The encounter between a long-established transdisciplinary practice and generative computation poses a problem distinct from the production of artificial images or the automation of scholarly writing. What changes when a documentary artistic work, developed across decades, acquires machine-readable addresses and becomes available to computational interpretation? This essay proposes Socioplastic Form as an indirect organisation of knowledge constituted through the relation, fixation, rotation and revision of heterogeneous materials. The form is neither a metaphorical organism nor a database exhausted by its records. Drawing critically on recent work concerning the epistemology of artificial intelligence, media cognition, artistic research, human–AI creative practice and boundary objects, the argument distinguishes computational capacity from autonomous understanding, and procedural reciprocity from biological symbiosis. It suggests that the consequential unit is not an AI-generated output but a documented alteration in how a conceptual problem can be posed, compared and reconsidered. The resulting research programme joins artistic authorship, archival precision and experimentally assessable interpretation without presuming their equivalence.
1. THE FORM IS NOT THE COLLECTION
The accumulation of documents does not, by itself, establish an intellectual form. A collection may be immense, carefully classified and publicly accessible while leaving the conditions of its own organisation unexamined. Equally, an artwork may consist of written materials without treating those materials as transparent carriers of a thesis. The distinction matters for a practice in which decades of exhibitions, urban observations, conceptual writing, collaborative projects and digital publication have generated an extensive documentary record. Socioplastics is not adequately described as an archive of interdisciplinary interests. Its artistic proposition lies in the operations through which heterogeneous material becomes addressable, is brought into consequential relation and is subsequently interpreted under altered conditions. The resulting structure is indirect: neither any single publication nor a complete inventory can show all the connections that make a particular reading possible. Yet this indirectness must not be romanticised as mystery. It is made from identifiable practices of writing, numbering, editing, attribution, publication and cross-reference. The work is textual and documentary; it has no independent life, intention or biological development. Its formal continuity arises from the decisions and activities of those who produce, maintain and read it. The difficult claim is therefore not that a sufficiently large archive becomes intelligent. It is that a sustained artistic practice can establish a distinctive field of epistemic operations whose history is recoverable and whose propositions remain open to criticism. The point of departure is a distinction between size as a quantitative characteristic and form as a consequential organisation of differences.
2. BEYOND THE METAPHOR OF THE LIVING MACHINE
Matteo Pasquinelli's comparative epistemology of machine, organism and language offers a way to resist an easy vocabulary of technological vitality. His historical argument does not present artificial intelligence as the belated arrival of a synthetic organism. Instead, it examines how models of machinery, life and language have travelled across disciplines and acquired different political purposes. Most consequentially, he proposes that AI systems automate relational structures sedimented in cooperation, labour and culture (Pasquinelli, 2026). This account redirects attention from claims about machine interiority towards the histories of abstraction and social organisation incorporated into computational models. It is particularly useful for distinguishing two different kinds of relation. Socioplastics organises a historically situated corpus through editorial and conceptual decisions; contemporary language models perform statistical and computational operations through architectures shaped by large-scale training and deployment. Their interaction can change what a researcher notices and writes, but it does not make the two systems equivalent, and it confers neither consciousness on a document nor ownership of a thought on a model. Reciprocation is operational rather than physiological: an interpretation produced with computational assistance can lead to the redefinition of a Marker, the reconsideration of a reference or the discovery of an overlooked contradiction. Those editorial changes alter the documentary conditions of later interpretation. The biological term symbiosis is unnecessary here: a document is not an organism, and model-generated language is not evidence of subjective experience. Nor is mutual benefit automatic. Researchers may gain useful new comparisons; system designers may gain actionable feedback when users choose to provide it. These benefits occur through different institutional arrangements and cannot be inferred simply from a dialogue. The more exact term is computational reciprocity: a sequence of technically mediated exchanges in which human decisions remain attributable and the contribution of the model remains examinable.
3. WHAT THE COMPUTATIONAL MEDIUM ALLOWS AND RESISTS
The novelty of generative computation cannot be explained entirely through a theory of networks or archives. It is also a medium with formal constraints. Lev Manovich identifies probabilistic generation as a trade-off between variability and precise control, alongside an expansive range of creative degrees of freedom, the persistence of conventional patterns and an entanglement of style and content in model outputs (Manovich, 2026). His provocative term 'media cognition' identifies the unusual breadth of cultural operations available to users of trained models. That formulation should be treated analytically, not as proof of biological thinking. A language model can produce a convincing correspondence between spatial morphology and linguistic classification while overlooking the local history that makes the correspondence misleading. It can also identify a recurring conceptual distinction dispersed among hundreds of publications and return it in a form that makes scholarly reconsideration possible. These are not opposite verdicts on intelligence; they are different consequences of a probabilistic medium interacting with documentary constraints. For Socioplastics, a computationally suggested relation should therefore be distinguished from an established relation, an editorially accepted revision and an empirically supported inference. The machine-readable address can help retrieve a text; it cannot settle whether two texts instantiate the same operation. The productive tension is between scale and discrimination. An enlarged capacity to compare materials is valuable only if it retains the possibility of saying that an analogy fails, that a source is absent, or that a concept changes significance when transferred between fields. The alternative is a seamless linguistic synthesis whose fluency conceals the very differences the work was designed to maintain.
