Global Adaptation in Networks of Selfish Components: Emergent Associative Memory at the System Scale

The conditions under which numerous independently-motivated components or agents in a complex system create globally efficient structures, behaviours or functions remains a fundamental open question for domains such as ecology, sociology, economics, organismic biology and many others. Here we show that if agents modify their relationships with other agents slowly compared to relatively rapid changes in their own behaviours then a system of selfish agents can self-organise its interaction structure such that more efficient combinations of behaviours arise. At the individual level we find that selfish changes to relationships cause behaviours that frequently co-occur under the same conditions to become more strongly connected. At the system level this means that modified connections effect an implicit distributed associative memory of behavioural combinations that are frequently visited. The improved global efficiency that results can thus be understood via the inherent ability of associative learning to generalise by idealising stored patterns and/or creating new combinations of learned features. These findings suggest that distributed complex adaptive systems of self-interested components as diverse as social networks, ecosystems, economic markets and organisms composed of selfish genes may exhibit organisational principles in common with those familiar in connectionist models of organismic learning, developing an associative memory of their past behaviour that enhances system-level efficiency