Bayesian Reasoning Runs AI

Many AI systems lean on Bayesian reasoning. Spam filters use it to weigh which words signal junk, and self-driving cars use it to fuse uncertain readings from cameras, radar, and lidar into a single best guess. A car sees a shape on the road. Prior belief: the object ahead is probably a pedestrian. Likelihood: this shape matches a pedestrian at 85%. Posterior: the updated probability. As new sensor data arrives, the estimate is refined again.

Forensic DNA matching works the same way. The prior reflects how common the DNA profile is, and the likelihood reflects how well the evidence matches. The posterior tells a jury how confident to be. In the 1990s, misusing this math, an error known as the prosecutor's fallacy, led to wrongful convictions.