Is Complexity just an illusion?
(I am dictating this while jogging around a pretty lake)
Most of what we call complexity probably isn't a property of reality itself. it's a property of the language we're using to describe it. if I point at a dog and say that's a golden retriever, I've compressed a ridiculous amount of information into two words. if you don't have that concept, you need a much longer description of the exact same dog. the dog didn't become more complicated. your map did.
I think expertise works in basically the same way. a chess beginner sees a board full of pieces and an absurd number of possible interactions. a grandmaster sees a few familiar structures and a handful of moves worth thinking about. they aren't seeing less. they just have better compression. years of experience have turned recurring patterns into concepts, so they don't need to reason about every detail from scratch. which is kinda what an abstraction is. compression.
and once you think about it this way, learning starts to look different. we usually treat learning like we're trying to stuff more information into our heads. read more, memorize more, collect more facts. but learning something complicated is probably closer to representation engineering in your own head. you're trying to acquire concepts that turn a lot of details into a few useful handles. without those handles, everything feels like a special case. once you have them, entire groups of problems suddenly become the same problem.
This also changed how I think about "simple explanations." people like the idea that the simplest explanation is usually the best one, but I don't think shortness is actually what we care about. we care about explanations that keep working. ones that survive new information and don't assume more than the evidence gives us. a short explanation can be simple because it found the right abstraction, or it can be simple because it quietly threw away everything inconvenient. those are very different things.
and weirdly, this might explain why simplicity does seem to correlate with truth in mature fields. when we discover some structure that keeps showing up, eventually we give it a name. something that once required a paragraph can become a word or two. so I think we sometimes get the causality backwards. it's not necessarily that simple ideas generalize. it's that useful, general structures eventually get compressed into language, and once that happens they look simple.
there's a pretty practical lesson here for learning. when you're entering a new field, one of the highest leverage things you can do is steal its primitives first. figure out which concepts the people who are actually good at it use constantly. without those concepts, you're going to spend a lot of time rediscovering the same structures yourself. then you need enough exposure for those concepts to stop being definitions and start becoming things you actually see.
this is why examples matter, but probably not in the way we usually think. you don't need fifty examples because more examples are inherently better. you need enough of them for the pattern to become obvious. a programmer eventually stops seeing every weird state bug as a unique mystery and starts thinking race condition? shared state? stale data? the search space gets smaller because experience has given them better categories.
but recognition alone isn't enough. the abstraction has to predict something. if you can only read an explanation and think yeah, that makes sense, you might just be recognizing a good story. actually understanding something means the model in your head constrains what you expect to happen next. prediction is where you find out whether your compression preserved the important parts.
another thing experts seem good at is not committing too early. when something breaks, it's tempting to immediately find the cleanest explanation and start digging. but good reasoning usually keeps several possibilities alive until the evidence starts killing them. what does this actually rule out? is a much better question than what story fits this? you're still compressing the search space, just without pretending you know more than you do.
and over time, you start building vocabulary of your own. if you keep encountering the same cluster of details, you eventually give it a name, even if only in your head. now you can think with that whole structure as one unit. and this compounds. every good abstraction makes the next thing a little cheaper to understand.
So is complexity an illusion? idk, maybe partially. obviously some things really are complicated. some systems are genuinely hard to predict and some problems really do have enormous search spaces. but I think a surprising amount of what we call complexity is really the feeling of not having the right representation yet.
once you find it, the problem can suddenly feel obvious. not because reality changed, or because you removed the difficult parts, but because you stopped translating everything from scratch every time you looked at it.
which is why I increasingly think intelligence has a lot to do with finding the right level of abstraction. compress enough that the structure becomes visible, but not so much that you throw away the details that matter.
the territory is what it is. a lot of the complexity tax gets paid somewhere between the territory and the symbols we're using to represent it.
sometimes you don't need to think harder.
you need a better map.