Machine Learning (ML) is a sub-area in Artificial Intelligence (AI) that aims at training the computer in order to preform specific tasks. The use of ML became almost compulsory with the appearance of big data in social, technological and scientific context. Great successes of ML include automatic translation and image recognition. The application of AI and ML in physics has started both in the classical and quantum level and several possible applications have been discussed. In the present talk we will give an introduction to ML describing also general applications such as speech recognition and focus on specific examples involving neural networks. Subsequently we will discuss how these ideas can be applied in physics, in phenomena emerging in complex systems. We pick two extreme cases involving chimera states, i.e. states of spatial coexistence of order and stochasticity as well as branching, i.e. generation of random caustic surfaces in wave propagation. We use a variety of ML methods (feed forward networks, reservoir computing, LSTMs) properly augmented for the specific problems, and show that we can have good machine predictability for these systems. We close by discussing possible applications of these methods in physics.