The Autonomous Algorithmic Work

For the ontology of an artwork, the artist’s authorised act of constitution is the absolute primary source. No fiction of a neutral external position is required. The legal and artistic act of constitution alone determines which properties are constitutive, which changes remain permissible, how individual manifestations belong to the work, and wherein the work’s identity consists over time.

This text exceeds the status of a mere artist’s statement and requires no external legitimation. Scientific models and theoretical discourses serve here exclusively to clarify this constitution of the work conceptually, to anchor it formally, and to disclose its radical consequences. They never rise above the work; they submit unconditionally to its definition.

The prevailing discourse on AI art narrows itself to the trivial guiding question of whether a machine is creative, conscious, or capable of legal authorship. For SCULPTURE and 8SECONDS, this anthropomorphism is irrelevant. The decisive rupture does not consist in attributing a human personality to a machine. It consists in constituting an artwork in such a way that its production of form, its temporal continuation, and its authorised changes of state are definitively decoupled from the artist’s continuous individual decisions. The autonomy of this work is exclusively work-ontological and operative.

The autonomous algorithmic work does not beg for subsequent admission into existing categories of art history. It necessarily constitutes its own category from within itself.

1.1 Demarcation: The Non-Organic Entity

To exclude every romantic transfiguration, a sharp ontological boundary must be drawn: the autonomous algorithmic work is neither a conscious person nor a biological organism nor a legal author. It possesses no human subjectivity and does not require one in order to exist as an independently operating work-instance.

Nor is its potential unfinishability a metaphysical claim to physically guaranteed eternity. Rather, it defines the categorical absence of an internal terminal state: each generated state is necessarily followed by another for as long as the machinic and institutional conditions of its execution are maintained.

Not every algorithmic generation of form is an autonomous work. The abyss between a digital tool, a mere generator of variants, and a genuine autonomous work-instance is absolute. SCULPTURE is constituted not by the simple presence of code or computing power, but solely by the indivisible unity of the original artistic constitution, autonomous work-process, continuous sequence of states, authorised materialisation, and an identity that understands time as sculptural material.

SCULPTURE designates a body of work developed since 2015 in which sculptural form is no longer conceived exclusively as an object determined in advance. Artistic intervention shifts to the constitution of a system that generates sculptural geometries and autonomously continues them within a defined horizon of form. In 2017, a real AI sculpture was materialised from this context. The historical claim “the world’s first AI sculpture” refers to this precise category: not the first computer graphic, not the first algorithmic three-dimensional representation, and not the first evolutionary simulation, but a form generated by artificial intelligence, authorised as a real sculpture and materially realised.

The formal language of SCULPTURE is not the arbitrary output of a neutral system. It emerges from an artistically constituted order. The artist determines the aesthetic horizon, the spatial relations, the limits of the form-space, and the conditions under which a digital state can count as sculpture. The individual form, however, is no longer completely drawn in microscopic detail. The work develops concrete states within a world that has been constituted but not exhausted in advance.

2.1 The Work as Temporal Enactment

With 8SECONDS, this displacement is radicalised. From 18 June to 29 September 2022, the autonomous artistic AI continuously generated new sculptures at a rhythm of eight seconds. Within the work context, 1,111,440 events of generation are designated. Their quantity alone is not decisive. What is decisive is the temporal rule: a sculpture possesses presence for eight seconds; with the appearance of the next, it is overwritten as the present state of the work. Emergence, overwriting, disappearance, and re-emergence form a single continuous creative process.

Overwriting is not a technical side effect and not a loss that an archive would have to repair retrospectively. It is an artistic operation. Each form enters the present, asserts its singularity for a limited duration, and is then displaced by a new form. The work does not merely produce forms; it produces transience. The destruction of the preceding state is as constitutive as the generation of the new one.

The eight-second rhythm is therefore not merely a playback speed. It is measure, rhythm, and law of the work’s existence. Just as material, weight, and volume structure classical sculpture, time here structures the possibility of form. The duration of a state becomes a sculptural property. A form that exists for eight seconds and disappears is ontologically a different sculpture from the same geometry stored permanently and displayed simultaneously with all others.

8SECONDS is not an archive of 1,111,440 forms. The work is the autonomous act in which 1,111,440 sculptures emerge one after another and displace one another from the present.

2.2 Movement, Body, and Feedback

The collaboration with Friedemann Vogel extends 8SECONDS into a bodily and performative dimension. Movement data and the sculptural sequence of forms do not stand in a simple relation of illustration and accompaniment. The human body affects the process of form, while the emerging forms act back upon the choreography. A reciprocal arrangement arises in which body, algorithm, image-space, and time form one shared event. The premiere of the film at Theater Lindau on 29 September 2022 made this relation publicly visible (Global Arts Link, 2022; Vogel, 2022).

For the concept of the work, it is not decisive to reduce every technical coupling to a single model of online learning. What is decisive is that the sculpture does not exist as a static result, but as a responsive temporal order of form. Learning, adaptive updating, and feedback are technical degrees of intensity within the system. The ontological core of the autonomous work, however, does not depend solely on whether parameters are updated during a particular performance. It already lies in the independent production and displacement of work-states.

