Supplementing Machine Learning with Knowledge Models Towards Semantic Explainable AI

Explainable Artificial Intelligence (XAI) aims at making the results of Artificial Intelligence (AI) applications more understandable. It may also help to understand the applications themselves and to get an insight into how results are obtained. Such capabilities are particularly required with regard to Machine Learning approaches like Deep Learning which must be generally considered as black boxes, today. In the last years, different XAI approaches became available. However, many of them adopt a mainly technical perspective and do not sufficiently take into consideration that giving a well-comprehensible explanation means that the output has to be provided in a human understandable form. By supplementing Machine Learning with semantic knowledge models, Semantic XAI can fill some of these gaps. In this publication, we raise awareness for its potential and, taking Deep Learning for object recognition as an example, we present initial research results on how to achieve explainability on a semantic level.