Dissertation (Ph.D.) - Indiana University, School of Informatics, Computing, and Engineering, 2019Traditionally, cognitive science has focused on symbol-manipulation tasks in isolated problem domains. Implicit in this approach is the identification of intelligence with high-level, algorithmic, and uniquely human abstract reasoning. Arguably more essential, however, is the coordinated perception and action required of even the simplest organisms, whose survival depends on reacting appropriately and in real time to the constantly changing conditions of a messy and often hostile environment. Embracing this latter view, some modern approaches to cognitive science identify intelligence with adaptive behavior: an intelligent organism is one whose actions support the survival of the individual or species. In this dissertation, I seek to characterize the neural basis of intelligence by investigating the neural correlates of adaptive behavior. This work is performed in the context of Polyworld, an agent-based artificial life model. In Polyworld, a population of agents evolves in a simulated ecology. Each agent is endowed with a rudimentary sense of vision and is capable of a set of simple actions controlled by a neural network. The population’s long-term survival depends on the evolution of adaptive behavior via rewiring of these networks over successive generations. I identify the neural mechanisms underlying adaptive behavior by investigating trends in neural properties, both over evolutionary time and across experimental conditions. Applying the tools of graph theory, dynamical systems theory, and information theory, I analyze Polyworld’s neural networks using metrics with demonstrated relevance in other studies of complex systems. Many properties of interest are amplified over evolutionary time: the emergence of a successful survival strategy is accompanied by trends toward small-world neural structure, critical neural dynamics, and effective neural information processing. These trends are statistically significant when compared to a neutrally evolving null model, suggesting correlations with adaptive behavior. More direct evidence for these properties’ adaptive significance comes from correlations with environmental complexity: as experimental conditions become increasingly difficult, each property is further amplified, fostering the increasingly complex adaptations required for survival.