Compression
The world is too complex to reason about directly. Every useful thought requires compression — reducing a sprawling reality into a model small enough to hold in mind.
A map compresses terrain. A price compresses value. A name compresses identity. A scientific law compresses observations. The question is never whether to compress, but how well.
Compression has two failure modes:
Lossy in the wrong places. You kept the details that don’t matter and lost the ones that do. The map shows roads but not elevation; you’re hiking. The financial model tracks revenue but ignores cash flow; you go bankrupt while profitable.
False fidelity. The compression feels complete but isn’t. You think you’ve captured the system; you’ve actually captured your assumptions about it. The model makes predictions that work until they catastrophically don’t.
Good compression is lossy in the right places: it discards what’s irrelevant to the decision at hand while preserving what’s load-bearing.
Vaclav Smil’s writing demonstrates compression by constraint. He doesn’t argue first; he measures. Energy flows, material throughputs, time constants. These physical realities constrain what’s possible, so they’re what survives compression. The vibes get cut; the joules remain.
Kahneman compresses cognition into two systems: fast/intuitive and slow/deliberate. Obviously the brain is more complex than this. But the compression captures something real — the phenomenology of thinking, the characteristic failure modes, the conditions under which each system dominates. It’s useful because it’s usably wrong.
Einstein’s thought experiments compress physics into scenarios a human can simulate. The elevator in free fall. The train with lightning strikes. The riding of a light beam. These aren’t the math; they’re compressions of the math that preserve its implications.
The craft of compression is knowing what to preserve.
For decisions: preserve the variables that change the outcome. Everything else is noise.
For prediction: preserve the mechanisms that generate the pattern. Surface features are misleading.
For teaching: preserve the structure that transfers. Details can be filled in later; the scaffold can’t.
For communication: preserve what survives retelling. If someone can’t repeat your point at dinner, you haven’t compressed it — you’ve just shrunk it.
Compression interacts with expertise in counterintuitive ways.
Experts see more detail than novices, but they also compress more aggressively. What looks like “intuition” is often highly compressed pattern recognition — seeing a chess position as a familiar structure rather than as thirty-two pieces. The compression happens below conscious awareness.
This is why experts struggle to teach. They’ve compressed so thoroughly that they’ve lost access to the uncompressed version. The beginner’s confusion seems inexplicable because the expert has forgotten what it was like to hold all the pieces separately.
The best teachers can decompress on demand. They hold both versions: the compressed form they think with, and an expanded form they can walk others through.
Compression isn’t simplification. Simplification removes; compression re-encodes.
A jpeg isn’t a simpler image — it’s the same image stored differently, with predictable degradation in specific places. A good mental model isn’t a simpler version of reality — it’s a different encoding that preserves decision-relevant structure while reducing cognitive load.
The highest compliment to a compression: “I never saw it that way before, but now I can’t unsee it.”
Related: [[explanatory-writing]], [[chunking]], [[models]], [[legibility]], [[constraints]], [[cognitive-handholds]]