Using Learning Algorithms to Create, Exploit and Maintain Knowledge Bases: Principles of Constructivist Machine Learning.

Recently, interest has grown in connecting modern machine learning approaches with traditional expert systems. This can mean, e.g, to identify patterns with neural networks and integrate them with knowledge graphs. While such combined systems offer a variety of advantages, few domain-independent approaches are known to make a hybrid arti-ficial intelligence applicable without human interaction. To this end, we present the implementation of a constructivist machine learning framework (conML). This novel paradigm uses machine learning to manage a knowledge base and thereby allows for both raw data-based and symbolic information processing on the same internal knowledge representation. Based on axioms for a constructivist machine learning, we describe which operations are required to create, exploit and maintain a knowledge base and how these operations may be implemented with machine learning techniques. The major practical obstacle in this approach is to implement an automated deconstruction process that avoids ambiguity, handles continuous learning and allows knowledge abstraction. As we demonstrate, however, these obstacles can be overcome and constructivist machine learning can be put into practice.