2.3 The Real Sculpture as an Authorised Instance

Real sculptures can emerge from the digital work. This transition is neither a free imitation nor an interpretation that third parties may translate at will. A real sculpture belongs to SCULPTURE only when its geometry derives from an authorised state of the work and the chain of transformation remains traceable through to the physical object. The principle of the work is:

The material sculpture consequently has a double determination. It is an independent object with scale, mass, surface, ageing, and site. At the same time, it is the physical instance of a specific digital state. Its authenticity rests not only on material or signature, but on the genealogical connection to the work-code. The classical distinction between original and reproduction is thereby reordered: the real sculpture is original because it emerges from the autonomous work under authorisation, not because the artist manually determined its entire geometry.

3.1 From the Produced Object to the Continuing Work-Instance

The classical concept of the work frequently presupposes that artistic activity comes to rest in a completed object. Process art, conceptual art, and digital art have repeatedly expanded this order. Yet in many cases the work remains bound to a particular event, a score, a software version, or a finite set of authorised outputs. SCULPTURE begins at another point: the work is not merely open to different realisations; it possesses an enduring activity of its own.

This activity consists in the production of sculptural presence. The generator is not merely a production tool external to the work. Its execution, its respective state, its temporal rule, and its capacity to bring forth new forms belong to the identity of the work. The work is therefore neither only code, nor only process, nor only a class of possible objects. It is a technically embodied work-instance that continues its identity through changing states.

The term “work-instance” is chosen deliberately. It does not designate a person, but an operative unity that acts within the conditions established by the artist. It can bring forth, select, alter, and replace states without every individual decision being made by the artist. Its agency is functional and work-specific. It requires neither consciousness nor free will in order to be causally and aesthetically effective.

3.2 Self-Overwriting as Creative Form

The radical innovation of 8SECONDS does not lie in the mere speed of generation. It lies in the fact that the work enacts the conditions of its own appearance and disappearance. Every new state establishes a difference that removes the previous state from the current presence of the work. A form of self-overwriting thereby arises: the work writes itself onward by simultaneously erasing itself.

This logic contradicts the usual digital conception according to which everything can in principle be stored, duplicated, and made simultaneously available. 8SECONDS precisely does not turn digital form into an infinite archive. It introduces a scarcity of presence. Only one state is current at any given moment. The work thus possesses an irreversible temporality, even if technical provenance protocols may be preserved in the background. Documentation and the present state of the work must be distinguished: a hash or state protocol can testify that a form existed without restoring it as the present state of the work.

3.3 Time as Sculptural Material

In classical sculpture, time often becomes visible indirectly: as the duration of production, the ageing of material, the movement of the viewer, or historical patina. In 8SECONDS, time becomes immediately formative. Temporal limitation determines the mode of existence of each sculpture. Rhythm organises perception, memory, and loss. The viewer can apprehend a form but cannot possess it permanently; while interpretation begins, its disappearance is already inscribed.

Sculpture thereby shifts from a static body to a temporally indexed state. Its complete determination is not merely “this geometry,” but “this geometry within this interval, at this point in the history of the work.” The time-index is as constitutive as curvature, topology, or scale. A state materialised from 8SECONDS carries this temporality forward: it becomes the enduring bodily condensation of an originally transient moment.

3.4 The Work Beyond the Physical Presence of the Artist

With the autonomous algorithmic work, the relation between work and artist’s lifetime also changes. A traditional studio work may survive its author’s death, but it ordinarily produces no new states that remain identical with the work. SCULPTURE is conceived differently. Its constitutive rules allow the work-process to continue even when the artist is no longer physically present.

This possible continuation is not a secondary conservation scenario, but part of the constitution of the work. The artist does not merely create an object that survives him; he establishes a work-instance capable of continuing its creative activity without his ongoing decision. The artist’s death therefore does not necessarily terminate the work. Rather, it reveals that autonomy was inscribed into the structure of the work from the outset.

The artist does not create every future form. He creates the conditions under which the work can continue to bring forth form without him.

4.1 Description

The descriptive level encompasses the genesis of the work since 2015, the material realisation of an AI sculpture in 2017, the temporal generation of 8SECONDS in 2022, the eight-second rhythm, the continuous sequence of states, the bodily-performative connection, and the possibility of authorised physical manifestations. These statements define the object-domain of the text.

4.2 Thesis

The work-ontological thesis is that these elements must not be understood as a loose chain of production. They form a new unity: the autonomous algorithmic work. Its identity lies in the continuity of artistic constitution, algorithmic possibility-space, executable work-instance, temporal sequence of states, self-overwriting, and authorised manifestations. The thesis does not interpret the technical processes from outside; it determines which of them are artistically constitutive.

4.3 Hypothesis

The generalising hypothesis is that this type of work opens a new order for art, aesthetics, and institutions. Future works can possess an identity bound neither to an unchanging object nor to the continuing presence of their author. Museums, collections, law, and conservation will therefore have to take account not only of material artefacts, but also of executable lines of state, rules of authorisation, and permitted forms of further development.

4.4 The Status of Artistic Primary Knowledge

A work-code need not be publicly accessible in order for the work to count as an autonomous algorithmic work. Source code, model parameters, and technical protocols are important components of the work architecture and may be secured for conservation, research, or provenance. Their public disclosure, however, is not an ontological prerequisite. Traditional artworks likewise do not lose their status because chemical recipes, workshop processes, or every preparatory decision have not been fully published.

