Linking Motor Learning to Function Approximation: Learning in an Unlearnable Force Field