You are likely correct. I think most researchers would agree, however. The bigger issue is actually learning how to form complex models. People want networks to just learn this implicitly, believing that we would likely impose counterproductive models. Other people simply struggle to incorporate models into the training process.