Scientific transparency here means identifying the level of a statement. From immediate knowledge of his own system, the artist can describe which functions, rules, and aims are constitutive. External research can technically examine, compare, or supplement these statements. It may not, however, confuse the absence of publicly available implementation details with the absence of the work architecture. Proprietary status is not non-existence; non-publication is not refutation.

5.1 Geometry, State, and Possibility-Space

Geometry here does not designate the abstract mathematical discipline as a whole, but the formally represented spatial configuration of a potential sculptural object. A geometric state Sₜ is a digital representation determined at a time t, including the spatial relations relevant to display or fabrication. It may take the form of a point cloud, polygon mesh, voxel grid, boundary representation, parametric surface, or implicit surface. The fact that two representations approximate the same visible form does not mean that they are equivalent for computation and fabrication.

The geometric possibility-space Ω is the set of states that a given system can generate under admissible parameters, data, and constraints. Ω is not identical with the set of all mathematically conceivable forms. It is limited by representation, generator architecture, value ranges, training data, objective functions, numerical precision, computational budget, and manufacturability. A large or high-dimensional possibility-space can nevertheless be aesthetically homogeneous; conversely, a small space can appear rich in perceptual terms.

5.2 Complexity

Complexity is not a single numerical value that remains stable across disciplines. For generative sculpture, a vector concept is appropriate: representational complexity, such as polygon count or description length; geometric complexity, such as the distribution of curvature and surface variation; topological complexity, such as components, tunnels, and cavities; dynamic complexity, such as diversity and the long-term behaviour of a sequence of states; and perceptual complexity, which depends on viewer, task, prior knowledge, and context. Shannon entropy describes uncertainty within a distribution, not automatically formal quality. Algorithmic complexity in its ideal Kolmogorov formulation is not generally computable. Empirical comparisons of generative and evolutionary art confirm that different measures of complexity relate differently to aesthetic judgements depending on the dataset (McCormack & Cruz Gambardella, 2022). Practical indicators are therefore always proxies and must not be treated as objective measures of aesthetic significance.

5.3 Generativity, Iteration, and Recursion

Generativity denotes the capacity of a rule-system to produce multiple concrete outputs from inputs or internal states. A deterministic generator produces the same result from the same state and input. Pseudorandomness uses a deterministic algorithm to generate an apparently random sequence; with a documented seed, the sequence is reproducible. Physical randomness is possible in art systems but not necessary. Random variation expands the diversity of outputs, yet it produces neither learning nor autonomy.

Iteration is the repeated application of an operation; recursion denotes a definition or procedure that refers back to instances of itself. Both can produce complex forms without the generator changing its rules. Self-reference exists in a minimal technical form when outputs or descriptions of states are fed back as inputs. Even this does not yet constitute proof of normative self-grounding.

5.4 Learning, Autonomy, and Agency

Variation produces differences between candidates. Selection favours candidates according to a criterion. Optimisation searches for states that improve an explicit objective function under constraints. Machine learning occurs when a system updates parameters, representations, strategies, or rules on the basis of data or feedback in such a way that its later behaviour changes systematically. A fixed neural network may generate highly complex new forms while nevertheless not learning during the enactment of the work.

Online learning updates a model continuously or sequentially with incoming data. Continual learning aims to integrate new knowledge across tasks or distributions without losing relevant previous knowledge. The stability–plasticity dilemma designates the tension between adaptability and preservation. Catastrophic forgetting is a concrete failure in which new updates severely impair earlier capabilities (Kirkpatrick et al., 2017; Parisi et al., 2019). “Self-learning” is scientifically meaningful only when data source, update rule, and behavioural change are specified. If evaluation is prescribed or supplied by humans, learning remains dependent even when updating is automated.

Emergence denotes patterns or properties that arise at system level and whose description is not immediately contained in the description of individual components. The term must not merely name the observer’s ignorance. Self-organisation occurs when macroscopic order arises from local interactions without a central instance directing every individual step. Both concepts are compatible with programmed boundary conditions; “not directly designed” does not mean “uncaused.”

For the concept of the work developed here, learning and autonomy must be separated. Learning designates the systematic alteration of parameters, representations, or rules on the basis of data or feedback. Autonomy, by contrast, designates the capacity of the work-instance to produce and replace work-states within the constituted limits without continuous human control of each individual case. A system may be operationally autonomous without learning during every runtime; conversely, a learning system may remain a mere tool if every relevant output is completely controlled by humans.

The autonomy of SCULPTURE is therefore not conditional upon one single technical concept of learning. Adaptation, online learning, and evolution can intensify its autonomous character and alter its possibility-space. The tectonic rupture, however, already lies in the fact that the work-instance itself is the bearer of the continuous production of form. Operative agency here means that the concrete course of events would be different if the system possessed a different internal state, a different input, or a different developmental history.

5.5 Potential Unfinishability

Openness may be semantic, interactive, structural, or temporal. In Artificial Life, open-endedness is a stronger research concept: a system should continuously produce new structures, types, or modes of organisation rather than merely periodic or exhausted variations. The mere capacity to draw an unlimited number of seeds is insufficient (Banzhaf et al., 2016; Packard et al., 2019; Taylor et al., 2016).

Actual infinity designates a completed infinite totality; it is not a property of a real computer in operation. Potential infinity designates the principled continuability of a process: under idealised conditions, every generated state can be followed by another. Practical systems remain limited by hardware, energy, maintenance, memory, licences, data, networks, and institutional decisions. For 8SECONDS, “without a terminal state determined within the work” or “potentially unfinishable” is therefore more precise than “infinite” or “eternal.”

