Harold Jeffrey’s introduced the notion of small-worlds vs large-worlds. Small-worlds have only known unknowns, not unknowns unknowns. Examples of small-worlds include video/board games, poker, and sports. We may not know what will happen, but we know everything that can happen.
Statistics operates in small-worlds. We place distributional assumptions on the observations and then infer the unknown parameters (but we know those parameters exist!). This is statistical inference.
In large worlds, i.e., worlds with unknown unknowns, statistics stops to work. Things can happen that we can’t know a priori that may affect our inferences. This is why inference in the realm of, e.g., geopolitics is extremely precarious: either you have to simplify the world so substantially that your model becomes useless, or you have to try and model very chaotic events which is dubious.