Representing the World

Representing the World

Copyright: © 2021 |Pages: 27
DOI: 10.4018/978-1-7998-5653-5.ch001
OnDemand PDF Download:
Available
$37.50
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Chapter 1 describes how specifically organized, hierarchical structures of a neural network can create neural representations of perceived reality. The authors describe how, as a result of categorization and generalization, memory traces created in subsequent layers can represent the perceived world in all its complexity. Starting from the representation of direct sensual impressions in the lowest layers, closely connected to the sensors of individual senses, to the representation of increasingly complex objects, the feelings and knowledge about the observed world are built. They postulate that to achieve this goal imaginary natural and artificial brains must contain such semihierarchical structures capable of creating new connections and information transmission paths. By associating large areas of brain fields in multiple layers, it is possible to create representations of complex reality. The dominant mechanism of self-learning is correlation learning, during which simultaneous, synchronous arousal of different senses creates mutually correlated features of the observed object. Perceived objects excite neuronal stimulation patterns that allow the system to identify the object in the future. The re-stimulation of the memory structures from the top layers to the sensory fields, causes the recall and creation of sensations similar to those felt during the original experiences. By comparing new sensual impressions with those stored in memory, the perceived objects are recognized. Frequent, simultaneous co-occurrence of stimulations of mental representations results in associations of memory cells and synapses, and thus associations of mental facts. Order and sequences of their occurrence is the basis of episodic memory. Imagined neural network memory cells, like natural brain neurons, do not limit their role to just remembering the information that they receive. They actively process this information and change the structure of their connections. We put forward the thesis that the described memory cells, artificial neurons, can create brains with features such as natural brains. It is this semihierarchical structure of neurons, which arise from categorization, generalization and association processes that can create neural representations of perceived reality. Learning through life experiences allows us to give them the characteristics of psychological sensations and thus they also become mental correlates of perceptions. The knowledge that these structures represent is as hierarchical they are. This hierarchy starts from the representation of the simplest direct sensual features, to complex models of the environment and abstract concepts that can be defined by symbolic language. The presented model describes the creation of knowledge in the mind, pattern recognition, remembering and imagining objects and events, planning, and making decisions. The systems thus created yield minds with cognitive, intentional, and propositional awareness. Unfortunately, they are devoid of phenomenal awareness, which we write about in the following chapters.
Chapter Preview
Top

Structures Of Memory

To create a reductive model of consciousness, it is advisable to propose the form and functioning of neural representations of perceived reality in our brains. For this purpose, imagine that there are structures in the brain in which observations and our thoughts—concepts created and expressed with language, describing perceived objects and patterns of processes that are observed and concocted in our mind—can be remembered. Those patterns may concern transforming objects or groups of objects—enlarging, shrinking, stretching, rotating, mirroring, adding, folding, and covering them. They can also apply to complex objects and their sets representing the environment, the scene in front of our eyes or imagined models of the environment, and even the whole world. Imagined structures might be reflected by some small elements, memory cells, connecting in a specific way and remembering information that reaches them. Now, let’s imagine that these elements—cells—are arranged in layers. Elements are connected in such a way that a group of elements in a bottom layer connects through multiple links to a group of elements in a higher layer. They may also connect between each other. There may be a lot of these connections, because every cell of a lower layer connects to many cells in a higher layer. Therefore, the cell may transfer information received from the lower levels to higher or lower levels, or to neighboring elements.

We can call these elements memory cells because this is their main role. They should be connected in a way that allows the transfer of information. We guess that these memory cells are neurons, and a network of connected neurons creates a neural network. The concept of a neural network has a diametrically different meaning in modern artificial neural networks researched in artificial intelligence than in the real network of neurons in biological brains. We don’t have to consider this now, however. We will try to define the minimal requirements for the network to represent memory structures so that the system with this kind of network can think and be conscious. In our imagined network of cells, we can distinguish hierarchical layer structures from layers on the lowest level, connected directly with sensory cells, up to layers on top of this hierarchy. These layers can be divided into fields that, due to their functions of remembering information they receive, we will call memory fields.

For our considerations we can assume that all memory elements are identical and perform the same functions. However, they have to be connected in a specific way that affects the functioning of the created network. We will have to determine how elements—memory cells and their groups—should act when they receive different kinds of stimulation. These stimulations may come from sensory cells, other neighboring cells, or cells located at lower layers. The main direction of transferring information from the sensory or memory cells of lower layers to higher-layer cells is shown in figure 1. This way a structure of connections is formed in which large numbers of cells of the lower layer are linked to a smaller number of cells in a higher layer. Because this configuration of cells doesn’t include all cells, it’s possible in our layered architecture to create many similar structures, and cells of the lower levels may be part of many structures, as is shown in figure 1. The hierarchical character of memory structures emphasized here will matter in further description.

Figure 1.

Hierarchical structure of the memory.

978-1-7998-5653-5.ch001.f01
Only a few example connections are shown. The structure of semblions and subsemblions defined and described in chapter 2 may be observed. Couplings between semblions aren’t shown. Groups of memory cells in certain layers that are connected in a way that allows the transfer of stimulations between them are marked with the same color. Only existing tracks of stimulations leading from lower to higher layers are marked on the picture. Coming down, feedback stimulations and couplings between cells on the same level (layer) have been omitted. The figure shows that one cell of a higher level may be stimulated with signals coming from many of the lower-layer cells. Group of cells that are marked with the same color in certain layers symbolize complexes corresponding to specific features of objects. We call them memory fields. Small memory cell groups (or single cells) representing memory field groups of many lower layers are called concept cells (cells marked with pink, yellow, and blue in layer i+2).

Complete Chapter List

Search this Book:
Reset