Ideas do not live in the mind alone. They survive, travel, mutate, and acquire authority through grammars: not only linguistic grammars, but documentary, infrastructural, archival, computational, logistical, and institutional grammars that determine how something becomes visible, thinkable, retrievable, credible, or governable. Across the works considered here, one proposition emerges with increasing force: knowledge is never merely produced; it is formatted, routed, classified, stored, measured, accelerated, interrupted, repaired, and sometimes suppressed by the systems that carry it. The life of ideas is therefore inseparable from the material and symbolic infrastructures that give ideas form. A theory may appear to be abstract, but it enters the world through papers, datasets, models, cables, archives, citations, files, repositories, keywords, roads, containers, search systems, protocols, and standards. Grammar, in this larger sense, is the architecture that makes thought operational.
Paul Otlet’s Traité de documentation provides the deepest historical foundation for this argument. Otlet understood the document not simply as a book, but as any material support through which thought could be fixed, circulated, analysed, recombined, and made socially useful. His Bibliology or Documentology imagined books, archives, images, diagrams, films, sound recordings, museum objects, catalogues, files, and encyclopaedic repertories as parts of one vast documentary organism. For Otlet, modern civilisation faced a crisis of abundance: the proliferation of documents could become either intellectual flood or collective irrigation, depending on the rational organisation of knowledge. His proposed Universal Documentation Network was therefore more than a bibliographic scheme; it was a political project of global intellectual coordination, designed to make information universal, reliable, complete, rapid, current, accessible, and available to the greatest number. In Otlet, grammar becomes civilisational: classification, notation, measurement, indexing, and standardisation are not neutral clerical operations but the conditions under which collective intelligence can exist. Otlet, P. (1934) Traité de documentation: le livre sur le livre, théorie et pratique. Bruxelles: Éditions Mundaneum.
This documentary imagination returns, in contemporary form, in Gebru, Morgenstern, Vecchione, Vaughan, Wallach, Daumé and Crawford’s “Datasheets for Datasets” and Mitchell, Wu, Zaldivar, Barnes, Vasserman, Hutchinson, Spitzer, Raji and Gebru’s “Model Cards for Model Reporting”. Both texts argue that artificial intelligence systems cannot be responsibly understood without documentation that records their origins, limits, intended uses, risks, and social implications. Datasheets make datasets accountable by asking why they were created, how they were collected, what they contain, who may be affected by them, and what they should not be used for. Model cards perform a parallel function for trained models, demanding transparency about performance across demographic, cultural, phenotypic, environmental, and intersectional conditions. These frameworks extend Otlet’s documentary ambition into machine learning: they insist that AI systems require grammars of description before they can be evaluated as social artefacts. Without such grammars, datasets and models circulate as opaque technical objects, concealing the histories, exclusions, and assumptions embedded within them. Documentation here becomes a political technology of responsibility. Gebru, T. et al. (2018) ‘Datasheets for Datasets’, Proceedings of the 5th Workshop on Fairness, Accountability, and Transparency in Machine Learning. Mitchell, M. et al. (2019) ‘Model Cards for Model Reporting’, Proceedings of the Conference on Fairness, Accountability, and Transparency, pp. 220–229.
The same politics of grammar appears in archival theory. Schwartz and Cook dismantle the myth that archives are passive repositories of facts, arguing that archives actively produce memory, identity, legitimacy, and silence. Selection, appraisal, description, preservation, access, and interpretation are all acts of power. The archive does not merely receive the past; it helps decide which past becomes durable. Cifor and Gilliland deepen this argument by showing that archives are also affective environments. Records generate grief, anger, longing, trust, trauma, recognition, and hope; absence itself can exert emotional force when records are destroyed, missing, imagined, or inaccessible. Caswell then radicalises archival practice through the idea of liberatory memory work, arguing that community archives must not simply include marginalised histories within dominant institutions, but actively intervene against symbolic annihilation. In these works, archival grammar is ethical and political at once: categories, finding aids, custody arrangements, metadata, access protocols, and silences shape the lives of communities. The life of ideas depends on whether memory is merely stored or made available for recognition, repair, and action. Schwartz, J.M. and Cook, T. (2002) ‘Archives, Records, and Power: The Making of Modern Memory’, Archival Science, 2, pp. 1–19. Cifor, M. and Gilliland, A.J. (2015) ‘Affect and the archive, archives and their affects: an introduction to the special issue’, Archival Science, 16(1), pp. 1–6. Caswell, M. (2021) Urgent Archives: Enacting Liberatory Memory Work. Abingdon and New York: Routledge.
