Herself’s Artificial Intelligence

John Holland in the mid 1970s designed the first ‘classifier systems’. These systems take inputs, match them to known conditions then perform the action requested.

Inputs can be true, negative, or not relevant ( 1, -1, 0 ). Say we have a game character who can be hungry, thirsty both or neither. We would then have inputs:

Inside our program we will have a table of four possible conditions and a specific action for each.

I. If -1, -1 keep doing what we were doing
II. If -1, 1 stop and get a drink
III. If 1, -1 stop and eat
IV. If 1, 1 stop eat and drink.

If some conditions over lap we can use statistics to find the closest match. In newer classifiers we also allow for learning new conditions.

If no existing condition matches our state create a new condition. Usually we start by adding zero ( not relevant ) to existing conditions in various inputs and building a rule based on that. If we run out of rule space we drop a little used rule from our database.

Currently these are used in determining what type of melanoma a person has; Price Grabber uses these algorithms to sort and fetch prices; and in NLP ( natural language processing ) applications.

For more information:
Natural Language Processing
Classifier System Abstracts
What’s a Classifier System
Wiki, Learning Classifier System