4. ARTISTIC KNOWLEDGE REQUIRES DURATION
Hedvig Jalhed and Mattias Rylander make a consequential distinction between the aesthetic qualities of AI-generated outputs and their contribution to the continuing development of artistic problems. Their account, informed by Popper's World 3 and Lakatos's thinking about research programmes, suggests that epistemic importance cannot be measured through immediate novelty, popularity or technical accomplishment. The more demanding criterion concerns the emergence and transformation of problems across extended periods (Jalhed and Rylander, 2026). They argue that the long-term epistemic contribution of present AI art has not yet been established merely by the production of compelling outputs. That caution also applies to Socioplastics. Twenty-five years of practice and a large number of words do not automatically validate the originality of its conceptual distinctions. What the duration does provide is the possibility of reconstructing when questions emerged, how their formulations changed, which earlier materials became newly relevant and where a line of thought ceased to be productive. The current act of assigning numbers, persistent identifiers and explicit cross-references is therefore not simply a belated act of preservation. It creates improved conditions for examining a pre-existing practice longitudinally. Fixation must not be confused with proof: a DOI identifies a publication; a title marks an editorial proposition; a Marker distinguishes a concept. Their usefulness depends on subsequent readings capable of testing, correcting or rejecting what they name. An artwork that makes knowledge through its form must also permit the history of its mistakes and failed analogies to remain legible. Otherwise recursion would be merely the repeated assertion of its own importance.
5. THE PROBLEM OF AUTHORSHIP WITHOUT ANTHROPOMORPHISM
Federico Bomba and Antonella De Angeli investigate agency and authorship in AI-related artistic practices through interviews and critical readings of three artists' work. Their account brings speech-act theory into contact with agential realism, questioning a simple opposition between sovereign artist and passive computational tool (Bomba and De Angeli, 2025). Sonja Rozental and Alwin de Rooij similarly challenge the binary of tool and collaborator, using Material Engagement Theory and an interpretative analysis of eighteen artists' practices to describe how creative possibilities are discovered through sustained interaction with machine-learning systems (Rozental and de Rooij, 2026). Both studies are useful, but their vocabularies should be handled carefully. Relational agency in a philosophical or analytic framework does not establish that a model possesses intentions, accountability or personhood comparable to those of an artist. Nor does recognising the contribution of data and algorithms absolve a publisher of responsibility for claims, omissions and attribution. In a textual work such as Socioplastics, the most useful distinction concerns the provenance of operations: who selected the source, which computational procedure proposed a classification, who accepted it, what evidence supported the revision and which version preserves that history. Authorship need not imply that the author personally generated every intermediate sentence or association; it does require an intelligible account of editorial responsibility. Computational participation can be substantial without being mystified. The new critical object is not a competition over who deserves the label of creator, but the documentary record of how a proposition was assembled, challenged and made public. That record can accommodate assistance without falsely attributing consciousness to an algorithm or surrendering the historical specificity of the work.
6. SOCIOPLASTIC FORM AS AN INDIRECT ARCHITECTURE
Socioplastic Form can now be defined more exactly as the indirect, distributed and revisable organisation of a body of work through operations that make conceptual differences recoverable across materials, formats and times. Relation establishes an encounter without asserting that the materials are equivalent. Fixation gives a distinction a stable documentary address without declaring its meaning final. Rotation changes the disciplinary, territorial or temporal position from which that distinction is examined. Return permits comparison with earlier formulations while preserving the record of intervening decisions. These operations do not function as a linear production pipeline. An urban drawing may be revisited through an essay; a historical exhibition may clarify the origin of a contemporary Marker; an external text may expose a contradiction between a conceptual definition and a field observation. The numbered node is one documentary unit, not the complete artwork. Neither the numerical sequence of the index nor the coexistence of eleven thousand addresses should be read as a direct chronological record of twenty-five years: older work can be indexed retrospectively, and publication order can differ from the period of artistic production. The Marker offers a localisable conceptual proposition, not a universal explanation. The Gravitational Corpus provides a cartography of intellectual interlocutors without assigning automatic authority to the most frequently cited. Distributed channels establish different publishing situations and editorial histories, while their interconnections permit readings no individual channel can contain. Such architecture is neither formless nor fully centralised. Its unity is procedural, while its content remains heterogeneous. Computational interpretation adds a new capacity for movement across this structure, but the legitimacy of that movement must be judged by the distinctions it makes possible and by the evidence it preserves. The indirect form is therefore not a hidden entity waiting to be discovered by sufficiently powerful machines; it is an accountable arrangement whose possible interpretations exceed any present description of it.
