By building some knowledge of physiology and muscle dynamics into a machine learning system, we can make it extrapolate farther from each training point.   That leads to better accuracy and/or it allows the model to be simpler and trained with less data.