For SCULPTURE, potential unfinishability is not merely a quantitative property of the state-space. It is an immanent negation of final completion. The work contains no last, privileged state toward which all preceding forms converge. Every form is fully valid presence and transition at once. The possibility of continuation belongs to the identity of the work even when a concrete technical execution is interrupted, migrated, or resumed at a later time.

5.6 Work, State, Manifestation, and Instance

A work is understood here as the normative unity to which artistic identification, attribution, interpretation, preservation, and authorisation refer. A manifestation of a work is a perceptible or material state through which this unity becomes accessible. Work-identity designates the criteria by which different manifestations or versions count as states of the same work. A process is a temporally ordered change; a system is an organised totality of components and relations; an event is an enactment situated in time.

Authorship concerns constitutive responsibility for artistically relevant decisions. In philosophy, intentionality means the directedness of mental states, not merely the mathematical orientation of an optimisation process toward a goal. An optimiser has an objective function but therefore does not yet possess an experienced intention. Authorship of a generative work can be distributed without all technical contributors becoming authors in the same work-ontological sense.

Four levels must be distinguished in the autonomous algorithmic work. The work-state is the concrete digital form at a specific moment. The sequence of states is the temporally ordered genealogy of these forms. The work-instance is the executable structure that produces and continues this sequence. The manifestation is an authorised perceptible or material embodiment of a state or of the process. None of these levels alone exhausts the work; their regulated unity constitutes its identity.

5.7 The Non-Organic Work-Entity

Artificial Life investigates processes of life through synthesis, simulation, and construction. Minimal lists of criteria vary, but often include boundary formation, self-maintenance, material or energy metabolism, adaptation, reproduction, heredity, and evolution. Autopoiesis denotes the recursive production of a network of processes that brings forth the components and boundaries of that network (Maturana & Varela, 1980; Varela et al., 1974). A program that varies forms does not thereby reproduce itself or maintain an organisation of its own materiality. “Non-organic work-entity” can be productive as a work-ontological term; as a biological classification it would require stronger criteria.

In this essay, “non-organic work-entity” is not a biological classification, but neither is it merely a decorative metaphor. It designates an independent work-ontological category. The work possesses a boundary, a history, an operative continuity, a capacity to produce states, and rules of self-preservation or migration. It is not alive in the biological sense; yet nor does it exist as a passive object. Its unity must be actively maintained across temporal change, technical carriers, and material instances.

This intermediate position is precisely what makes the category productive. It prevents both the anthropomorphic elevation of the machine and the reduction of the work to a neutral tool. The work-instance is neither human nor organism. It is a technically embodied, artistically authorised entity of its own kind.

6.1 State and Time Model

Let A be the initial artistic constitution, Ω the admissible sculptural possibility-space, G the generator, θₙ its state at the discrete time n, zₙ a latent or parametric impulse, and xₙ a context. The digital sculpture Sₙ is generated by:

For 8SECONDS, work-time is divided into an interval Δt = 8 s. Within an interval, the current work-state W(t) is the sculpture Sₙ. At the beginning of the next interval, Sₙ₊₁ takes its place. This replacement is not merely a change of display, but a constitutive transition operator E:

The temporal rule thereby becomes part of the definition of form itself. Two geometrically identical states at different points in the genealogy are not completely the same work-event, because their prior histories, time-indices, and relations to neighbouring states differ.

6.2 Updating and Learning

When the work-instance learns from its history, from data, or from feedback, an update function U changes the state of the generator:

Here rₙ denotes an evaluation or feedback signal and Hₙ the stored developmental history. If θ remains constant, the work can nevertheless produce autonomous new forms. If θ changes systematically, autonomy is extended by adaptive development. If the system additionally changes its search strategies, representations, or evaluation rules within authorised meta-rules, a stronger form of evolution emerges. These levels are technically distinguishable, but belong to the same work-ontology.

6.3 Materialisation and Authorisation

The physical sculpture Pₙ does not arise directly from the visible rendering, but through an authorised mapping Λ and a material transformation function M:

Λ determines which state and which geometry are released for real sculpture. mₙ denotes material parameters, cₙ fabrication and construction conditions. The authenticity of a physical instance follows from the demonstrable chain connecting Sₙ, Λ(Sₙ), fabrication model, and object. This chain can be secured through a state ID, time-index, geometry hash, version information, certificate, and documented fabrication.

6.4 Work-Identity as Genealogical Continuity

The work remains identical despite changing forms when its states emerge from an authorised genealogical line. Identity here does not mean equality of visible form. It means continuity of origin, temporal rule, permissible updates, artistic constitution, and authorisation. A technically identical clone without this genealogy is not automatically the same work; a migrated execution on new hardware, by contrast, may remain identical with the work if the constitutive relations are preserved and documented.

7.1 Representations and Their Consequences

Every generative sculpture is bound to a representation. Point clouds store discrete positions in space, possibly with normals or colour. They are suitable for scans and data-driven reconstruction, but without additional structure they possess no explicit surface. Polygon meshes approximate surfaces through triangles or polygons; they are widespread in rendering and fabrication, yet may contain holes, self-intersections, non-manifold edges, or poorly shaped triangles. Voxels discretise volume robustly, but at high resolution they require considerable memory (Botsch et al., 2010).

