Essays on the exploration-exploitation dilemma.

Summary Over the last two decades, numerous empirical studies have highlighted the differences in individual choices when they are made on the basis of a description of the space of states of the world and their underlying probabilities (decision from description), and when they result from experimentation with this space via sampling (decision from experience). Indeed, in the first case, the individual has a perfect knowledge of the space of states of the world. In contrast, in the second case, the individual does not know in advance all the possible alternatives and/or their probabilities of occurrence. This divergence between the individual choices observed in these two configurations is commonly referred to as the description/experience gap. The phenomenon of undersearch is one of the causes put forward in the literature to explain this gap. Given the importance of the issue of choice under uncertainty in economics, the search process deserves further attention. This thesis aims to contribute to the theoretical and experimental literature that studies this process and the exploration-exploitation dilemma that is inherent to it, both at the individual and collective level. The thesis is composed of 3 essays combining theoretical modeling, multi-agent modeling, evolutionary algorithms and laboratory experiments. The first chapter of this thesis examines the determinants of the search process in the context of an individual optimal stopping problem. The results show that this process depends to a large extent on the degree of certainty of the information and that regret and anticipation play an important role. The second chapter studies the information sharing behavior in the context of a competitive collective search using multi-agent simulations and evolutionary algorithms. It highlights the existence of individual benefits to sharing, even when others do not share in return, provided that two mechanisms are present: imitation with a certain level of innovation and local visibility. The third chapter tests and validates experimentally these results and underlines the crucial role of learning.