Linearly separable problems are easy to solve
Linearly separable problems can be solved with relatively crude classifiers, trained with very few examples. This is because linearly separable spaces can be divided by a single hyperplane, and the number of points of data required to categorize regions around this hyperplane scales with the dimensionality of the hyperplane. So a binary classifier only requires two data points to train if the representation linearly separates the downstream task. This is why, With the right representation, judgement is cheap