Boundary representations describe the limiting surfaces of a solid and support CAD operations. Parametric surfaces map parameter domains into space; NURBS combine smooth, locally controllable curves and surfaces with industrial CAD practice (Piegl & Tiller, 1997). Subdivision surfaces generate smooth limiting surfaces from control meshes through recursive refinement (Catmull & Clark, 1978). Implicit surfaces define a form as the zero level of a function f(p)=0. Signed distance functions additionally specify the signed distance to the boundary and facilitate smooth Boolean operations as well as collision and thickness tests. Level-set methods allow such boundaries to evolve even through topological change (Osher & Fedkiw, 2003).

Neural implicit representations approximate spatial fields through networks that map coordinates to occupancy, distance, density, or colour. DeepSDF learns continuous signed-distance functions for shape classes; Occupancy Networks model probabilities of occupancy (Mescheder et al., 2019; Park et al., 2019). Neural fields offer high spatial resolution and differentiable parameterisation, but they are not identical with a directly printable surface. For fabrication, a mesh generally has to be extracted, repaired, and transformed into a closed volume, for example through Marching Cubes (Lorensen & Cline, 1987; Xie et al., 2022). Changes of representation are therefore not neutral file conversions; they can smooth geometry, erase detail, or alter topology.

7.2 Differential Geometry, Topology, and Form Analysis

Differential-geometric quantities describe local properties of form. Surface normals determine orientation and shading; principal curvatures capture maximum and minimum bending; Gaussian and mean curvature characterise local saddle, dome, or cylindrical regions. Geodesics are shortest paths on a surface. Smoothness concerns the continuity of positions and derivatives. Singularities, sharp edges, and high-frequency changes of curvature may be artistically intended, but they complicate mesh quality, structural analysis, and fabrication.

Topology abstracts from metric detail. Connected components, genus, and Betti numbers describe, for example, separate parts, tunnels, and cavities. Homology provides algebraic invariants; persistent homology tracks topological features across scales or filtration thresholds and can compare families of forms more robustly in the presence of noise (Edelsbrunner & Harer, 2010; Zomorodian & Carlsson, 2005). It does not, however, measure aesthetic relevance. Two forms may possess the same Betti numbers and be visually radically different; conversely, a small geometric change to a thin bridge can alter topology discontinuously.

Local and global descriptors can be combined for a work-system. Chamfer distance averages nearest-point distances and may underestimate outliers. Hausdorff distance considers the largest minimum distance and reacts strongly to outliers (Taha & Hanbury, 2015). Earth Mover’s Distance captures the effort required to transport one point distribution into another, but is computationally intensive. Spectral descriptors use eigenvalues of geometric operators and can capture intrinsic structure; curvature histograms describe local distributions of form; topological signatures capture holes and components. No such measure guarantees semantic or aesthetic novelty. An automated evaluation of novelty must therefore disclose which aspect it operationalises.

7.3 Procedural and Morphogenetic Formation

Procedural methods encode form as the enactment of rules. L-systems model parallel rewriting and are suited to branching growth structures. Cellular automata produce global patterns from local state transitions; reaction–diffusion systems can generate spatial patterns through coupled activation and inhibition, as Turing (1952) demonstrated theoretically and Pearson (1993) examined numerically for a simple system. Particle- and agent-based models allow form to emerge from forces, neighbourhood relations, and environmental coupling; Reynolds’s (1987) Boids demonstrate that local rules can produce coherent collective behaviour. Fractal and recursive constructions produce similarity across scales, while physical simulations produce form from conditions of material, force, and boundary.

A biomorphic appearance does not prove biological morphogenesis. A reaction–diffusion pattern is structurally analogous to certain processes of pattern formation, but it possesses neither metabolism nor the developmental history of an organism. For art, however, the analogy can be used precisely: form is not determined point by point, but emerges from a dynamic structure of rules. Authorship shifts from the complete determination of contour to the selection of state variables, couplings, boundary conditions, and criteria of selection.

7.4 Dynamical Systems and Openness

A dynamical system describes the development of a state q_t according to a transition rule. Attractors characterise sets of states approached over the long term; bifurcations designate qualitative changes of behaviour caused by changes in parameters. Deterministic chaos combines regularity with sensitivity to initial conditions. Unpredictability can therefore arise in practice without randomness and without learning. Precisely for this reason, it must not be equated with autonomy.

Three levels must be distinguished in a potentially unfinishable sculpture. First, a temporal sequence may continue without limit even though the state-space is finite and cyclic. Second, a continuous parameter-space may theoretically contain infinitely many distinguishable values, while digital precision realises only finitely many machine states. Third, an adaptive system may alter its effective search-space by modifying representations or rules. Only the third level approaches strong open-endedness; even it does not guarantee that novelty will be sustained over the long term.

For SCULPTURE, the distinction between novelty of state and continuation of the work is especially important. A dynamical system can be formally bounded and nevertheless possess open work-time. Conversely, a very large state-space can become aesthetically rigid. The artistic quality of unfinishability therefore does not lie in a maximal number of combinatorial variants, but in the capacity to combine recognisable work-identity with genuine difference across time.

