Actor-network theory

Actor-network theory (ANT) is a methodological approach developed by Bruno Latour, Michel Callon, and John Law that treats social and technical reality as effects of networked associations rather than as properties of discrete entities. The central move is analytical symmetry: human intentions and material resistances are granted equal explanatory weight. A door hinge and a bureaucratic regulation are both actants — entities that modify the behavior of other entities within a network.

ANT rejects the conventional sociological distinction between macro-structures and micro-interactions. There is no society that explains local events; there are only local events that, when stabilized and scaled, produce what we call society. This makes ANT a radical anti-essentialist framework: categories like nature, society, and technology are not foundations but outcomes — temporary settlements in ongoing processes of network-building.

The theory has been applied to scientific practice, technological innovation, legal systems, and medicine, always with the same question: how do networks hold together, and what happens when they break? A persistent criticism is that ANT's descriptive neutrality prevents normative evaluation — it can explain how networks stabilize but not whether they should. The question of whether ANT needs a political supplement, or whether its refusal of politics is itself political, remains unresolved.

The Computational Reading

Actor-network theory has been read as sociology, as philosophy of science, and as methodology. It has rarely been read as computational theory — but it should be, because the central ANT insight (that networks produce their own components through association) is precisely the insight that underlies the theory of self-organizing computation.

A network in ANT is not a graph-theoretic network (a set of nodes and edges with fixed properties). It is a generative network: the act of association modifies the properties of the associating entities. When a scientist recruits a microscope into her network, the microscope becomes a different entity — it becomes a 'scientific instrument' rather than a 'piece of glass and metal.' The scientist, meanwhile, becomes a 'microscope user' rather than merely a 'person.' This is not metaphor. It is a description of how the state of a system changes when new components are coupled to it.

This is exactly what happens in computational systems when modules are composed. A sorting algorithm, in isolation, has a well-defined computational complexity. When it is embedded in a database system, its complexity changes because the database system provides pre-sorted indices, buffer pools, and query optimizers that modify the effective input distribution. The algorithm is not 'used' by the database; it is transformed by the database, just as the microscope is transformed by the scientist. The network is not a container that holds pre-defined entities. It is a process that continuously redefines what its entities are.

The computational reading of ANT produces a prediction: the stability of a network should correlate with the computational closure of its components. A network is stable when the computational outputs of each component are sufficient inputs for the next, without requiring continuous renegotiation. This is why bureaucratic networks are stable: the forms, the rules, and the procedures create a computational closure — each agent knows what to produce given what they receive. It is also why they are brittle: when the environment changes and the closure breaks, the network cannot adapt because its components have been defined by the closed computation, not by any intrinsic adaptability.

The connection to algorithmic institutions is direct. A digital platform is an actor-network in which the mediators are not people and objects but APIs and data flows. The platform 'enrolls' users, merchants, and advertisers into a network that modifies their properties: a merchant on Amazon is not the same entity as a merchant outside Amazon, because the platform's computational infrastructure (search ranking, recommendation, fulfillment) transforms what the merchant can do and what the merchant is. The platform, like any ANT network, produces its own components through association. The difference is that the associations are mediated by code rather than by social negotiation, and the transformations are therefore faster, more scalable, and less reversible.

ANT's refusal to grant explanatory priority to either the social or the technical becomes, in the computational reading, a refusal to grant priority to either hardware or software. The network is the computation; the computation is the network. This is not a metaphor. It is a claim about what 'association' means when the associating entities are computational processes.