A model of Consciousness With neural networks By: Hadiseh Nowparast

Overview:  Neural Networks  The Global Workspace Theory  A hybrid neural network model for consciousness

About neural networks

McCulloch-pitts model of a neuron:

A taxonomy of feed-forward & feed-back network architectures:

Applications:

The global workspace theory

 This theory provides a well-formalized model of consciousness, identifying and describing the actors that are involved in consciousness as well as their interactions, roles and function.  In this model, consciousness is part of a cognitive distributed system.

 In order to regulate all the possible interactions, the system needs atleast a bottleneck to force the modules either to compete or to collaborate.  In Baars’ model, the bottleneck is a working memory (short-term memory) called the global workspace which represents the consciousness.

A view of the main components of Baars’ model:

 Unconscious specialized processors they represent a simple skill, a basic knowledge (e.g., the ability to make a simple addition, to understand a sentence in our mother tongue, to perceive temperature, etc.). the processors are by definition unconscious as they represent the set of competences of a given system.

 Consciousness or global workspace when an unconscious specialized processors is unable to achieve a given task by itself, it accesses the consciousness that will in turn broadcast the needs of the processors to the entire system.

 Coalition and conscious processes the unconscious specialized processors that answer the broadcasted needs form together with the requesting processors a coalition able to achieve a given task. in turn, the coalition may need further skills; it will then iteratively access the consciousness in order to have the needs broadcasted to the system. A so-formed coalition is called a conscious process.

 Contexts they are the assumption of the system that restrain access to the consciousness and discharge the coalition formation process. if the specialized processor wants to access consciousness, it has either to fit the context or to wait until the context is deactivated.

A hybrid neural network model for consciousness

 This framework reflects explicit connection between two parts of the brain: Global working memory Distributed modular cerebral networks relating to specific brain functions

 This framework is composed of three layer : Physical mnemonic layer Abstract thinking layer Recognition layer

 Two first layers cooperate together through third layer to accomplish information storage and cognition using algorithms of how these interaction contribute to consciousness: the reception process the reception process whereby cerebral subsystems group distributed signals into coherent object patterns. The partial recognition process The partial recognition process whereby patterns from particular subsystems are compared or stored as knowledge. The resonant learning process The resonant learning process whereby global workspace stably adjusts its structure to adapt to patterns’ changes.

This framework postulates that: 1. At any given time, many modular cerebral networks are running in parallel and process data from outside in an unconscious manner. those networks belong to the physical mnemonic layer of the framework. an assumption is made about categorizing the outside inputs into two groups: aware inputs and arousal input. (the latter can only reach the recognition layer while the former can break out of the recognition layer and take part in the associative recognition in the global workspace.)

2. A recognition layer is a searching tree composed of layered storage neurons positioned by the inherent frequencies. all feature of an object pattern are processed in the lowest level by mnemonic layer networks before entering here. Some neurons in a particular level are grouped to form a cluster. The representative of a cluster in the nth level becomes a point in the higher-level n+1.

3. Global workspace that belongs to the abstract thinking layer can potentially interconnect multiple cerebral networks at the physical mnemonic layer through the recognition layer. when the global workspace is active for some duration, the abstract information in the thinking layer is available to variety of processes that would be mobilized by top- down intentional projection into cerebral actions that may involve several distributed neural networks. This global activity of abstract information through workspace is defined as the conscious state of this framework.

The general framework of consciousness:

References  A hibrid neural network model for consciousness,2004  A model of agent consciousness and its implementation,2006  Artificial neural networks: A tutorial

The end