The Analysis of Ontology-Based Neuro-Symbolic Intelligence Methods for Collaborative Decision Support

Abstract

The neural network approach to AI, which has become especially widespread in the last decade, has two significant limitations – training of a neural network, as a rule, requires a very large number of samples (not always available), and the resulting models often are not well interpretable, which can reduce their credibility. The use of symbols as the basis of collaborative processes, on the one hand, and the proliferation of neural network AI, on the other hand, necessitate the synthesis of neural network and symbolic paradigms in relation to the creation of collaborative decision support systems. The article presents the results of an analytical review in the field of ontology-oriented neuro-symbolic artificial intelligence with an emphasis on solving problems of knowledge exchange during collaborative decision support. Specifically, the review attempts to answer two questions: 1. how symbolic knowledge, represented as an ontology, can be used to improve AI agents operating on the basis of neural networks (knowledge transfer from a person to AI agents); 2. how symbolic knowledge, represented as an ontology, can be used to interpret decisions made by AI agents and explain these decisions (transfer of knowledge from an AI agent to a person). As a result of the review, recommendations were formulated on the choice of methods for introducing symbolic knowledge into neural network models, and promising areas of ontology-oriented methods for explaining neural networks were identified.