New York on the Map vs in Idea Space
Concept · vector_semantics · hypothesis · visibility: working
The cleanest single illustration of the whole thesis — and the thing she named as ‘the real thing to
illustrate.’ On a map, New York City is a point at coordinates (~40.7°N, 74.0°W): position fixed,
handed over by the rock, measured from Greenwich, neighbors decided by physical adjacency — Newark,
Yonkers, the Atlantic. One position, one neighborhood, imposed. In idea / latent space, New York is
a vector learned from everything ever said about it: neighbors are whatever it is related to —
Tokyo, London, Paris (peer world-cities), Gotham, Wall Street, hip-hop, ‘the city that never sleeps.’
Position isn’t given, it’s learned from relations, and it can sit in many neighborhoods at once
(financial-NYC near Hong Kong; cultural-NYC near LA; the-idea-of-NYC near ‘ambition’). The map says NYC
is near Newark; idea space says NYC is near Tokyo. Both true — only one is interesting for most
questions. The map’s neighbors are decided by a rock; idea space’s neighbors are decided by meaning.
Bridge: the P31/P279* query (NYC → city → settlement → geographic location) is idea-space showing
through the graph — Wikidata is a discrete, hand-built approximation (named edges) of what an
embedding holds continuously (learned coordinates). Same intuition, two implementations. And a rhyme
worth keeping: even visualizing latent space means projecting it down to 2D (t-SNE / UMAP), which
distorts — the Theorema Egregium lie repeats in ML. You cannot flatten the idea-manifold honestly
either.
Context: arose 2026-06-17; likely the first explorable interaction to build (NYC’s two neighborhoods,
side by side).
Connections
- → A.geography-fails-at-both · illustrates
- → C.welded-to-one-shape · contrasts_with
- → C.latent-space · rests_on
- → P.sanctioned-topology · conceptually_related_to
source: live code session (2026-06-17) · extraction: auto (schema v3.1)