Escaping from the Box

This paper presents a solution to the problem of the creation of new terms by suggesting that representations can come from the environment, and that through experience an agent discovers which representations are useful. If suitable feature building mechanisms are built into a system then it can build a large number of features of an object in the environment, many of which may be redundant. Then, through experience, the system can learn which of these features are relevant to the domain by allowing useless features to atrophy, and by building new features to form necessary distinctions. These ideas are implemented in the Contextual Memory System (CMS) which is part of the Mathematics Understander (MU) program.