Suppose you have N variables x_1, ..., x_10 and you want to predict y_1, ..., y_10. You know that each y_i depend on each x_i in a complex, non-linear way.
How many samples would you need to to make sense of distribution? How does number of samples grow with N?