Mathematics
Bayes' Theorem and Updating Beliefs with New Evidence
Quick fact
Spam filters use Bayes' theorem to classify emails: they start with a prior probability that any email is spam, then update it using the likelihood of certain words appearing in spam versus legitimate mail. This same equation underlies medical tests, A/B testing, and even the way your brain processes sensory information.
Why this is interesting
You hear a weather forecast that says rain for tomorrow, but you look outside and see clear skies. Should you change your belief? Bayes' theorem is the mathematical rule that tells you exactly how much to update your beliefs when new evidence appears.
Read the full explanation
Understanding Bayes' Theorem and Updating Beliefs with New Evidence
Imagine you're a doctor. You know that a certain disease affects 1 in 1000 people (prior probability). You have a test that detects the disease with 99% accuracy, but it also has a 5% false positive rate. A patient comes in and tests positive. How likely is it that they actually have the disease? Most people guess around 95%, but Bayes' theorem shows the real answer is much lower—about 2%. Here's why: Bayes' theorem says that the probability of a hypothesis (disease) given evidence (positive test) equals the probability of the evidence given the hypothesis (true positive rate) multiplied by the prior probability of the hypothesis, all divided by the overall probability of the evidence. In symbols: P(H|E) = P(E|H) P(H) / P(E). The denominator, P(E), accounts for all the ways the test could come back positive—both when the disease is present and when it's not. Because the disease is rare, most positive tests are false positives, so the posterior probability remains small. This process is called belief updating: you start with a prior, adjust it as evidence arrives, and get a posterior—which becomes your new prior for the next update.
A deeper explanation
The power of Bayes' theorem lies in its ability to incorporate multiple pieces of evidence sequentially. When you receive new evidence, you use your current posterior as the prior, and apply the theorem again, refining your belief with each step. Mathematically, the denominator P(E) serves as a normalizing constant ensuring the posterior sums to 1, and it can be computed as the sum (or integral) of the likelihood times the prior over all possible hypotheses. This mechanism mirrors scientific reasoning: we have hypotheses with prior plausibility, we conduct experiments to gather data, and we update our confidence. It also reveals why evidence can be weak: if the likelihood ratio P(E|H1)/P(E|H2) is close to 1, the evidence barely changes our beliefs. This principle is the foundation of Bayesian statistics, which increasingly dominates fields from machine learning (Bayesian networks, spam filters) to clinical decision-making and even machine learning model calibration. Understanding this mechanism is key to interpreting medical test results, evaluating scientific claims, and making rational decisions under uncertainty.