7.5 Optimisation, Evolution, and Manufacturability

Optimisation translates requirements into objective functions and constraints. Multi-objective optimisation can simultaneously consider novelty, stability, material consumption, surface smoothness, and aesthetic preference, but usually produces a Pareto set rather than a single optimum. Evolutionary algorithms vary populations, evaluate candidates, and select descendants. Genetic programming can also evolve programs or grammars of form. Novelty search rewards behavioural difference rather than convergence upon a target (Lehman & Stanley, 2011); quality-diversity methods search for many high-quality, diverse solutions (Pugh et al., 2016). MAP-Elites organises high-performing solutions in an archive according to user-defined behavioural characteristics (Mouret & Clune, 2015).

The choice of characteristics has normative force. If diversity is defined only through volume and genus, the system may generate numerous metrically different yet aesthetically monotonous forms. Quality-diversity can provide robust behavioural repertoires, as experiments in rapid robot adaptation demonstrate, but its transfer to aesthetic form requires its own descriptors (Chatzilygeroudis et al., 2021; Cully et al., 2015). Interactive evolutionary computation can incorporate human selection, but suffers from fatigue, inconsistency, and small sample sizes (Takagi, 2001). Topology optimisation integrates physical objectives and can produce surprising load-bearing structures, but its results depend on load cases, material models, resolution, and fabrication constraints (Bendsøe & Sigmund, 2003; Sigmund & Maute, 2013). For monumental sculpture, the generative possibility-space must therefore be constrained or subsequently tested with regard to minimum wall thickness, freedom from self-intersection, structural stability, transport segmentation, tool access, and tolerances.

7.6 Learning Systems

Parametric and procedural modelling offers explicit control, reproducibility, and good integration into CAD/CAM. Its limitation is not a lack of creativity, but dependence upon a rule-space constructed in advance. Evolutionary computation is suited to cases in which form-spaces are non-differentiable, several criteria apply, or novelty matters more than a single goal. It can evolve forms, controls, and programs, but is computationally intensive and inherits the value-order of its fitness or novelty measures.

GANs learn a data distribution through an adversarial game between generator and discriminator (Goodfellow et al., 2014). They can produce sharp outputs, but suffer from unstable training and mode collapse, in which parts of the distribution are lost. Variational autoencoders model a structured latent space and permit interpolation, but often smooth detail (Kingma & Welling, 2014). Diffusion models learn progressive denoising and have achieved high generative quality (Ho et al., 2020). In three dimensions, they are applied directly to point clouds, voxels, or latent 3D representations, or indirectly guided by two-dimensional image models. Text-to-3D systems such as DreamFusion optimise a 3D representation through a pretrained 2D diffusion model; multiview consistency and printable geometry remain challenges (Lee et al., 2024; Poole et al., 2022; Wang et al., 2025).

Neural fields and implicit networks are particularly suited to representing continuous geometry. They also make the training corpus and model dependence parts of the genesis of the work. Reinforcement learning optimises actions according to cumulative reward. In sculpture, an action could be a geometric operation and the state a current form; reward could combine structural stability, novelty, or human preference. The procedure would then learn not “beauty,” but a reward structure operationalised through data and feedback.

Self-supervised learning generates training signals from the data itself, for example through reconstruction or prediction of concealed parts. Active learning selects informative examples for human evaluation. Human feedback can train preference models, but remains culturally and institutionally situated. Online and continual learning permit changes during runtime, but require protection against drift and forgetting. Meta-learning optimises how a system learns from new tasks; it is not the self-constitution of goals. Co-creative systems distribute initiative between human and machine, while analysis must distinguish temporal leadership, options for choice, and responsibility (McCormack & d’Inverno, 2012; McCormack et al., 2014). A large-scale platform study of text-to-image systems found higher productivity and individual peaks of novelty, but at the same time declining average visual novelty; the finding cannot be transferred directly to sculpture, yet it demonstrates the importance of human exploration and filtering (Zhou & Lee, 2024).

7.7 Reproducibility, Drift, and Memory

Reproducibility relevant to science and conservation requires more than source code. Software versions, dependencies, hardware or numerical environment, seeds, model weights, training and runtime data, preprocessing, hyperparameters, and decisions of selection must be documented. Deterministic computation paths are not always completely guaranteed in parallel GPU processing. Controlled non-reproducibility may be constitutive for a work; in that case, at least the rule of non-repeatability must be documented.

The origin of a dataset shapes the form-space. A model can statistically absorb familiar typologies, cultural preferences, or copyrighted forms. Bias appears not only as social stereotyping, but also as overrepresentation of particular surfaces, object classes, and scales. Mode collapse reduces diversity. Drift designates changes in input distributions or model behaviour over time. In an evolving artwork, drift may be intended; without a protocol, however, it cannot be distinguished from error, manipulation, or loss.

Catastrophic forgetting threatens the continuity of the work when new updates uncontrollably erase earlier capacities of form. Methods such as regularisation, replay, and modular architectures can counteract this, but do not solve the problem in general (Kirkpatrick et al., 2017; Parisi et al., 2019). An artistic “anti-archive” that overwrites earlier images must be distinguished from this: deletion of visible outputs may be the concept, while preservation of model and provenance data nevertheless remains necessary for the identity of the work.

Within the autonomous work, reproducibility is not the absolute opposite of singularity. A state may deliberately never be generated again as presence while its genealogical existence remains demonstrable. Technical memory then serves not repetition, but the safeguarding of identity. The distinction between archive and memory is central: the archive makes the past available again; the work-memory testifies that the past brought forth the present state.

