail. Those categories are satisfyingly clean. Reality rarely cooperates. We often have to judge with incomplete information and then live with the consequences of being wrong.

Bayesian probability deals with exactly this condition.

Strictly speaking, Bayesian reasoning is not one algorithm. It is a way of revising a judgment when evidence changes. It does not require us to know the answer at the beginning. It asks only that we adjust our original view reasonably when new information arrives.

From prior to posterior

Bayes' theorem can be written as:

P(H | D) = P(D | H) × P(H) / P(D)

H is a hypothesis and D is new evidence.

P(H) is the prior probability, our degree of belief in the hypothesis before seeing the new evidence.

P(D | H) is the likelihood. It asks how probable the evidence would be if the hypothesis were true.

P(H | D) is the posterior probability, our revised judgment after taking the evidence into account.

The formula looks abstract, but ordinary life is full of Bayesian updates.

Suppose a friend who is usually reliable does not reply to a message. Two explanations may occur to you: your friend is busy, or is deliberately avoiding you.

Before this silence, you already knew something about the person. Your history together forms the prior. If the friend has been reliable for years, "busy" should begin with a higher probability.

Only then do we examine the new evidence: no reply.

The question is not how uncomfortable the silence makes you feel. It is how strongly that silence distinguishes the two explanations. Busy people may not answer. People avoiding you may not answer either. The evidence fits both hypotheses, so it is less discriminating than the emotion it causes.

If more evidence appears, perhaps repeated cancelled plans with no explanation, then the balance may move decisively toward the second hypothesis. Revising your judgment is now reasonable.

Bayesian reasoning cannot tell us what the friend is thinking. It reminds us not to reverse a long-standing judgment because of one ambiguous signal.

A prior is not a bias that must be eliminated

Prior assumptions are often treated as enemies of good judgment. From a Bayesian point of view, judgment without a prior is almost impossible.

When we think about an unfamiliar industry, a relationship or a new opportunity, we inevitably use accumulated experience. A person who claims complete objectivity still relies on default assumptions. They may simply remain unspoken.

The question is not whether we have priors. It is where they came from and whether evidence is allowed to change them.

A prior built from long experience is usually more reliable than a passing mood. A prior built from stereotype deserves more caution. Nor should a prior be fixed at zero or one. Once we decide that something is absolutely impossible or certain, no new evidence can move the conclusion.

One of the most dangerous habits in life is treating a provisional view as an unchangeable fact.

The strength of evidence is not the shock it produces

News can be startling without being informative.

To judge the strength of evidence, compare two questions. If my hypothesis is true, how likely am I to see this evidence? If another hypothesis is true, how likely am I to see the same thing?

When a phenomenon appears under many explanations, it does little to prove any one of them. A person's silence may come from anger, exhaustion, work or simple forgetfulness. The silence may provoke emotion without identifying the cause.

Other evidence looks ordinary but discriminates sharply. If an outcome is common under one hypothesis and rare under its alternatives, it deserves to change our judgment substantially.

This is useful when reading news, understanding another person or observing ourselves. Ask less often, "Did this shock me?" Ask instead, "If my original judgment were wrong, would I still be likely to see the same thing?"

One update should not become a final verdict

Bayesian probability allows belief to keep changing. Today's posterior becomes tomorrow's prior. The next piece of evidence triggers another update. Knowledge is not a cabinet of fixed conclusions. It is a condition under continuous revision.

Human beings do not find this easy.

We turn opinions into identities. Once we have supported a view in public, revising it can feel like admitting personal defeat. To defend the position, we ignore hostile evidence or invent a special explanation for every counterexample.

Bayesian thinking asks us to separate belief from self. My judgment can be wrong without making my entire person a failure. Changing my view is not inconsistency when the change follows new evidence.

Updating can also go too far. One new fact should not erase years of experience automatically. The proper size of an update depends on the quality of the evidence and how much information it adds to what we already knew.

A Bayesian person neither clings to an old judgment nor lets the latest headline drag it around.

Look for information that can really change your mind

Bayesian reasoning suggests a practical question:

What evidence would make me change my view?

If the answer is "nothing," then we are probably defending an identity, a wish or a faith rather than a testable judgment.

If such evidence exists, the next step is to seek it.

When considering a job, do not spend all your time imagining the future. Investigate facts that distinguish one future from another. When a relationship is trapped in misunderstanding, do not keep guessing the other person's motives. Look for a conversation that separates the competing explanations. When debating whether a project will work, run a small and controlled experiment that produces a great deal of information.

Information does not have equal value. The most valuable information makes different hypotheses predict different results. Only then can an observation move our judgment substantially.

Probability cannot make a decision for us

Knowing how likely an event is does not tell us what to do. Action also depends on reward, cost and tolerance for risk.

Even if rain is unlikely, carrying an umbrella may be sensible because the cost is low. Even if a project's chance of success is respectable, a failure that would destroy the whole organization calls for a more conservative plan.

Bayesian inference updates our understanding of the world. A decision combines that understanding with the consequences of action.

This is why two rational people can agree on the probability and still choose differently. They may agree about the facts but face different costs.

Admitting uncertainty is not weakness

Certainty of tone is often mistaken for competence. The more confidently someone speaks, the more credible that person can appear.

Bayesian thinking uses another standard. A reliable person need not sound certain about every question. The person should know which evidence supports a judgment and what would cause it to change.

Mature judgment is not permanent correctness. It is a form of revisable honesty: reach a provisional conclusion from the available evidence, update it when evidence changes, and do not pretend that you knew the answer from the start.

We cannot remove uncertainty from life. We can keep our beliefs fluid and make sure that every revision has a reason.

That may be the most important lesson Bayesian probability offers everyday life.