Auto-encoders can be used in model-based learning to predict the next state given a state and an action