Escalation of Coevolutionary Arms Races

<div> <div> <p>The biological complexity present in Nature has been attributed to a number of processes, here we consider coevolutionary arms races. Previous coevolutionary research has shown that such monotonic progress is possible, but these existing monotonic algorithms employ engineered techniques not likely to have influenced Natural evolution, such as comparisons to archives of historical individuals. We study Markovian coevolution, where the only preserved information is contained within the evolving population itself. In a signaling game from evolutionary game theory we show how parameters of evolutionary dynamics influence the reachability of evolutionary equilibria. We build on these ideas to show how a model of escalation in evolutionary arms races can continue to increase in complexity under finite population dynamics, but is vulnerable to coevolutionary pathologies even at large population sizes. When using a coevolutionary game that incorporates an external environment we find that more complex external environments better facilitate coevolutionary escalation. Finally, this idea is extended to consider how population diversity can play the same role as exogenous environmental complexity by facilitating escalation in 2 models: a minimalist coevolutionary game, and a swarm game. We conclude that competitive coevolution is a productive evolutionary mechanism capable of escalation, and show that both exogenous and endogenous complexity can facilitate the escalation of coevolutionary arms races. </p><p>Harrington, K. I. S, (2014) "Escalation of Coevolutionary Arms Races". In <i>Department of Computer Science, Brandeis University</i>.</p> </div>