An opaque predictive model can be useful in the clinic, if it is revalidated, and possibly retrained, everywhere it is used.

Better understanding of a risk factor, however, can be communicated instantly, for free, around the world.

Model interpretability is important.