The initially selected outliers can prevent the model from converging to the desired predictive model in the active learning process.