Bayesian Updating

Bayesian updating is a formal way to change your mind. It’s the process of adjusting the strength of your belief in an idea when you encounter new information. You start with an initial belief, and then, as you gather evidence, you incrementally update that belief to be more accurate. For example, if you believe a new restaurant is great but then read three negative reviews, you should rationally become less confident in your initial positive opinion.

Try this: What new information have I seen today, and how should it slightly adjust one of my existing beliefs?

Think of your beliefs not as on/off switches, but as dials of confidence. Bayesian updating is the formal method for turning that dial. The process involves three core components:

• Prior: Your starting belief before you see new evidence (e.g., your initial confidence level from 0% to 100%).
• Evidence: The new information or data you observe.
• Posterior: Your updated belief after considering the evidence. Yesterday’s posterior becomes today's prior.

A doctor might have a low prior belief a patient has a rare disease. A positive lab test (evidence) forces her to update to a much higher posterior belief, preventing her from dismissing an unlikely but possible diagnosis.

Similarly, a business team with high confidence in a new product should lower its belief if early sales data is weak. This transforms 'being wrong' from a failure into a necessary, data-driven update.

The strength of your prior matters; a very strong initial belief requires very strong counter-evidence to change it significantly.

Bayesian updating provides a mathematical rule for how a rational agent should change their degree of belief given new evidence, making it a cornerstone of modern statistics, machine learning, and cognitive science.

• Its origins trace to Thomas Bayes' 18th-century work on conditional probability, which laid the groundwork for the theorem bearing his name. [1] (1763)
• Some cognitive scientists argue the human mind is an intuitive "Bayesian machine" that approximates these principles for everyday reasoning under uncertainty. [2] (2011)
• In modern statistics, it provides a complete framework for building models probabilistically, emphasizing flexible thinking over rigid formulas. [3] (2020)

As a description of human psychology, Bayesian updating is imperfect. People often fail to update rationally due to cognitive biases like confirmation bias. The model also requires a 'prior' belief, and the choice of this starting point can be subjective and heavily influence the final conclusion—a long-standing debate in statistics. [4]