When prediction fails: a computational complexity view of stress.

Living organisms are increasingly understood as predictive systems that anticipate near-future, fitness-relevant demands and proactively adjust physiology and behavior. Then, stress arises when environmental challenges exceed the range normally anticipated by the organism, eliciting stress responses that push regulatory systems toward or beyond their functional limits. These conditions are often associated with uncertainty, unpredictability and uncontrollability. We focus on the functioning of control systems when they fail to predict the environment and successfully control behavior. We hypothesize that under such conditions feedback controllers become increasingly obsolete and can be dynamically discounted, providing a fitness benefit. Computational complexity of adaptive control is expected to reduce the size of the controller program, the complexity of its output and therefore the complexity of adaptive behavior. We briefly review the literature suggesting that reduced complexity of behavior may serve as an indicator of developing stress. Finally, we outline several approaches and R software packages for the measurement of complexity.