← All writing Final · Mar 2026 · Updated Sep 2026

Why Explanation Comes in Layers

Against reductionism: higher-level explanations are often the only forms in which reality becomes intelligible to finite minds. Compression and connectedness are why the explanatory hierarchy cannot be collapsed.

Introduction

Why does a pot of boiling water have more than one explanation? You can explain it by molecular motion. You can explain it by vapor pressure and the air pressure above the pot. Or you can say that Buddy wanted ramen and turned on the stove. All three are true, and none cancels the others.

The question matters because our answer shapes how we think about knowledge. One common answer is that every higher-level explanation is shorthand for a deeper one. On that view we talk about agents, goals, organisms, and markets only for convenience, and real understanding always lies further down.

I think that view is mistaken. Lower-level mechanisms are real, of course. The mistake is assuming that lower means more explanatory. A more fundamental description is not automatically a more illuminating one.

My conjecture has two parts. The first is compression: finite minds cannot reason over reality in its full detail, so explanation must throw away most of that detail while keeping the structure that matters. Higher-level objects like pots, colonies, goals, and markets are what successful compression produces. The second is connectedness: as descriptions go lower, they do not become simpler and more separable. They become more dependent on their surroundings, until at the bottom the context implicates everything. Together, these two reasons explain why the layers of explanation can’t be collapsed into the lowest one.

1. The Temptation of Reductionism

Reductionism, as I use the term, is not the claim that things have parts, or that lower-level processes are real. Both are true. It is the view that the best explanation of anything must come from analyzing it into components, so that higher-level explanations are less fundamental or less real.

The view has real appeal. If water is made of molecules, molecular description seems closer to the truth than talk of pots and dinner. If a colony is made of ants, the real explanation seems to be the ants. The higher-level story looks like a verbal summary of what is really happening underneath.

But this quietly assumes that breaking a thing into parts reveals the structure that explains it. That needs to be argued, not assumed. The structure that matters may live in patterns among the parts, in history, in boundary conditions, or in how a system interacts with its environment. Decomposition can keep every causal ingredient and still destroy the explanatory form.

None of this is a rejection of causal explanation. Tracing causes is often how we explain things. But causation operates at many levels. Molecules cause phase transitions, intentions cause actions, markets cause prices. The question is not whether causation is real at each level. It is which level of causal description makes a given phenomenon intelligible. Reductionism assumes the answer is always the bottom. I will argue that it often isn’t.

2. Compression

Compression is representing a phenomenon with less information while preserving what matters for explanation. It is not mere abbreviation. It selects structure.

Finite minds need it because a complete microphysical description of even a mundane event would be hopelessly vast. But size is not the deepest problem. A microphysical description does not interpret itself. It does not mark which cluster of particles is a pot, which pattern is a person, which sequence of events counts as cooking, or which behavior expresses a goal. Those are not written on the microstate. They are abstractions over it.

So knowledge does not grow simply by accumulating detail. A complete inventory is not yet an explanation. It becomes one only when structure is selected and organized into a form a mind can use. We do not start with reality in its full detail and then choose to simplify it. We start with no understanding at all, and we reach reality through abstraction. Compression is not a stylistic aid added after understanding. It is part of what understanding is.

This is also why higher-level objects are not illusions produced by laziness. A higher-level object earns its place when it captures stable structure better than an ocean of lower-level detail does.

3. Connectedness

Compression explains why we need abstractions, but not why they can’t be traded for lower-level ones given enough effort. For that we need a second idea: connectedness, the degree to which a phenomenon depends for its intelligibility on relations to other things, conditions, histories, or environments.

At higher levels, things look fairly separable: the pot, the stove, the ant, the kitchen. That separability is useful and often deeply explanatory. But it is an achievement of abstraction, not a basic feature of reality.

Go lower and the dependencies multiply. To describe a molecule in boiling water, you must specify temperature and pressure, which are collective properties of enormous numbers of molecules. You must specify the container, the atmosphere, the heat source, the gravitational field. At the quantum level, particles can be entangled so that no particle’s state can be described on its own, whatever the distance between them. In general relativity, the geometry of spacetime in one region depends on mass and energy elsewhere. Each attempt to isolate a local description pulls in more of the world.

Here is the second conjecture. Following this descent to its end, you do not arrive at simple, self-sufficient units. You arrive at a description in which nothing can be specified without specifying more, until the context implicates the whole. The lower limit of reductionism is not simplicity. It is holism.

One clarification keeps this from being misread. The fundamental laws of physics may well be simple. The claim is about describing a particular event at that level. A simple law applied to one pot of water still needs the state of everything that pot is connected to.

That is what higher-level explanation does for us. Each layer draws a boundary around what is relevant and treats everything outside it as fixed background. I call this managed disconnection: the deliberate severing of relations that don’t matter to the phenomenon at hand. It is what makes tractable explanation possible in a deeply connected world. Lower-level description does not only add detail. It undoes the managed disconnection and restores the dependencies the higher layer had screened off.

