Cognitive Cartography

Let’s be honest: most information exploration systems still make us do the filing. We build folders, tags, and categories, then forget our own logic a week later. A single-parent hierarchy is especially awkward when one note belongs to several questions at once.

(Update 2024): Oh great, now we have “AI-driven” note-taking apps that use “smart assistants” to… CATEGORIZE of all things. As if slapping three tags on a book would solve the problem. We’re automating the same habit.

Spatial organization offers another way to orient ourselves. Think of tracing a shape with your fingers in the dark: recognition depends on the sequence of touches, not one isolated point. A knowledge map could likewise retain paths and neighboring ideas rather than showing only the item you searched for. That is a design hypothesis, not proof that every abstract relationship has a natural location.

We use movement to talk about ideas: we circle back, move closer, lose our bearings. People also gesture when explaining size, proximity, or a process. These observations suggest design possibilities, though a gesture or metaphor alone does not establish how the brain represents an abstract concept.

I am interested in three operations. Navigation lets someone move among related ideas while seeing the path they took. Zoom changes the level of detail without dropping the wider context. Landmarks make a space recognizable on a later visit. Together they might help with questions that folders handle poorly, particularly when material belongs to overlapping contexts.

The complementary question is how those relationships form and change. Auto-Associative Workspaces explores that underlying structure; cognitive cartography asks how we might move through it.

A spatial interface may help people orient themselves in a complex information space when its layout reflects relationships they can inspect and remember. We should compare it with a clear hierarchical or list-based interface on actual tasks: finding a known item, returning to an earlier line of inquiry, and discovering a relevant connection. Speed matters, but so does whether someone can explain what the map means after using it.

The possibility of serendipity is harder to measure. Finding more links is not automatically a benefit; the links must matter to the question at hand. A map that looks rich while making claims harder to verify has failed.

Cartographers choose what a map shows and what it suppresses. We face the same decision with information. Position could express a particular relationship, such as topical similarity or chronology, but it cannot express every relationship at once. Color might show source or status; size might show scope. Each choice shapes what a reader is likely to notice.

The most useful move may be semantic zooming. From a distance, a reader sees regions and paths. Up close, they can inspect the source passage, its claims, and its links. The transition has to preserve orientation: if every zoom rearranges the terrain without warning, the map becomes another source of confusion.

A layout can also show a trail through the material. The route someone followed to reach a claim may be as useful as the claim itself, particularly when they return days later or invite someone else into the inquiry.

Visual parameters carry arguments of their own. Put items close together and readers will infer kinship; make one larger and they may assume importance. That inference could be useful when proximity reflects an explicit citation or repeated co-use. It becomes deceptive when the layout only reflects a model’s uncertain similarity score. I would label the relationship and let the reader inspect the items behind it.

Information has several valid organizations. An article may belong to a research question, a historical period, and a disagreement between two people. Any single arrangement privileges one of those readings. A useful map must let people change the view and still understand what moved.

Scale creates another problem. A few dozen items may form memorable neighborhoods; thousands can collapse into clutter. Progressive disclosure and landmarks help only if they remain stable enough to learn. Layout performance matters too, but I would first test whether people can explain the relationships the layout is supposed to show.

I am wary of making terrain decorative. Elevation, motion, or color should encode something a reader needs to know. Mapping height to popularity, for instance, would give already prominent material a literal mountain. The choice is not neutral. A different map might foreground uncertainty, an overlooked source, or a contradiction.

Shared spaces raise further questions. A frequently used route could become visible to others, but popularity need not determine where everyone travels. I would want traces to support exploration without turning collective habit into a fixed path.

There is also a tension between a stable personal map and a changing body of knowledge. Familiar landmarks help me return to work, while new sources and relationships should be able to alter the terrain. I do not think there is one universal balance. It depends on whether the map is supporting recall, open exploration, or a shared investigation.

Metasphere explored procedural terrain for knowledge navigation. ECCHR Explore applied semantic navigation to legal podcast material, and Trails pursued paths through associated information. These projects tested parts of the idea in different contexts; they do not amount to general validation that spatial interfaces outperform other forms of navigation.

The practical test is narrower. Does a map help someone find their way through a particular body of material, maintain context while changing scale, and notice a connection they can then verify? If it cannot do those things, a good list may be the better interface.

I would begin by arranging a small set of real material by hand. Ask people to find a known source, explain how two items relate, and resume an interrupted inquiry. Watch where they expect a path or landmark and where the layout misleads them. Compare the result with the existing search or folder view before adding procedural terrain or automatic clustering.

I would also ask them to retrace a route after the layout changes. If they cannot find their way back, visual novelty has cost them continuity. If they notice more connections but cannot explain why those items were adjacent, the map may be generating interest without understanding. Both failures are easy to hide behind an impressive landscape.

The important constraint is the material itself. Some collections have useful spatial relationships; others gain little from being placed on a map. The interface should make those differences visible rather than forcing the metaphor to win.