Genetic algorithms that predict a time series, for example, the affairs of a company on the stock market, are 'trained' by running a huge amount of information through them and selecting ones that produce the closest representation of what will happen soon on in the series. With stock markets, it is possible to get records from many years back, and algorithms can be trained with this data before actually risking anything in real life.

A research team created a genetic algorithm at some time in 1995-96 to predict the activity on the stock market, and it has made some impressive predictions. One day, the program suggested that the researchers "buy Apple", and, sure enough, Apple Computer's share price soared several days later. The researchers had absolutely no idea how the program had deduced this, or whether it was completely random, but the program had evidently been trained well from the previous data!