Why would I care if the classification into disease vs polymorphism was automatic or not? Wouldn't the probability of the model being correct all that I need?

I am looking at whole exome sequence data in a sample of affected subjects with a rare disease. The sequences were compared to GRCH38.p2 107 and variant files were derived and then run through annovar. I can see that I have two variables related to MutationTaster: MutationTaster_score and MutationTaster_pred.

the score is the probability of the model being correct from 0 to 1
the prediction classifies as four possible predictions:

1. "A" ("disease_causing_automatic")
2. "D" ("disease_causing")
3. "N" ("polymorphism")
4. "P" ("polymorphism_automatic")

I read http://www.mutationtaster.org/info/documentation.html but am none the wiser.