Yes, and the fact that a deck of cards is finite is why I can use precise numerical probabilities to describe it, but not with MBTI; however, the principle of increased predictability is still the same, and still based entirely on the concept of generalizations.
As for your next paragraph, well, I think you've just restated my point, and at this point I suspect that you may misunderstand the definition of "generalization":
1. the act or process of generalizing.
2. a result of this process; a general statement, idea, or principle.
A general statement, idea, or principle.
A generalization doesn't have
to describe ALL members of a given group or class in order to have validity. As long as it describes a majority of them a majority of the time, it's useful.
For another example, here's a generalization: Black people are better at basketball than white people.
Of course, there are certainly great white basketball players. There are probably even more black people who suck at basketball, but neither of these facts reduces the value of said generalization. Obviously examples abound of white people who play basketball better than black people, and yet the generalization itself still rings true. It's a question of averages, not of applying exact or rigid labels to every individual member of a particular group. This is the very definition of a generalization.