Mark Granovetter

In the [YOU] on AI Field Guide

The cycle’s account of who benefits from AI and who does not is, at its structural foundation, a Granovetterian account. The communities that crossed the threshold of recognition fastest—software developers, open-source contributors, startup builders—are communities that are, by the standards of network theory, extraordinarily rich in weak ties. Open-source projects connect strangers across organizational boundaries. Forums and social platforms function as massive weak-tie generators. Conference culture bridges the gaps between companies and specializations. In such a network, the orange pill moment diffused with the speed of fire through dry grassland: each builder who experienced the threshold crossing became a bridge, carrying the recognition to clusters she was connected to through weak ties that the recognition had not yet reached.

Other professional communities, less rich in weak ties, experienced the diffusion more slowly. The legal profession, with its strong internal culture and its resistance to external information, maintained higher barriers. The academic community, organized around disciplinary silos with limited cross-boundary connection, received the information unevenly. Granovetter’s threshold model—developed in 1978 to explain why similar groups produce dramatically different outcomes in collective action—explains this pattern with precision: the same average threshold can produce rapid cascade in one network and complete stagnation in another, depending on the distribution of thresholds and the density of bridging capital.

Granovetter’s most provocative contribution to the cycle, however, is not the explanation of diffusion but the explanation of trust. His 2022 observation that personal knowledge—the kind that comes from having worked with someone, having seen them under pressure—is a form of information that statistical processing cannot replicate maps precisely onto the cycle’s diagnosis of what AI cannot provide and what human relationships must continue to supply. The builder who trusts Claude the way she would trust a close colleague has confused the subjective intensity of the interaction with the evidential basis of the relationship. AI is structurally a weak tie, and the appropriate trust calibration is a weak-tie calibration: appreciate the novelty, verify independently.

His concept of embeddedness—the insistence that all economic action is embedded in concrete social relations—provides the deepest critique of the AI infrastructure that Zuckerberg is building. The decisions about what to include in a training corpus, how to weight different sources, which outputs to reinforce—these are social decisions, made by specific people in specific institutional contexts, reflecting specific values. The AI tool that appears to offer disembedded knowledge is itself embedded in the social relations that produced it. Understanding this embeddedness is the condition for using the tool wisely rather than being used by it.

Origin

Born in 1943, Granovetter completed his doctorate at Harvard in 1970 with a dissertation on how people find jobs—a question that seemed sociologically modest and turned out to be philosophically rich. The dissertation revealed the pattern that would become the 1973 paper: among the professionals who had changed jobs and found new positions through personal contacts, the overwhelming majority found them through people they saw occasionally or rarely, not through their closest associates. This empirical finding contradicted the intuitive assumption that intimate relationships are the most valuable—and it demanded a structural explanation rather than a psychological one.

The structural explanation Granovetter provided was the concept of bridging: weak ties are valuable because they connect different social clusters, giving each side access to information that circulates only within the other. Strong ties, by contrast, tend to connect people within the same cluster, recycling information that is already known. The paper transformed the sociology of social networks and laid the groundwork for every subsequent network-theoretic analysis of how innovation diffuses, how communities adopt new technologies, and how inequality is reproduced through the structure of connection rather than the deficiencies of individuals. Ronald Burt’s later concept of structural holes extended Granovetter’s framework into organization theory, giving managers a precise vocabulary for the bridging positions that capture disproportionate value.

Granovetter’s 1985 paper “Economic Action and Social Structure: The Problem of Embeddedness” added a second major contribution: the argument that economic sociology had to recover from the twin errors of over-socialized and under-socialized accounts of human behavior. The neoclassical actor who responds only to prices is under-socialized—abstracted from the relational context that shapes every real economic decision. The over-socialized actor who simply enacts social norms is equally wrong. What actually exists is something in between: a concrete, historically specific actor whose economic decisions are embedded in ongoing social relationships that cannot be abstracted away without distorting what the decisions actually are.

Key Ideas

The Strength of Weak Ties. The paradox at the heart of Granovetter’s work: the relationships that feel least important are structurally the most valuable for accessing novel information. Close friends inhabit the same social world and carry largely redundant information; acquaintances inhabit different worlds and carry maximally non-redundant information. AI, understood through this framework, is the most powerful synthetic weak tie in history—a connection to the entire documented knowledge landscape that delivers non-redundant information at scale. The structural advantage of AI adoption is not primarily about productivity; it is about network position. Access to the strongest weak tie ever built is access to more novel information, which compounds over time exactly as network advantage always does.

Structural Holes and the Value of Bridging. Ronald Burt extended Granovetter’s framework into a theory of structural holes—the gaps between disconnected groups where the greatest creative and commercial opportunities reside. The person who bridges a structural hole controls the flow of novel information between disconnected communities and captures disproportionate value. AI demolishes the human monopoly on bridging: when any builder can access a tool that bridges effectively all documented structural holes simultaneously, the structural advantage of human bridging diminishes. The floor of creative possibility rises for those who were previously structurally disadvantaged; the ceiling drops for those whose advantage came primarily from their bridging position.

The Threshold Model of Collective Behavior. Granovetter’s 1978 model shows that the distribution of individual thresholds in a group determines whether a cascade of adoption occurs. Two groups with identical average thresholds can produce dramatically different outcomes depending on whether the thresholds are evenly distributed or clustered. Applied to AI adoption, the model predicts exactly the pattern the cycle documents: rapid cascade in communities with skewed-low threshold distributions (developer communities), stagnation in communities with clustered thresholds (established professional guilds). The cascade is not about individual intelligence or openness to change; it is a structural property of the network.

Embeddedness. All economic action is embedded in concrete social relations—this is Granovetter’s deepest claim, and it applies with full force to AI. The training corpus is not the entire landscape of human thought; it is a historically contingent sample reflecting the biases of the institutions that produced the documents it contains. The connections the AI can make are bounded by the connections that exist in the documented, digitized, predominantly English-language knowledge landscape. The connections it cannot make—the bridges to knowledge that was never written down, never digitized, never published—are invisible to the builder, because the absence of a connection is structurally undetectable from within the system that fails to make it.