Even the best maps (any abstraction of reality, including descriptions, theories, models, etc) are imperfect and often inaccurate. This is because they are a reduction of reality.
Maps can also be a snapshot of a point in time, thus representing something that doesn't exist anymore.
Maps and territories
Reality is complex. A reduction or abstraction of it is necessary for humans to understand it.
Abstraction is so necessary for us that we will rather use an incorrect abstraction than none.
Every map has its limitations because:
1) in order for information to be reduced, some will inevitably be lost.
2) relying on a map means we have no way of knowing if it is correct or not
3) maps are made and read with interpretation, a process that can cause major errors.
The human brain often takes great shortcuts to make sense of its surroundings. That means in the process of understanding reality, we tend to go for the easiest way to do so and automatically apply it even in situations where it doesn't hold.
An example could be applying something that worked before in another situation to a new one. The old map may not work in the new one. "The terrain changed, but the old idea stuck."
An example is the [[VAR theory]], or value-at-risk, used in the banking community to find quantify risk. The problem is models use a finite set of parameters, while in reality, there are infinite risks, as Nassim Taleb says.
We should endeavour to build systems that reduce model error. A model like VAR got people trusting it so much that the financial world used it to optimize, not guide. The world usually does not operate necessarily as the model predicts.
So what do we do?
Use informed common sense than rely completely on a model. Models can hide risks, it is after all just a model.
Falling back on simpler heuristics is also an option, like how Warren Buffet has never used a computer model in his life yet manages to grow an institution so successfully, solidly built with backup systems and margins of safety at multiple levels. "Extra cash, not extra leverage."
Instead of optimizing to a model, accept that there are limitations to seeing the future.
The trade-off is that in the short run, using an optimized model will work much better than one that isn't.
Don't confuse reality with useful models. Models are simply one way of interpreting a set of information.
Reality is messy, there aren't neatly set parameters like in a map. The map itself is flawed because it has to be abstracted.
Understand where the map is useful and where it isn't. Use the map when you understand its limitations and its usefulness.