Neural representations of expected reward and risk during gambling

Abstract

Organisms continuously monitor the stimuli they encounter and the outcome of their actions. To survive in an uncertain world they aim for rewards and try to avoid punishments. Research in neuroscience, ecology, and economics implies that organisms base their decisions in uncertain situations on expected rewards and risk. Neuroscience focuses on reward prediction learning based on reward prediction errors. In contrast, economic studies emphasize risk in addition to expected reward. We used functional imaging in humans during gambling tasks to understand how the brain represents expected reward and risk. We find that brain activity in subcortical dopaminoceptive structures can be separated, both spatially and temporally, into signals that correlate with (mathematical) expectation of reward, and with reward variance (risk) – two fundamental parameters in financial decision theory. Our results suggest that the primary function of the dopaminergic system extends beyond its established role in learning, motivation, and salience: it signals different aspects of upcoming stochastic rewards – expected reward and risk. Based on financial decision theory we then hypothesized neural representations of prediction risk and prediction risk errors. We find that the insula represents both. In analogy with reward representations in subcortical structures, the signals are spatially and temporally differentiated. These findings expand our understanding of the neural basis of decision making under uncertainty by adding prediction risk estimation. Finally, we investigated where and how expected reward and risk are combined into the neural representation of a gamble’s overall value. Using canonical correlation analysis, we find a new predictor that – contrary to expected utility theory – adds risk to expected reward. This sum may define a metric of conflict or attention. This metric significantly correlates with activation in the anterior cingulate cortex a structure associated with conflict monitoring. Drawing on financial theories, we show how the brain represents expected reward and risk. Our results suggest that the earlier understanding of decision making under uncertainty needs to be expanded to include (prediction) risk as measured by variance as well as prediction risk errors. Such integration has far-reaching implications, in particular for pathological decision making.