Cognitive Maps

In the AI Story

Cognitive maps are not literal maps. Tversky's research shows they are more accurately called cognitive collages: piecemeal, distorted, schematized assemblies of spatial information that nonetheless support effective reasoning. We remember routes as sequences of landmarks and turns rather than as geometrically accurate layouts. We organize social networks as proximity rather than precise coordinates. We structure knowledge domains through containment relations, adjacency, and hierarchy — spatial primitives that the mind deploys fluently even when the underlying domain has no literal spatial character.

For AI collaboration, the cognitive map framework illuminates why certain prompts succeed and others fail. A prompt that matches the user's cognitive map of the problem — describing the architecture as a flow, the system as a hierarchy, the data as a network — activates the user's existing spatial model and allows the model's response to plug into that structure. A prompt that demands a translation into an unfamiliar spatial structure imposes representational mismatch at the outset.

The framework also explains why expertise transfers unevenly. An expert in one domain has built elaborate cognitive maps of its territory — knowing which ideas are adjacent, which are central, which are peripheral. When the same expert works in an adjacent domain, some of the spatial structure transfers; when she works in a distant domain, she must build a new map from scratch. AI collaboration can accelerate the map-building but cannot replace it — the map lives in the user's head, not in the tool.

Origin

Tolman's 1948 paper "Cognitive Maps in Rats and Men" introduced the term. Tversky's work from the 1980s onward extended the concept to human knowledge organization and demonstrated the systematic distortions — alignment, rotation, hierarchical organization — that characterize human spatial memory.

Key Ideas

Maps as metaphor engines. Spatial structure provides the primitives through which we reason about abstract domains.

Cognitive collages. Our mental maps are schematic and distorted, not geometric — but their distortions are systematic and support effective reasoning.

Knowledge domains as territories. Expert knowledge is organized as spatial structure, which is why expertise feels like familiarity with a place.

Map transfer and mismatch. The quality of AI collaboration depends on whether the user's cognitive map of the problem matches the representation the tool produces.