Organizations operate in task environments where the same problem situation hardly repeats itself over time. In this context, the decision-makers face a distinct learning challenge which may be referred to as the problem of generalization: how to generalize from past experiences to deal with new situations that are distinct but similar to the past? The first part of the dissertation presents a computational model that explores some aspects of the generalization problem: when an organization experiences a largely idiosyncratic series of events, at what level of granularity should these events, and the associated actions and outcomes, be encoded? How does generalizing from experience impact the wisdom of future choices, and what are the boundary conditions or factors that might mitigate the degree of desired generalization? The model incorporates how characteristics of opportunities might be encoded so that experiential learning is possible even when the organization’s experience is a series of unique events. The results highlight the power of learning through generalization in a world of novelty and the features of the problem environment that reduce this “power.” The second half of the dissertation presents an experimental investigation of a set of interrelated questions on learning and decision-making: when presented with a stream of novel opportunities that generates risky outcomes, how do human decision-makers infer the value of the opportunities, and how do they decide to capture the opportunities based on their evaluations? The study provides novel findings about generalization and risk-taking using a web-based experiment. When holding positive beliefs, decision-makers commit to the opportunities almost irrespective of the magnitude of the belief. In contrast, when having negative beliefs, decision-makers execute the reward opportunities to a noticeable degree, in proportion to the belief magnitude. Such findings suggest that decision-makers take into consideration the generalizability of their risky choice outcomes when they have a limited understanding of the task environment. Overall, this dissertation offers new theoretical insights into the role of generalization in the process of learning from past experiences.