The evolution of social dominance through reinforcement learning

Abstract

Groups of social animals are often organised into dominance hierarchies that are formed through pairwise interactions. There is much experimental data on hierarchies, examining such things as winner, loser, and bystander effects, as well as the linearity and replicability of hierarchies, but there is a lack evolutionary analyses of these basic observations. Here I present a game-theory model of hierarchy formation in which individuals adjust their aggressive behaviour towards other group members through reinforcement learning. Individual traits such as the tendency to generalise learning between interactions with different individuals, the rate of learning, and the initial tendency to be aggressive are genetically determined and can be tuned by evolution. I find that evolution favours individuals with high social competence, making use of individual recognition, bystander learning and, to a limited extent, generalising learned behaviour between opponents when adjusting their behaviour towards other group members. The results are in good agreement with experimental data, for instance in finding weaker winner effects compared to loser effects.

Dynamic neurogenomic responses to social interactions and dominance outcomes in female paper wasps

“…Consistent with the previous analyses (Fig 4), we find that there are more extreme changes in losers compared to winners, shown by the greater spread along the y-axis (Fig 5). Interestingly, this observation fits with theoretical results that loser-effects should be stronger than winner-effects 68 .…”

“…Over a span of 4 hours the initial similarity between winners and losers decreases, as loser gene expression patterns show larger shifts consistent with theoretical predictions of larger loser effects compared to winner effects 68 . The most dramatic shifts in expression over the course of four hours in losers are due to a mixture of increasing or decreasing expression compared over time (Fig 5b).…”