In this paper, we introduced the temporal and information representation properties of the human mirror neuron system (MNS) that have been revealed by using noninvasive imaging techniques (magnetic resonance imaging (MRI), positron emission tomography (PET), and noninvasive brain stimulation such as transcranial magnetic stimulation (TMS) or electrical stimulation). Next, we discussed the neural networks involved in action understanding and affordance. In this paper, we especially emphasized the role of MNS in prediction and action recognition, and introduced recent brain theories and a neural network model, based on several MNS properties, including the prediction function, the free energy principle, and the minimization of prediction error.