Reservoir computing (RC) is a machine learning (ML) framework that has gained attention in recent years as the interest in alternative computing paradigms has grown. RC allows the utilization of physical systems to solve ML tasks. We demonstrate the use of the nonlinear photorefractive reservoir computer and perform tasks requiring both nonlinearity and memory, such as chaotic time series prediction. Changing the photorefractive response by adjusting the applied field and laser power controls the characteristics of the reservoir. Optimizing the characteristics of the reservoir for performing a 10-step Mackey–Glass (MG) time series prediction, we achieve a mean square error (MSE) of 5x10−4.