An automated learning system for improving graphical user interfaces. Information about a user transaction which caused an error is obtained from an input to an element in a graphical user interface. Associations are created between the error and one or more elements in the graphical user interface which generated the error to form association rules. The association rules are applied to a set of control rules to create user interface preference rules. A corrective action may then be performed on the one or more elements in the graphical user interface according to the user interface preference rules.