The problem of survival from an algorithmic point of view

摘要

Our goal is to go deeper into the many writings on Behavior-Based Artificial Intelligence [Meyer et al., From Animals to Animats, MIT Press, 1992] and to understand the interest—rather than the mechanisms—of learning. Our intention is to study the complexity of the behavior of living beings from a theoretical point of view. To do so, we introduce formal environments that model the survival issue. Then we prove in this formal context that, many times, the extra cost imposed by the conservation of information, even if it is relevant, is greater than the benefit of knowing it. Consequently, in order to survive in our abstract worlds, one must manage his knowledge in a way that fits the evolution of the environment. Furthermore, physiological observations corroborate these purely theoretical results. Thus, we use these results to design a parallel system in which each module manages its knowledge in a specific way. This enables us to obtain a virtual creature whose behavior evokes that of a biological hen.