The overarching goal of the project is to develop AI surrogates for molecular dynamics (MD). At least in the beginning, my work will focus on implicit transfer operators (ITOs). We intend to create working examples of ITOs that can be provided as supplemental material in a survey/tutorial paper. We also intend to make novel methodological contributions to the field. We are particularly interested in describing available and developing new evaluation metrics. Both the accuracy of the models, as well as the speed advantages when compared to state of the art approaches from classical statistics should be evaluated. We are also interested in finding improvements to the model or the trainig routine, as may be be motivated via scientific principles from physics and chemistry. We have started investigating the use of coarse-grained data for training ITOs, and will investigate the properties of training ITOs with masked data, which is one form of coarse graining. Potential applications may include the inclusion of experimental data in the training, or identifying the most salient features in molecular data. We expect our research to advance the state of the art in generative models for molecular dynamics. In particular, we hope to produce work that will make ITOs more accessible to newcomers and we aim to develop new methodology for both, model design and model evaluation.