Artificial societies and information theory: modelling of subsystem formation based on Luhmann's autopoietic theory

QA75 Electronic computers. Computer science;HM Sociology;QA76 Computer software;Q Science (General)

Di Prodi, Paolo ; Porr, Bernd

University:University of Glasgow

Department:School of Engineering

关键词: Multi agent systems, social systems, self organizing systems, autonomous social systems, information theory for adaptive agents, adaptive learning;   

Others  :  http://theses.gla.ac.uk/2869/1/2012DiProdiPhD.pdf

来源: University of Glasgow

【 摘 要 】

This thesis develops a theoretical framework for the generation of artificial societies. In particularit shows how sub-systems emerge when the agents are able to learn and have the abilityto communicate.This novel theoretical framework integrates the autopoietic hypothesis of human societies, formulatedoriginally by the German sociologist Luhmann, with concepts of Shannon's informationtheory applied to adaptive learning agents.Simulations were executed using Multi-Agent-Based Modelling (ABM), a relatively new computationalmodelling paradigm involving the modelling of phenomena as dynamical systems ofinteracting agents. The thesis in particular, investigates the functions and properties necessaryto reproduce the paradigm of society by using the mentioned ABM approach.Luhmann has proposed that in society subsystems are formed to reduce uncertainty. Subsystemscan then be composed by agents with a reduced behavioural complexity. For example insociety there are people who produce goods and other who distribute them.Both the behaviour and communication is learned by the agent and not imposed. The simulatedtask is to collect food, keep it and eat it until sated. Every agent communicates its energy stateto the neighbouring agents. This results in two subsystems whereas agents in the first collectfood and in the latter steal food from others. The ratio between the number of agents thatbelongs to the first system and to the second system, depends on the number of food resources.Simulations are in accordance with Luhmann, who suggested that adaptive agents self-organiseby reducing the amount of sensory information or, equivalently, reducing the complexity of theperceived environment from the agent's perspective. Shannon's information theorem is usedto assess the performance of the simulated learning agents. A practical measure, based on theconcept of Shannon's information ow, is developed and applied to adaptive controllers whichuse Hebbian learning, input correlation learning (ICO/ISO) and temporal difference learning.The behavioural complexity is measured with a novel information measure, called PredictivePerformance, which is able to measure at a subjective level how good an agent is performinga task. This is then used to quantify the social division of tasks in a social group of honest,cooperative food foraging, communicating agents.