In this paper, an overview of techniques to estimate the visual complexity of the different parts of a scene from a given point of view, in order to be used in radiosity computation, are presented. This information is then used to improve the Monte Carlo based radiosity calculation by distributing the rays leaving a patch according to the estimated visual complexity of each part of the scene. All the proposed techniques to estimate the visual complexity are based on artificial intelligence methods, especially heuristic search, and divide the scene in regions via a hemisphere associated to each patch and divided in spherical triangles. The visual complexity in a region is considered constant.