From seismic we not only need to obtain the structure that could contain hydrocarbons, but also the rock properties so we can decide on whether we are dealing with reservoir rocks (sandstone, carbonates, even shales), sealing rocks (shales, salt) or source rocks (shales, coals). To know what type of rock is present is important, but also what its porosity is and whether it is fractured, as that is important for permeability. To obtain accurate information on the rock properties we need, in principle, to have applied two-way elastic wave imaging. Considering elastic propagation, which includes mode conversion, is necessary when we analyze the (pre-stack) amplitude variation with angle of incidence (AVA). From quantitative analysis of pre-stack seismic data, elastic properties of the reservoir will be derived. But these need to be translated into rock properties relevant for production, that is porosity and fluid saturations. That means that a rock-physics model need to be chosen. For clastic reservoirs that is relatively easy, for carbonate reservoirs it is much more non-unique. Machine Learning, which is part of Artificial Intelligence is applied more and more in all domains of the geosciences, including reservoir characterization. Therefore, I have included applications for classification and clustering of seismic reservoirs, using open-source software.