The story
In December 1763 the Royal Society in London heard a paper called An Essay towards Solving a Problem in the Doctrine of Chances. Its author, the Presbyterian minister Thomas Bayes of Tunbridge Wells, had died in 1761, and his friend Richard Price had found the manuscript, edited it and added an introduction. It showed how to work backward from evidence to the likely cause: given what we have observed, how should we revise our estimate of what is true? That idea is now called Bayes' theorem.
Probability theory before Bayes mostly ran forward. Jacob Bernoulli's Ars Conjectandi (1713) and Abraham de Moivre's work could tell you the chance of drawing certain cards or tossing certain results from a known setup. They could not easily answer the reverse question that scientists and ordinary people face: given a handful of results, what can be said about the unseen setup behind them? Bayes imagined a ball rolled onto a table at an unknown spot, and then further balls rolled, with each reported only as falling to the left or the right of the first ball. How should one guess where the first ball lies? His method combines a starting belief with new data.
In modern form the rule says that the probability of a hypothesis after seeing evidence equals its probability before, multiplied by how well it predicts the evidence, and scaled to total one. Bayes did not write it in this shape. The French mathematician Pierre-Simon Laplace independently found and generalized it in the 1770s and later. For much of the twentieth century statisticians preferred 'frequentist' methods that avoid prior beliefs, but computers revived Bayesian methods from the 1980s and 1990s. Today Bayesian inference is used in medical tests, spam filters, weather and search models, and many machine learning systems.
Why it mattered
- Bayesian inference became a core method of statistics, letting analysts update estimates as new data arrives.
- Laplace's use of the rule in astronomy and later work spread the idea to the sciences.
- Machine learning and artificial intelligence use Bayesian ideas for spam filters, speech recognition and uncertainty estimates.
- Doctors and researchers use it to interpret test results, since the same result means different things depending on how common a disease is.
Sources
- An Essay towards solving a problem in the Doctrine of Chances (Philosophical Transactions, 1763) Royal Society (reprint at Georgia Tech)
- Thomas Bayes Advances Probability Theory EBSCO Research Starters
Drafted by AI (Claude) from the sources above and checked against them by AI; no historian has reviewed this page. Spot a mistake? Suggest a correction. How pages are made
Community notes
Loading notes…