Embeddedness

In the AI Story

Classical economics treats the market as a disembedded mechanism — a neutral coordination device operating above social relations. Granovetter demonstrated, through extensive empirical work on industries, firms, and labor markets, that this picture is fiction. Actual economic behavior is conducted through specific, named relationships with histories, obligations, and expectations. Trust, reputation, and personal loyalty are not frictions on market efficiency; they are the substance of market function.

The framework generalizes beyond economics. Any social process that appears to transcend its specific relational context — pure science, universal law, algorithmic judgment — is, on close inspection, embedded in concrete social relations that shape both its production and its application. The appearance of disembeddedness is itself a social achievement, maintained by specific institutional arrangements that obscure their own contingency.

Applied to AI, the framework reveals what the technology's apparent universality conceals. The large language model appears to offer disembedded knowledge — information from any domain, available to anyone, without the social context that would normally accompany it. But the model is trained on specific documents produced by specific institutions in specific languages reflecting specific cultural traditions. The training data is socially embedded; therefore the outputs are socially embedded.

Granovetter himself made this point in a 2022 interview: No matter what kind of big data or artificial intelligence or machine learning that employers are able to draw on, they will never know as much about a person as someone who actually knows them and has worked with them. The statement is not nostalgia for human connection. It is a structural claim about what different information sources can and cannot carry — personal knowledge is embedded in sustained relational engagement, and statistical processing cannot reproduce it.

Origin

The embeddedness argument appeared in Granovetter's 1985 American Journal of Sociology paper, which became nearly as cited as his 1973 weak-ties paper. The framework drew on Karl Polanyi's earlier use of the term in The Great Transformation (1944) but sharpened it into a structural claim about the inseparability of economic and social action.

The framework became foundational to economic sociology and has been extended into studies of corporate governance, industrial districts, international trade, and — now — algorithmic knowledge production.

Key Ideas

Action is always relational. Individual decisions occur within networks of ongoing relationships that shape preferences, constrain options, and structure consequences.

Markets are social constructions. The coordination attributed to market forces is produced by specific, describable relational arrangements — not by abstract mechanisms.

Trust is structurally generated. Reliable cooperation emerges from the network position of the parties, not from personal virtue or rational calculation.

AI is embedded knowledge. The apparent universality of language-model outputs conceals the specific social structures encoded in training data and corporate governance.

Disembeddedness is a myth. The claim that any process transcends its social context is itself a social achievement maintained by institutions whose contingency is systematically obscured.

Debates & Critiques

How thick the embeddedness claim should be — whether it describes a tendency or a necessary feature of economic action — has been debated for four decades. The strong version holds that disembeddedness is impossible in principle; the weak version holds that disembeddedness is rare in practice. Granovetter himself tended toward the strong version.