7. BOUNDARY OBJECTS AND THE CONDITIONS OF TRANSFER
A transdisciplinary claim becomes weak whenever the movement of terms across fields is mistaken for knowledge integration. Paraskeva Wlazlak and Kristina Säfsten's study of boundary objects in product realisation is instructive precisely because it is operational and situated. Their framework assesses how an object can support the exchange of specialised knowledge in a particular collaborative environment, and why its usefulness depends on the complexity and requirements of that environment (Wlazlak and Säfsten, 2026). Their domain is engineering, not the artistic work examined here; this distance is productive. A Socioplastics Marker can sometimes act as a boundary object between an architectural observation, a curatorial problem and a computational classification, but not simply because it appears in all three. Its terminology must be comprehensible enough to permit contact while specific enough to preserve the different evidentiary standards of those situations. A social interpretation of access cannot be inferred from a section drawing alone; a model-generated analogy is not a resident's testimony; an artwork can register a conflict without becoming a representative survey. The relevant form of transfer is conditional and may properly end in a documented refusal of equivalence. This can be observed in spatial research. The proposed AgonisticSpace WalkAlong work, for example, defines ten walking conversations as a curatorial inquiry in which movement, disagreement and situated observation produce distinct records, subject to consent and later critical writing (Lloveras, 2026). It remains a proposal rather than evidence of completed encounters. Its methodological relevance lies in showing how a documentary form can specify an encounter without pre-empting what participants will say. Likewise, a Dutch citizen-science study of co-developed walking routes demonstrates that residents' observations and deliberations can influence a realised neighbourhood intervention, but its empirical results cannot automatically validate another territory or research design (van Holland et al., 2025). Transdisciplinarity needs such restrictions if it is to remain an operation rather than a flattering description.
8. A RECIPROCITY THAT CAN BE TESTED
The relation between Socioplastics and computational systems should be evaluated through changes in interpretative practice rather than declarations of technological partnership. A tractable experiment would compare how a model answers the same research questions when given isolated documents, documents with provenance and chronological links, and documents with explicit conceptual distinctions and revision histories. The questions should require discriminating between similar terms, identifying incompatible positions, tracing changes in a Marker and refusing a proposed transfer where the source does not warrant it. Human evaluation would examine recoverable citations, false associations, omissions, preservation of disagreement and the capacity to distinguish documentary fact from generated interpretation. Cases should be chosen in advance, with independent reviewers and known source passages, so that a fluent answer cannot be mistaken for a successful reconstruction. The test must include negative controls: pairs of documents that appear lexically similar but have incompatible theoretical or evidentiary commitments. Improvements under richer conditions would support the narrow claim that explicit architecture enhances certain forms of machine-assisted reading; they would not demonstrate that the archive has become conscious, that the model understands as a person does or that artistic value is computationally measurable. Failures would be equally informative. If added metadata produces premature synthesis, the structure may need to represent exclusions and contested relations more clearly. If a model invents a genealogy, the documentary address must expose rather than absorb the error. The practical question is thus how an existing artistic and intellectual practice can participate in the technological present without surrendering its critical criteria to software performance. Participation may involve testing tools, reporting their failures, proposing new interfaces and revising the written corpus in response to defensible findings. Such participation need not grant the technology a privileged status. It is one contemporary extension of a longer practice of putting artistic, scientific and architectural methods under reciprocal scrutiny. The time of the artwork remains open because further human work and new technical conditions may change its possible interpretations. Its continuity depends on attributable human work, on accessible documentation and on the possibility that later readers will reject some of its proposed relations. Socioplastic Form names the accountable, transdisciplinary arrangement through which those changes can be documented, disputed and used to formulate problems that neither a closed theory nor a mere collection of URLs could make available in the same way.
REFERENCES
Bomba, F. and De Angeli, A. (2025) ‘Agency and authorship in AI art: Transformational practices for epistemic troubles’, International Journal of Human–Computer Studies, 205, 103652. https://doi.org/10.1016/j.ijhcs.2025.103652.
Jalhed, H. and Rylander, M. (2026) ‘World 3 revisited: Artistic research in the age of (de)generative AI’, Humanities and Social Sciences Communications, 13, article 1519. https://doi.org/10.1057/s41599-026-08964-5.
Manovich, L. (2026) ‘A medium that thinks: Generative AI and media cognition’, Emerging Media, 4(2), pp. 225–232. https://doi.org/10.1177/27523543261454458.
Pasquinelli, M. (2026) ‘Machine, organism and language: A comparative epistemology of AI models’, AI & Society, 41, pp. 7833–7845. https://doi.org/10.1007/s00146-026-03094-7.
Rozental, S. and de Rooij, A. (2026) ‘Neither tool nor collaborator: Rethinking human–AI co-creativity in artistic practice with Material Engagement Theory’, Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3791752.
van Holland, B.J., Schuling, R., Jepkema, N., Doueiri, Z.N., King, A.C. and de Jong, J. (2025) ‘Leveraging citizen science to improve exercise-friendliness in a low socio-economic status neighborhood: A first-generation study of the co-development of walking routes in the Netherlands’, Citizen Science: Theory and Practice, 10(1), article 27. https://doi.org/10.5334/cstp.820.
Wlazlak, P. and Säfsten, K. (2026) ‘Development of a boundary object assessment framework supporting knowledge integration in product realization’, Journal of Integrated Design and Process Science, 29(4), pp. 282–302. https://doi.org/10.1177/10920617251333333.