Multilevel selection in causal models: the multiplicative nature of evolutionary and probabilistic selection processes as the general driver for the irreversibility emergence of cooperation and specialization.

To explain why major evolutionary transitions are so common in the history of life we need to find the causes that systematically generate the irreversible emergence of cooperation and specialization. For this purpose it is necessary to consider selection at both the individual and group level. The co-author of the concept of major evolutionary transitions (Szathmáry) recently proposed to analyze the evolution of populations subject to multilevel selection by means of Bayesian hierarchical models, making use of the isomorphism between evolutionary theory and Bayesian inference. However, the proposal remains open. In this paper we specify a probabilistic causal model, in which individuals are affected by the environment and by the social behaviors of cooperation and defection of their context. Even under this minimal set of hypotheses, where we consider unconditionally cooperative individuals who generate a common good that can be exploited by defecting individuals without receiving some kind of punishment in return (e.g. end of cooperation), probabilistic inference shows that cooperative individuals are favored by multilevel selection. In addition, we show that as soon as cooperation emerges, an advantage in favor of specialist strategies appears. Since the specialist strategies are individually poorly adapted to the environment, an irreversibility of the evolutionary transition is created. The reason why an advantage in favor of cooperation and specialization arises in our simple causal model is due to the multiplicative (non-ergodic) nature of probability theory and its isomorphism with evolutionary theory. https://github.com/glandfried/TrueSkillThroughTime/releases/download/doc/landfried-learning.pdfhttps://github.com/glandfried/TrueSkillThroughTime/releases/download/doc/landfried-learning.pd