It helps to realize that "human reality" is nondeterministic in the sense that we lack sufficient computing power and modelling capability to work with it. It's not random, it's AI-complete. We literally can't handle it formally, and that's why we need to defer to the only "general AI" we have - or as its ordinarily called, human judgement. And human judgement is not random - it's taking into account vast amounts of data, like ethics/morality and social context, which for now escape our attempts at formalizing them.

And the concept of case law / precedent is an attempt to synthesize a point of view from multiple different judges. “Training data,” if you will.