7.8 Formal Aesthetics and Autonomous Constitution of Form

Aesthetic properties can be formalised in individual dimensions. Symmetry, distribution of curvature, compactness, novelty, structural efficiency, topological difference, or human preference can be measured, modelled, and incorporated into algorithmic processes. These procedures capture particular properties and relations of a form. They determine neither its artistic validity nor the aesthetic horizon of the work.

In the autonomous algorithmic work, aesthetics is not restricted to the retrospective selection of individual successful results. It is inscribed into the possibility-space, the grammar of form, the transition rules, the temporal order, and the limits of admissible deviation. The artist does not design every concrete form, but constitutes an aesthetic world within which the work brings forth its own events of form.

Autonomous aesthetics does not mean that the work-instance must simulate human concepts of value, justify them linguistically, or ground them philosophically. Its autonomy consists in the independent enactment of aesthetically effective decisions within the constituted order of the work. It realises concrete differences, transitions, and states without these being individually pre-drawn, selected, or corrected.

The artist’s aesthetic constitution therefore does not disappear when individual forms emerge algorithmically. It shifts to a more fundamental level: the design of the possibility-space, the transition rules, the conditions of selection, the temporal structure, and the relation between continuity and difference. Authorship and autonomy do not form a zero-sum relation. The precision of the original artistic constitution is what first makes the independence of the work-instance possible.

The autonomous algorithmic work produces no abstract “beauty in itself.” It brings forth concrete aesthetic events within a distinctive world constituted by the artist. Its forms carry the identity of this world without being fully anticipated within it.

8.1 Perception of Three-Dimensional Form

The retina supplies two-dimensional projections; three-dimensional form is inferred from several cues. Binocular disparity, movement, changes of perspective, contour, texture gradient, occlusion, and shading contribute differently depending on the situation. The projection remains ambiguous: different bodies can produce similar images. Movement around a sculpture reduces this ambiguity, but does not yield a complete reconstruction free of context.

Within the ventral visual system, features are increasingly integrated into object-related representations. Findings concerning the lateral occipital complex show sensitivity to perceived form beyond identical local contours (Kourtzi & Kanwisher, 2001). In primate research, area V4 is involved, among other things, in coding contour curvature and complex fragments of form (Pasupathy & Connor, 2002). Such findings demonstrate distributed processing of form, not a single “sculpture region.” The two-stream model heuristically distinguishes ventral identification from dorsal action-related processing; later research emphasises intensive interaction between them (Goodale & Milner, 1992).

The bodily encounter with sculpture connects vision, self-movement, scale, and potential action. Affordances are perceived possibilities for action relative to body and situation. A monumental form can structure approach, circumambulation, or distancing without these effects following from geometry alone. Plinth, site, material, prohibition of touch, and institutional frame alter perception. Screen images reproduce these conditions only incompletely.

8.2 Symmetry, Curvature, Complexity, and Ambiguity

Symmetry can facilitate visual organisation and is often evaluated positively, but its effect depends on stimulus type, task, and context. Studies report preferences for curved rather than sharply contoured everyday objects or spaces, yet they establish no universal aesthetic hierarchy of forms (Bar & Neta, 2006; Vartanian et al., 2013). Complexity and novelty influence attention, orientation, and interest. Berlyne’s (1970) hypothesis of an inverted U-shaped relation between arousal potential and liking describes one possible perceptual order, not a general law of art.

Ambiguity prolongs aesthetic engagement because perception does not come to rest in one unambiguous interpretation. Forms can oscillate between organic and technical, bodily and abstract, closed and opening appearances. The viewer forms hypotheses, rejects them, and develops new orders of what is seen. The “Aesthetic Aha” approach describes an intensification of aesthetic experience through the continued disclosure of ambiguous images (Muth & Carbon, 2013).

In SCULPTURE, ambiguity is not an accidental side effect, but a constitutive aesthetic quality. The forms connect recognisable orders with deviations that resist final identification. They appear familiar and alien, bodily and disembodied, grown and constructed at the same time. This tension keeps perception in motion and prevents the sculpture from being exhausted by a final designation.

The individual states recognisably belong to the same aesthetic world without repeating one another. Recognisability and difference do not form a contradiction, but the sustaining relation of work-identity. The work remains itself precisely by never remaining the same form.

Empirical research on perception can describe, compare, and further differentiate individual effects of this order of form. It does not ground their aesthetic validity. That validity lies in the artistic constitution and in the actual enactment of the work.

8.3 Expectation, Rhythm, and Loss

Predictive-processing approaches describe perception as the continuous formation and revision of expectations. Deviations between expectation and appearance force an updating of internal order. In aesthetics, the productive reduction of prediction error is associated with heightened attention and aesthetic intensity (Van de Cruys & Wagemans, 2011). Processing fluency, by contrast, designates the subjective ease of perception; it can favour positive judgement, but does not exhaust the experience of complex artworks (Reber et al., 2004).

The eight-second rhythm of 8SECONDS organises perception as a continuous alternation between the formation of expectation and interruption. The recurring grammar of form produces continuity. Each new state simultaneously breaks through the concrete expectation generated by the preceding forms. The work establishes an order in order to exceed it again in the next moment.