4. From Boiling Water to Wanting Ramen

Return to the pot. One explanation says the water boils because its molecules have enough energy to escape into the gas phase. Another says it boils when its vapor pressure equals the surrounding atmospheric pressure. A third says Buddy wanted ramen and turned on the stove.

These are not rivals. They sit at different levels of emergence, a term I take from David Deutsch: a set of phenomena that can be explained well in terms of each other without being analyzed into their constituents. (This is different from the complexity-science use of “emergence” for bottom-up causal patterns; the two ideas are independent, and confusing them is an easy way to misread this essay.) Vapor pressure, boiling point, and atmospheric pressure explain one another without forcing us down to atoms. Wanting, planning, cooking, and eating form another such domain. Neither is made illusory by the existence of molecules.

What would it take to derive “Buddy wanted ramen” from a pure microphysical description? Far more than a snapshot of matter. You would need a long stretch of time, a way to track the same entities across it, concepts for persons, beliefs, intentions, and artifacts, and a way to tell relevant structure from meaningless detail. In other words, you would need much of the world and its history already organized into higher-level concepts. The lower-level account can’t produce this explanation without first borrowing the higher level.

The higher-level account is not a lapse in rigor. It compresses what is central and ignores oceans of irrelevant variation. That is what rigor looks like when explanation is the goal.

5. Why Higher-Level Explanations Are Real

A natural objection arises here. If layers exist because of how minds must engage with reality, aren’t they artifacts of cognition rather than features of the world?

Part of the answer is yes: the need for layers comes from minds. An intellect with unlimited memory and processing might track every particle and predict every event without them. But the structure a good layer captures does not come from minds. Vapor pressure does predict boiling, whoever is watching. Intentions do organize behavior. A layer is useful only because there is a real regularity there to compress.

And minds are themselves part of reality. The layers they need are real features of how reality can be understood: not fundamental to the substrate, but not invented or arbitrary either. That is a more modest claim than saying higher-level objects exist apart from all minds, and a more defensible one.

What makes a higher-level explanation real in this sense is explanatory reach: it illuminates many related phenomena, answers why-questions, and tells us what would have happened otherwise, without rebuilding the whole substrate each time. That reach is a genuine property of the explanation, and it is grounded in genuine structure.

That is why “Buddy wanted ramen” is a valid explanation of boiling water. It doesn’t deny physics. It explains at a level where agency, preference, and intention organize the event more economically and more informatively than a particle-by-particle history could. Rejecting it for not being microphysical is not rigor. It is a confusion about what explanation is for.

A colony is not unreal because it is made of ants. A market is not unreal because it is made of transactions. A self is not unreal because it is implemented in matter.

6. The Reductionist Mistake

We can now state the mistake plainly. Reductionism expects that breaking a phenomenon into components will bottom out in simple, separable units that explain everything above them. The connectedness argument says it won’t. Descriptions of particular events, pushed downward, lead toward holism, where nothing can be specified without specifying more. The reductionist pursues simplicity and finds its opposite.

That makes reductionism, as a method of explanation, self-undermining. Decomposition works well at higher levels because managed disconnection has already done the hard part. We can isolate a pot from a kitchen, an ant from a colony, or a buyer from a market because abstraction has made those isolations tractable. The success of decomposition up there is not evidence that reality is made of separable units. It is evidence that higher-level abstraction has done its job. Carried downward, into domains where connectedness has not been screened off, decomposition strips away the very relations the phenomenon depends on.

Ant colonies make this vivid. Individual ants follow local rules, many mediated by pheromones. But the colony finds food, builds bridges, defends itself, and adapts in ways no ant plans or understands. The colony-level pattern has explanatory autonomy. Insisting that only the ant is explanatorily real means missing the phenomenon entirely.

The same mistake shows up in discussions of mind. Predictive processing may describe something real about the lower-level architecture of cognition. It does not follow that creative explanation, criticism, or agency are best explained by it. A lower-level mechanism can be real without supplying the best explanation at a higher level.

Conclusion

So why does explanation come in layers? Because finite minds cannot reason over uncompressed reality, and because going deeper does not lead to simpler, more separable units. It leads toward holism. A complete lower-level inventory is not yet an explanation. It becomes one only through compression and managed disconnection, which keep the structure that matters and set the rest aside.

That is why higher-level explanations are not mere conveniences. For finite knowers they are often the only forms in which reality becomes intelligible. They don’t float free of lower-level reality, but they aren’t dissolved by it either.

Components matter, and mechanisms matter. But explanation is not a contest to find the smallest pieces. It is the search for the right managed disconnection in a deeply connected world. The hierarchy of explanation is not an embarrassment to be eliminated. It is one of the ways truth becomes knowable.