It was remarked earlier (Section 2.2.1) that a US Political Analyst would ``obviously'' be able to determine a Congressman's party given their votes. However, would this necessarily be true? Put another way, would this expert's knowledge base necessarily yield better results than a learning algorithm operating on an observed dataset? After all, there are always exceptions to rules, and people can be detrimentally biased by their knowledge base, so does the knowledge base really help? The person does, however, have the advantage that if they don't know the answer to a given question, they can attempt to learn what is required to answer it.
This then begs the question of can eager learning algorithms discover what they need to learn in order to make correct conclusions? An interesting experiment would be to compare the results of a human expert with standard ML algorithms and eager learning expert systems.