The essay “Living Archives at Scale” synthesises these archival debates into the concept of archival metabolism. Its central claim is that contemporary archives, repositories, datasets, and post-digital knowledge systems can no longer be governed by preservation alone. Once archives grow beyond immediate human comprehension, they must ingest, digest, anchor, expose, shelter, delay, recombine, and reactivate materials. This metabolic model gives archival politics a scalar grammar. Care must become navigable; abundance must become inhabitable; machine legibility must not erase interpretive thickness. The five concepts proposed there—archival metabolism, scalar grammar, strategic porosity, differentiated speed, and stable nucleus / plastic periphery—offer an applied vocabulary for sustaining the life of ideas under conditions of digital saturation. A living archive needs stable anchors, such as DOIs, indexes, datasets, and core definitions, but it also needs plastic edges where concepts can branch, mutate, and return. Here grammar becomes ecological: it is the system by which ideas breathe, circulate, and remain available without dissolving into noise. Lloveras, A. (2026) ‘Living Archives at Scale: Reparative Care, Scalar Grammar and the Metabolism of Post-Digital Knowledge Infrastructures’.
Infrastructure theory extends this argument from archives to the built and technical systems of circulation. Brian Larkin shows that infrastructures are not neutral supports for movement but material forms of governance, imagination, and sensory experience. Roads, pipes, power grids, satellites, and transport systems produce subjects, expectations, exclusions, and fantasies of modernity. Infrastructure therefore has both politics and poetics: it organises circulation while also signifying progress, state power, sovereignty, or failure. Starosielski’s The Undersea Network similarly dismantles the fantasy of immaterial digital communication by showing that global connectivity depends on submarine fibre-optic cables embedded in oceans, beaches, colonial histories, military geographies, local conflicts, and environmental conditions. The internet does not float in a cloud; it passes through routed, fragile, territorial systems. Together, Larkin and Starosielski reveal that the life of ideas depends on physical pathways. Thought travels through pipes, cables, stations, landing sites, ports, and maintenance regimes. Even the most abstract digital exchange has a geography. Larkin, B. (2013) ‘The Politics and Poetics of Infrastructure’, Annual Review of Anthropology, 42, pp. 327–343. Starosielski, N. (2015) The Undersea Network. Durham and London: Duke University Press.
Cowen’s The Deadly Life of Logistics adds a sharper political economy to this infrastructural account. Logistics appears to be a neutral science of efficiency, yet Cowen shows that it descends from military supply and becomes a global technology for reorganising production, labour, territory, sovereignty, and security around the uninterrupted movement of capital. Containerisation, just-in-time production, intermodal transport, deregulation, and supply-chain management do not simply move goods; they restructure the world around circulation. This is crucial for understanding ideas as well. Academic concepts, digital systems, AI models, datasets, books, and archives all depend on logistical conditions: servers, shipping routes, institutional procurement, platform infrastructures, maintenance labour, and global supply chains. The politics of circulation determines which ideas travel rapidly, which remain stranded, which become profitable, and which are rendered invisible. Cowen makes clear that movement is never innocent. To circulate is to enter a regime of power. Cowen, D. (2014) The Deadly Life of Logistics: Mapping Violence in Global Trade. Minneapolis: University of Minnesota Press.
The computational works in this constellation complicate the life of ideas further. LeCun, Bengio and Hinton’s “Deep Learning” describes how neural networks learn representations from data through multiple layers of abstraction, transforming raw inputs into patterns, features, and classifications. Hayles’s Unthought expands cognition beyond consciousness, arguing that much of what enables perception, decision, adaptation, and meaning-making occurs through the cognitive nonconscious and through cognitive assemblages involving humans, machines, bodies, sensors, codes, and environments. Bender, Gebru, McMillan-Major and Shmitchell’s “On the Dangers of Stochastic Parrots” then interrupts the celebratory narrative of scale by warning that large language models recombine statistical patterns without grounded understanding, while reproducing environmental costs, social biases, and misleading fluency. These three texts form a tense conceptual triangle. Deep learning shows the technical power of learned representations; Hayles shows that cognition is distributed beyond the conscious human subject; Bender and colleagues warn that statistical fluency must not be mistaken for meaning, responsibility, or understanding. The grammar of AI is therefore both generative and dangerous: it can reorganise knowledge at scale, but it can also simulate intelligence while obscuring its material and political conditions. LeCun, Y., Bengio, Y. and Hinton, G. (2015) ‘Deep learning’, Nature, 521, pp. 436–444. Hayles, N.K. (2017) Unthought: The Power of the Cognitive Nonconscious. Chicago: University of Chicago Press. Bender, E.M. et al. (2021) ‘On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?’, Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 610–623.
