Artificial intelligence becomes governable only when it is understood as an infrastructure of conversions. Cultural works become training material; labour becomes workflow; electricity becomes prediction; public archives become model capacity; language becomes a commercial service; and territorial resources become the hidden substrate of apparently weightless outputs. Each conversion redistributes visibility and power. Some actors acquire scale, speed and abstraction, while others encounter extraction, opacity or the loss of control over their own contributions. Accountability begins by refusing the final interface as the true location of the system. The prompt, newsroom tool or generated song is only the visible threshold of a much longer arrangement composed of datasets, licences, standards, data centres, grids, organisational rules, metadata and maintenance labour.
The epistemic ground of accountable infrastructure is situated knowledge. No system is universal merely because it processes large quantities of material. Data carries histories of selection; categories preserve institutional hierarchies; models amplify the languages, regions and cultural forms most available to them. Feminist analysis makes this asymmetry explicit by asking who defines the problem, whose bodies absorb environmental and social costs, whose labour disappears and whose consent is assumed. Context is not supplementary information added to a finished model. It is the condition under which a technical operation can be judged. The same act of copying may serve research, competition, preservation or market substitution; the same automated function may assist an under-resourced newsroom while deepening dependence upon an opaque platform.
Infrastructure is political because it allocates scarce capacities. Data-centre growth places computational demand inside electricity systems planned for households, industry and public services. Training regimes place creative production inside markets governed by copyright and licensing. Newsroom automation places editorial judgment inside organisations with unequal resources and weak policy. Open-source publication places maintenance obligations upon communities that may never have been funded to carry them. These are not separate externalities surrounding AI. They are the system’s concrete form. Governance must therefore operate across scales: at the level of model design, organisational practice, regional energy planning, labour standards, licensing institutions and public regulation. Efficiency matters, but it cannot substitute for decisions about priority, ownership and legitimate use.
Aesthetic and archival life is equally infrastructural. Cultural material does not enter computation as inert content; it enters through acts of classification, digitisation, licensing, scraping, description and technical translation. A cultural commons is not produced by making files visible. It requires gateways through which people can reuse, reinterpret and govern shared resources, as well as an installed base of institutions, practices and standards capable of supporting those actions. Metadata becomes a political architecture because it determines whether provenance survives circulation and whether recognition can be converted into compensation. The archive is thus not behind AI as a passive reservoir. It is continuously redesigned by the mechanisms through which material becomes findable, interoperable, trainable and economically addressable.
Socioplastics can operationalise this condition by treating the corpus as a maintained public infrastructure rather than a mass of accumulated texts. Each node requires provenance, durable identity, intelligible relations and conditions of reuse; each interface should function as a gateway rather than a terminal display. Indexing must disclose rather than conceal the transformations that make material portable. Openness should be measured not only by public access but by usability, sustainability, equity and the ability of different communities to appropriate the corpus without surrendering context. Such a field does not reject scale, automation or machine legibility. It builds counterweights into them: traceable sources, distributed formats, explicit permissions, redundant repositories and human-readable paths through computational organisation.
The decisive contribution is a shift from ethical commentary to infrastructural composition. Accountability is not a statement attached to a system after deployment; it is embedded in how energy is procured, how data is acquired, how consent is interpreted, how labour is recognised, how software is maintained, how cultural objects retain provenance and how institutions permit contestation. A corpus becomes public when its technical architecture supports durable participation and when its abstractions remain answerable to the territories, people and works from which they derive. What accountable infrastructuring adds to Socioplastics is a precise discipline for making scale legible, use negotiable and continuity collectively sustainable.