It's not really "magic". It's just that interpretation is not really that important sometimes. It's the approach of "Who cares what these coefficients mean, the

It's not really "magic". It's just that interpretation is not really that important sometimes. It's the approach of "Who cares what these coefficients mean, the model performs well on the test set". This is true for most of predictive modelling.