Learning under uncertainty is a common task that people face in their daily life. This process relies on the cognitive ability to adjust behavior to environmental demands. Although the biological underpinnings of those cognitive processes have been extensively studied, there has been little work in formal models seeking to capture the fundamental dynamic of learning under uncertainty. In the present work, we aimed to understand the basic cognitive mechanisms of outcome processing involved in decisions under uncertainty and to evaluate the relevance of previous experiences in enhancing learning processes within such uncertain context. We propose a formal model that emulates the behavior of people playing a well established paradigm (Iowa Gambling Task - IGT) and compare its outcome with a behavioral experiment. We further explored whether it was possible to emulate maladaptive behavior observed in clinical samples by modifying the model parameter which controls the update of expected outcomes distributions. Results showed that the performance of the model resembles the observed participant performance as well as IGT performance by healthy subjects described in the literature. Interestingly, the model converges faster than some subjects on the decks with higher net expected outcome. Furthermore, the modified version of the model replicated the trend observed in clinical samples performing the task. We argue that the basic cognitive component underlying learning under uncertainty can be represented as a differential equation that considers the outcomes of previous decisions for guiding the agent to an adaptive strategy.