Local knowledge is Geertz's name for the form of understanding that cannot be extracted from context without loss. The claim that all societies have kinship systems is universally true and practically empty. The consequential knowledge about kinship — how it shapes inheritance, marriage, the distribution of obligation, the texture of emotional life — is always local, embedded in specific histories and specific institutions. The present volume applies the concept to the AI discourse, which is saturated with universal claims — AI democratizes capability, AI increases productivity, AI threatens employment — that conceal the radical variation in what these claims actually mean for specific populations in specific contexts.
The distinction between local and universal knowledge is not a sentimental preference for the particular. It is an epistemological argument about where consequential understanding lives. Universal claims provide a vocabulary. Local knowledge provides the grammar. And a language cannot be spoken with vocabulary alone.
The AI transition presents the universalist temptation in unusually acute form because the technology is genuinely global while its effects are radically local. The same tool is available in Lagos, Trivandrum, and San Francisco. The experience of the tool is not the same in these places. The developer in Lagos operates within a web of significance that includes unreliable infrastructure, economic precarity, and distance from the centers of capital that convert prototypes into products. The engineer in Trivandrum operates within a web shaped by decades of software industry development in India. The San Francisco founder operates within a web of venture capital culture and disruption mythology. The universal claim — AI democratizes capability — is true for all three. It means different things in each context.
The failure to recognize this is not merely intellectual. It produces policy responses that work in one context and fail catastrophically in another. The dams that direct the AI current must be built locally, by people who understand the local landscape, because the same current produces rapids in one terrain and marshes in another. A dam designed for one landscape may fail in the other.
Geertz's insistence on local knowledge was never anti-comparative. Comparison was essential to his method — insights from Java illuminated later observations in Morocco, and the differences between the two contexts were as instructive as the similarities. The argument was against premature universalization: extracting general principles from specific cases before the cases had been described thickly enough to reveal what the general principle was actually supposed to capture.
The concept is articulated across Geertz's career but receives its most systematic treatment in his 1983 collection Local Knowledge: Further Essays in Interpretive Anthropology. The book extends the interpretive method from the single-society focus of The Interpretation of Cultures into comparative territory, developing the argument that genuine comparison requires thick description of both sides of the comparison rather than the application of a universal framework to multiple contexts.
The distinction has been adopted across fields from legal studies (where local legal cultures resist universal legal principles) to software engineering (where local code practices resist universal design patterns). Its transferability testifies to the generality of the insight: understanding that travels well is usually understanding that has been stripped of the specificity that made it consequential.
Universal claims are frames, not contents. They tell us what to look for; they do not tell us what we will find in any particular place.
Meaning is always realized locally. The same event, experienced in different webs of significance, produces different meanings and different consequences.
Premature universalization obscures. Generalizing before the specific has been thickly described produces empty vocabulary dressed as explanation.
Dams must be built locally. Institutional responses to the AI transition cannot be universally specified; they must be adapted to the specific landscapes in which they operate.
Comparison requires thick description on both sides. Genuine comparison reveals differences rather than confirming uniformity.