Risk-sensitivity: crossroads for theories of decision-making

Most actions result in one of a set of possible outcomes. To understand how this uncertainty, or risk, affects animals' decision-making some researchers take a normative approach, asking how an animal should respond to risk if it is maximizing its fitness. Others focus on predicting responses to risk by generalizing from regularities in behavioural data, without reference to cognitive processes. Yet others infer cognitive processes from observed behaviour and ask what actions are predicted when these processes interact with risk. The normative approach (Risk-sensitivity Theory; RST) is unique in predicting a shift in a subject's response to risk as a function of its resource budget, but the predictions of this theory are not yet widely confirmed. In fact, evidence suggests a strong bias towards risk-proneness when delay to reward is risky and risk-aversion when amount of reward is risky, a pattern not readily explained by RST. Extensions of learning theory and of Scalar Expectancy Theory provide process-based explanations for these findings but do not handle preference shifts or provide evolutionary justification for the processes assumed. In this review we defend the view that risk-sensitivity must be studied with theoretical plurality.