Internal model representations that encode what tasks do, allowing comparison of task similarity and prediction of learning trajectories.