Computational models of attention competition

Abstract

In the new digital age, information has grown increasingly abundant and immediately available. Thus it is easy to see that the scarcest resource of today is not information but rather attention. We design several complex adaptive social systems in which agents with limited attention capacity confront a wealth of information. We take rather an exploratory approach and use primarily agent-based models to study the dynamics of competitive endeavours, such as artificial markets and games. Our purpose is to study the dynamics of cooperation and competition among boundedly rational artificial agents. (i) First we built a simple model in which cultural items compete for the limited attention of agents and we investigate the impact of advertisement pressure. We observe that the market share of the advertised item improves as a result of an increase in the standard items. (ii) Secondly, we work on attention games in a specific context of Iterated Prisoners Dilemma. We find out it is best for agents to pay attention to defectors in order to achieve a higher social welfare. Hence, cooperators becomes more prudent to the defective moves. (iii) Thirdly, we investigate the evolution of cooperation. This time agents are “hard-wired” to pay attention to defectors. Agents have limited memory size and refuse to play with defectors. As opposed to what we expect, we observe that subsequent generations loose their memory and are ultimately invaded by defectors, when playing with a defector brings non-negative payo↵s. We reformulate the payo↵ matrix structure to incorporate negative payo↵s and show how threat (of receiving negative payo↵s) fosters greater memory size and cooperation. We also observe how memory acts like an immune response of the subsequent generations against aggressive defection. This functionality of self-immunization has emerged as a result of the co-evolutionary process.