Models are run by specific mathematical equations that have to make assumptions about the state of the atmosphere. These assumptions result in model "biases" over time (i.e. the NW trend -- out in time, models forecast the storm too far to the east. They "see" the atmosphere better as the lead time decreases. As the "see" better, they "correct" their solutions/storm tracks. Usually, this corrects NW, as the modal bias is to model the storm too far to the east). Humans can observe these trends over time, recognize them, and adjust for them in their forecasting. Until we get to a state where the entire atmosphere can be observed and modeled, models will always have certain biases. It's up to us to recognize them and use them to our advantage when forecasting.