Context Distance: Why Intelligence Cannot Be Centralized the Way Data Can

Consolidating data into central data stores is taken to be a precondition for insight. For AI agents this assumption is wrong in principle: the context from which an agent reasons should sit as close as possible to where it originated. This essay introduces context distance as a metric, decomposes it into three components (dilution, distortion, and loss of the ability to query back), and argues that data tolerates copies whereas intelligence tolerates only references. Centralizing compute is unaffected by this; centralizing context, however, degrades intelligence into data about a former why. Knowledge graphs follow the same ordering with one exception: reference knowledge may be central, whereas observation, interpretation, and evidence must stay local, because a merged graph, once a reasoner runs over it, produces statements that hold in no source. What remains open is how higher-order insight arises when context stays local.