Discovery is seeing the particular as an instance of the general
Type: note · Status: seedling
A common framing in Zettelkasten-adjacent writing contrasts "topical" linking (filing by category) with "semantic" or "mechanistic" linking (connecting by shared deep structure). The claim is that topical linking produces no surprise — you open the economics drawer and find economics — while mechanism-based linking engineers serendipity.
This framing is useful but the dichotomy is false. Topic and mechanism are not different kinds of link. They are the same cognitive operation — recognizing similarity — applied at different depths of abstraction. Topic is shallow recognition (these are both about economics). Mechanism is deeper (these both describe systems that degrade under structural overload). But once you see the shared mechanism, linking by it is exactly as easy as linking by topic. The bottleneck is never the linking. The bottleneck is the recognition.
This matters because controlled disorder engineers serendipity through semantic rather than topical linking gets the prescription slightly wrong. Luhmann's actual move wasn't "link by mechanism instead of topic." It was link by judgment instead of category. A filing cabinet uses pre-determined classification. Luhmann's Zettelkasten replaces that with individual decisions: THIS note connects to THAT note for THIS specific reason. The reason might be topical, mechanistic, analogical, or contrastive — what matters is that a human judged the connection worth making and articulated why.
The dual structure of discovery
Discovery — whether in mathematics, science, or a knowledge base — has a dual structure that can't be decomposed:
- You posit something general that didn't previously exist as a named concept
- You simultaneously recognize that things you already knew are instances of it
The general doesn't exist until you see the particular as an instance of it. And the particular wasn't legible as an instance until you posited the general. They co-arise.
Mathematical lemma extraction
A mathematician has Theorem A and Theorem B, apparently unrelated. She notices the proofs share structure — similar moves, similar appeals. She extracts the common structure as a Lemma. The Lemma is a new node in the knowledge graph that didn't exist before. Both theorems now link to it, and the link is both factual (they depend on it) and illuminating (seeing the shared foundation changes understanding of both).
Category theory takes this to its logical extreme — abstracting so aggressively that group theory and topology become instances of the same structure. The abstraction makes the similarity nameable, and once named, recognizing further instances becomes cheap.
Darwin's theory of evolution
Darwin took four observations that were separately unremarkable — variation exists, more offspring are born than survive, traits are heritable, environment exerts pressure — and saw that taken together as axioms, they produce a theorem: populations adapt over time. Each observation was known to breeders, Malthus, naturalists. Nobody was hiding information. The creative act was seeing that these statements from different domains are premises of the same argument.
Crucially, Darwin proposed the model without knowing the mechanism of any of its components. He didn't know what generates variation (mutations), how heredity works (DNA), or the biochemistry of selection pressure. The model operates at a level of abstraction above all those mechanisms. It says: I don't care how variation is produced, only that it exists. Given these abstract properties, adaptation follows necessarily.
This is mathematical in its core. Evolution by natural selection is a theorem about any system satisfying those axioms — which is why the same logic appears in genetic algorithms, immune system antibody selection, market competition, and cultural evolution. The empirical work was necessary not to prove the theorem but to identify the right axioms. The Galapagos finches didn't prove natural selection; they showed Darwin which abstract properties to attend to.
Darwin's move goes beyond lemma extraction: he didn't just find shared structure, he proposed a generative model — an abstract machine that produces the observed phenomena as outputs. The model explains why the similarity exists and predicts where else it should appear.
Fleming's discovery of penicillin
This feels different — a contaminated petri dish, mold killing bacteria. Where's the "extraction"? But other microbiologists had seen mold contamination. Some had probably seen inhibition zones. Fleming's contribution was recognizing that this specific anomaly was an instance of something general — that there exists a category of substances (antimicrobials) that could be extracted, purified, and used therapeutically. The observation was particular; the insight was seeing the particular as an instance of a general category that didn't yet have a name.
Three depths of abstraction in connection
These examples suggest a hierarchy of connection depth:
| Depth | Operation | Example | Power |
|---|---|---|---|
| Shared feature | Name a surface similarity | "Both involve capacity limits" | Descriptive — organizes but doesn't explain |
| Shared structure | Extract a common pattern | The lemma; "both follow: load → graceful degradation → threshold → catastrophic failure" | Structural — reveals form but not cause |
| Generative model | Propose an abstract machine that produces both phenomena | Natural selection; "both are outputs of: systems that optimize for efficiency under normal load sacrifice resilience to overload" | Explanatory — explains why the similarity exists and predicts new instances |
Each level subsumes the previous. A generative model implies shared structure implies shared features. But they're increasingly powerful and increasingly hard to discover.
The discovery problem
The hard problem in knowledge systems is not linking (once you see a connection, articulating it is straightforward) but recognition — seeing that two things share structure at some level of abstraction.
Recognition cost scales with abstraction depth: - Surface similarity (topic): cheap. Embeddings, keywords, filenames get you there. - Structural similarity (pattern, proof technique): expensive. Requires understanding what a note is really about, then comparing that understanding across notes. - Generative similarity (shared causal process): very expensive. Requires not just comparing but inventing the dimension along which the comparison becomes visible.
The mathematical tradition offers a partial solution: develop vocabulary for naming structures. Once a structure has a name (a lemma, a category, an algebraic property), recognizing new instances of it becomes cheap. The naming amortizes the discovery cost. In a knowledge system, this means the highest-value act isn't linking two notes that share a mechanism — it's creating a new note that names the mechanism. Once you have a note titled "Systems degrade when you exceed their structural capacity," linking cognitive load and architecture to it is trivial. The hard creative work was seeing the pattern and giving it a name.
Open Questions
- Can you build practices that push connections from "shared feature" toward "generative model"? The elaborative encoding requirement pushes from bare adjacency to articulated connection. Is there an equivalent push toward deeper abstraction?
- Two distinct discovery problems get conflated: (1) noticing that existing notes share structure (O(N^2) comparison), and (2) inventing the abstraction that makes the similarity visible (genuine theoretical creativity). These probably need different system support. LLMs might help with (1); (2) is less clear.
- The hierarchy (feature → structure → generative model) maps loosely onto Alexander's three levels of connection (concrete → structural → vague/deep). Is this the same axis?
Relevant Notes:
- arscontexta — refines: the "controlled disorder" claim is right about judgment-based linking but the topic-vs-mechanism framing is a false dichotomy
- Notes Without Reasons — extends: the adjacency-vs-connection distinction maps to recognition depth, not link kind
- alexander-patterns-and-knowledge-system-design — parallels: Alexander's three levels of connection may map to the feature/structure/generative-model hierarchy
- stabilisation — analogous spectrum: stabilisation grades (store output → write description → extract code) parallel the abstraction depths here
- information value is observer-relative — grounds: the recognition cost hierarchy maps to computational bounds on structure extraction (epiplexity)
Topics: