Chunking

• When I was in graduate school, I worked with John Laird who has a cognitive architecture called Soar.
• One of the features of Soar is that it represents a lot of its knowledge in rules.
• It can also learn new rules in a process called chunking.
• While I dig Soar, the point is that systems can learn new rules.
• There are a lot of ways to do this.
• One type of system just learns association rules. The Tesco Nappy Beer example works here, but association rules work pretty well with big data.
• Another thing you can do is make rules more specific or more general.
• One of the nice things about rules is that people can generally understand them.