?
Presumptions of Semantic Representations Evolution
To date, traditional information technology and artificial intelligence technology have evolved independently of each other. Now is the time to fundamentally rethink the experience of using and evolution of traditional information technology and its integration with artificial intelligence technology. Currently, the key problem in the development of information technology in general and artificial intelligence technology in particular is the problem of ensuring the information compatibility of computer systems, including intelligent systems. One of the advantages of combining knowledge bases - the core of modern information systems - is the use of a more or less standard way (in the form of semantic networks) to represent knowledge. However, the sole use of semantic networks to represent knowledge does not solve the problem of standardization. For such a representation to become generally accepted, it must be based on the fundamental principles of describing semantics. The most fundamental concept in this sense is the representation of knowledge in human consciousness. Semantic representations in the modern sense - that is, thesauri, semantic networks, ontologies - are a very rough description of the highest level of reality where we live, which can only conditionally be called a model of the world. In order to understand the architecture of the semantic representations that adequately describe the human world (as the modeled object), it is necessary to reproduce the architecture of the human world model as it is formed in the human consciousness. To understand this architecture, one need to look at the human brain (a natural neural network that forms semantic representations), at its (brain) organs that form this architecture, at the cognitive networks that form in it in response to particular situations at its input and output in the process of solving specific problems, at the informatics of individual sensory and effector modalities, the levels of hierarchical representations of which contain images of events of the external and interoceptive world of human of varying degrees of complexity, at the combinations of these hierarchies in multimodal representations, including those in the description of entire situations, as well as the sequence of situations, how they are used in the process of the unconscious and purposeful behaviour - its planning and implementation control. Obviously, the brain is both very large and very heterogeneous neural network that is complex in architecture. Such representations are not only well supported by a comparison with the architecture and informatics of the brain, but are also effectively modelled in applications. TextAnalyst, a software technology for automatic semantic analysis of unstructured texts (which is based on an artificial neural network based on neurons with time summation of signals), effectively implements the functions of forming a homogeneous semantic network, automatic abstracting of texts, comparing texts by their meaning, as well as classifying and clustering texts. It can be assumed that this technology will also effectively analyze code sequences obtained in the analysis of video sequences, if this analysis is sufficiently bionic. A single approach to the processing of textual and visual information will enable constructing effective multimodal systems for processing and presenting information, which is the only accurate approach to the modelling of human intellectual functions.