Artificial Intelligence (AI) is currently one of the fastest growing areas of computer science. In the past few years the interest in this field has transcended academic development, achieving a variety of applications, potentially influencing every modern economic activity. Much of this development is due to the recent availability of large data sets, as well as of computational resources, and is linked to a single field of AI, namely Machine Learning (ML). This fast development brought notoriety to the field, however, being confined to a single subarea of AI, it is deemed to achieve a standpoint, since the outputs of ML systems do not have explicit justifications that allow them to be audited. Therefore, machine learning algorithms cannot be fully trusted when dealing with human (sensitive) data. The key reason for the lack of justification for the ML outputs is that these algorithms have been largely developed independently from the AI subfield of Knowledge Representation and Reasoning. The future development of AI should consider the development of machine learning systems with knowledge representation capabilities allowing them to provide an explanation to their computed decisions. A large part of my current research has been devoted exactly at developing feasible and efficient combinations of machine learning algorithms (specially convolutional neural networks and reinforcement learning) with high-level ontologies capable of describing the outputs of ML algorithms. This facilitates reasoning about ML decisions and the inference of new facts about the underlying domain. In particular, I have been investigating how the output of a convolutional neural network, that associates visual scenes to English sentences, could be explained by means of a spatial ontology representing the domain objects and their attributes.