The reality is messy. Sure, we can think in terms of feedback loops and interactive modules and it can be helpful sometimes. We just have to be aware we are imposing our puny models onto the infinitely complex beast.
The model will break down at some point. Recognizing the correct model and extrapolating from it is not useful unless we know the limit at which it breaks down.
It is also hard to be objective about our mental models, we tend to apply all kinds of favoritisms and observation biases when evaluating them. In machine learning (google maps) those biases are still present, but at least they are not our personal biases so they are easier to spot and correct for.
Sure enough, models are problematic for various reasons, a few of which you have enlisted.
It's just an entirely different conversation to have.
Also, many models can be used simultaneously and overlap in the analysis of a particular situation. It isn't necessary to Have It All Figured Out and keep it all in memory simultaneously, just to have a set of methods to draw quick conclusions. Ensemble methods tend to work best IME.
If the partner doesn't like you, you can say all the right, do all the right things, etc and still be passed-over.
Then another partner comes in and does all the wrong things. Still he gets not passed-over.
There's chemistry involved in relationships as well. From smell to much more subtle things. It's a function with infinite variables. We can create models that help us understand society, economy, etc. But these are social sciences, the outcome is affected by the observer ;-)
-of things you could say or do,
there are many right things,
many wrong ones too.
Only some are true.
Who said that line?