The parable of the hare and the tortoise: Network

Networks provide the architecture of systemic adaptation. In any social system, the units of that system seek success—whether those criteria for success are determined by evolutionary processes or by (boundedly rational) goal-seeking processes. Networks provide the architecture of success in social systems because information about success that flows through the network provides one of the primary mechanisms for improvement in the system. This paper explores the implications of this architecture for system-level success (e.g., Putnam 2000), rather than nodal-level success (e.g., Burt 1995). In particular, using simulation models, it demonstrates how network architecture affects the balance between exploration and exploitation (March 1991) within a system. It finds that small world (small number of degrees of separation) networks (the hare) are better at exploitation—they quickly converge on the best solution that exists in the network at the beginning of the simulation. However, small worlds perform very poorly at exploration, because they remain stuck on that solution. Larger world networks (the tortoise) perform better in the long run because they explore more of the solution space. This effect is greater the more local optima there are in the solution space. Further, controlling for connectedness, we find that the smaller the world the better the system in the short run, and the worse it is in the long run. Finally, we observe how social cliques and network density affect the social outcome. We argue that these findings are robust across a wide variety of settings, offering illustrations from: the diffusion of pre-written history agricultural innovations, genetic programming, and creative groups. The first section of the paper examines the role that networks play in the diffusion of information, focusing in particular how networks affect the aggregation of information. The second part of the paper develops a simulation model, basing the solution space on NK models (Kauffman 1995), to understand the implications of emulation processes, where it is assumed that actors emulate the most successful other