Concepts live by being stabilised enough to be recognised and plastic enough to be reinterpreted. This is the central lesson that unites the archive, the dataset, the model, the cable, the container, the road, and the document. A concept without infrastructure remains private intuition. A concept overdetermined by infrastructure becomes bureaucratic residue. The life of ideas therefore depends on a delicate balance between fixation and movement. Otlet’s fiches, Caswell’s reparative archives, Gebru’s datasheets, Mitchell’s model cards, Starosielski’s cables, Cowen’s logistics, Hayles’s cognitive assemblages, and the metabolic archive all describe different aspects of this balance. Ideas need containers, but they also need passages. They need names, but also reinterpretation. They need memory, but also reactivation. They need metrics, but not metric domination. They need machine legibility, but not algorithmic capture. They need politics because every grammar distributes power: who may speak, what may be retrieved, how evidence is valued, which bodies are recognised, and what futures can be imagined.
In conclusion, grammar is not a decorative order imposed after thought. Grammar is the condition through which thought becomes transmissible, durable, governable, and contestable. Politics is not external to this process, because every documentary, archival, computational, and infrastructural grammar allocates visibility and invisibility, speed and delay, authority and marginality, preservation and erasure. Concepts live only when they pass through systems capable of holding them without freezing them, exposing them without flattening them, and circulating them without severing them from responsibility. The future of knowledge therefore requires neither pure openness nor rigid control, but metabolic architectures: systems that can ingest abundance, preserve ethical continuity, sustain interpretive plurality, and keep ideas alive as they move across documents, archives, infrastructures, machines, and communities.
Bibliography
Bender, E.M., Gebru, T., McMillan-Major, A. and Shmitchell, S. (2021) ‘On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?’, Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pp. 610–623.
Caswell, M. (2021) Urgent Archives: Enacting Liberatory Memory Work. Abingdon and New York: Routledge.
Cifor, M. and Gilliland, A.J. (2015) ‘Affect and the archive, archives and their affects: an introduction to the special issue’, Archival Science, 16(1), pp. 1–6.
Cowen, D. (2014) The Deadly Life of Logistics: Mapping Violence in Global Trade. Minneapolis: University of Minnesota Press.
Gebru, T., Morgenstern, J., Vecchione, B., Vaughan, J.W., Wallach, H., Daumé III, H. and Crawford, K. (2018) ‘Datasheets for Datasets’, Proceedings of the 5th Workshop on Fairness, Accountability, and Transparency in Machine Learning.
Hayles, N.K. (2017) Unthought: The Power of the Cognitive Nonconscious. Chicago: University of Chicago Press.
Larkin, B. (2013) ‘The Politics and Poetics of Infrastructure’, Annual Review of Anthropology, 42, pp. 327–343.
LeCun, Y., Bengio, Y. and Hinton, G. (2015) ‘Deep learning’, Nature, 521, pp. 436–444.
Lloveras, A. (2026) ‘Living Archives at Scale: Reparative Care, Scalar Grammar and the Metabolism of Post-Digital Knowledge Infrastructures’.
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I.D. and Gebru, T. (2019) ‘Model Cards for Model Reporting’, Proceedings of the Conference on Fairness, Accountability, and Transparency, pp. 220–229.
Otlet, P. (1934) Traité de documentation: le livre sur le livre, théorie et pratique. Bruxelles: Éditions Mundaneum.
Schwartz, J.M. and Cook, T. (2002) ‘Archives, Records, and Power: The Making of Modern Memory’, Archival Science, 2, pp. 1–19.
Starosielski, N. (2015) The Undersea Network. Durham and London: Duke University Press.
