Opponent assessment and conflict resolution through mutual motor coordination

Abstract Evolutionary game theory has been the theoretical tool of choice for analysis of social interactions in general and aggressive interaction in particular. Many important game theory models of aggression are formulated based on assumptions about short timescale behavior but are tested trough measuring long timescale behavior. The resulting methodology often yields ambiguous conclusions, because very different short timescale behaviors can generate similar effects on the long timescale. Here we circumvent problems with traditional methods by developing a machine learning pipeline that measures and classifies the behavior of individual unmarked animals on a sub-second timescale. The resultant dataset allows us to directly measure activity correlations between the attacker and the defender and to identify the cumulative assessment model as an appropriate description of zebrafish, Danio rerio, aggression without reference to covariation between resource holding potential and fight duration. Further analysis of phase specific movement rules unexpectedly revealed display-like characteristics in an attack behavior. A measurement of differences in dynamics between pre-and post resolution attacks suggested that highly energetically costly resolution phase attacks occur because the winner is trying to increase its relative dominance over the loser. Overall, our data pipeline and associated framework can help researchers to develop improved descriptions of strategic behavior which span the full timescale from life history decisions down to the policies used for rapid motor coordination during individual behavioral acts.