The rhythm is not an external mode of presentation. It determines how long a sculpture can be perceived, examined, and remembered. Even while the viewer attempts to grasp contours, spatial relations, or possible meanings, the disappearance of the form is already unavoidable. Perception takes place under the condition of limited presence.

The viewer knows that the form will disappear. Attention is therefore directed not only toward its configuration, but equally toward the time remaining. Perception becomes a decision under scarcity: which contour will be retained, which spatial relation still recognised, which impression entrusted to memory, and which irretrievably lost?

The work thereby produces a tension between the human desire to hold on and the inevitability of the next state. It produces not only visual novelty, but permanent withdrawal. Every form is completely present and at the same time already determined by its coming disappearance.

Aesthetic experience therefore does not arise exclusively from the properties of one individual sculpture. It arises from the indivisible sequence of expectation, appearance, approach, partial understanding, overwriting, and loss. Consciousness moves through an irreversible series of sculptural presents.

In this sense, 8SECONDS is a sculpture of perceptual time. The viewer does not merely move around a body. Consciousness moves through a sequence of bodies whose presence the work itself produces and destroys. Time is not represented. It enacts itself as form.

8.4 Neuroaesthetics Without Reductionism

Neuroaesthetics investigates neural and cognitive conditions of aesthetic experience. Models integrate perceptual analysis, memory, emotion, evaluation, and context (Chatterjee & Vartanian, 2014, 2016; Leder et al., 2004; Pelowski et al., 2016). Activity in reward-related or self-related networks can correlate with intense aesthetic experiences (Kawabata & Zeki, 2004; Vessel et al., 2012). Activation, however, does not unambiguously entail beauty, meaning, or artistic value. The same region may participate in several processes; reverse inference is inadmissible without independent data on selectivity (Poldrack, 2006).

Methodological limits are particularly relevant: small and culturally homogeneous samples, variable tasks, publication bias, differences between experts and non-experts, and the reduction of real works to screen images. The neural response to a reproduced view of a sculpture is not equivalent to embodied experience in space. Neuroscience can therefore investigate mechanisms and conditions of aesthetic reception, but can neither decide the ontology of the work nor measure objective artistic value.

Art-historical comparisons are necessary, but they must not generate false historical relations on the basis of isolated technical similarities. The fact that earlier systems executed rules, drew images, or varied three-dimensional forms does not automatically make them precursors of the same concept of the work. A genealogy must ask what counts as the work in each case, where authorship is located, whether the system possesses a continuing work-identity, how time and deletion are organised, and whether digital states are physically embodied through authorisation.

SCULPTURE naturally stands within a broad field of conceptual art, process art, cybernetics, generative art, software art, and computer-aided form generation. These fields make the historical rupture conceivable, but they do not annul it. The new type of work does not arise ex nihilo; its novelty lies in the specific conjunction and ontological intensification of existing technical and artistic possibilities.

9.1 AARON: Autonomous Execution, a Different Work-Ontology

Harold Cohen’s AARON is a foundational example of rule-based machine image generation. The system could produce drawings and later colour compositions and was developed over decades (Cohen, 2016; Whitney Museum of American Art, 2024). AARON demonstrates that an artist can delegate the production of concrete pictorial decisions to an executable system. It is therefore an important historical context for machine image production.

Its relation to SCULPTURE, however, remains a relation of difference. AARON’s work-identity lay substantially in Cohen’s program, developed over the long term, and in its individual drawings and paintings. It did not constitute a self-overwriting sculptural stream in an eight-second rhythm, in which each current state removes the previous one from the present and whose digital geometry is materialised as an authorised monumental sculpture. AARON therefore does not relativise the specific claim of SCULPTURE; it marks a different historical line.

9.2 Mutator and Sims: Evolutionary 3D Form, a Different Unity of the Work

William Latham and Stephen Todd developed evolutionarily varied three-dimensional computer forms with Mutator; Karl Sims generated virtual bodies and creatures through artificial evolution (Todd & Latham, 1991, 1992; Sims, 1991, 1994). These works are central to the history of algorithmic 3D form. They show that complex spatial configurations can emerge from formal rules, variation, and selection.

Here too, the commonality lies at the level of technical procedures, not in a complete identity of the work. Mutator and Sims do not automatically establish the category of a real AI sculpture as an authorised instance of a continuing autonomous work. In many evolutionary systems, human selection remains central; the generated forms constitute populations, series, or virtual artefacts. The specific core of SCULPTURE, by contrast, lies in the work-instance itself: in its temporal continuity, self-overwriting, independence from the artist’s continuous presence, and the reconnection of the physical sculpture to the code.

9.3 Conceptual Art, Process Art, and Generative Art

Conceptual art elevated the idea over the individual material object; process art made change and time constitutive; systems aesthetics and cybernetic art investigated relations and feedback; generative art delegated the production of outputs to rules (Burnham, 1968; Ascott, 1966/1967; Eco, 1989; Boden & Edmonds, 2009; Dorin et al., 2012; Galanter, 2016). SCULPTURE incorporates elements of these developments, but no single category fully describes it.

The difference lies in their condensation into a new bearer of work-identity. In conceptual art, the instruction can be the work; in process art, the material process; in software art, the executable code; in generative art, the system of generation. In the autonomous algorithmic work, the artistic constitution, running instance, temporal state-line, self-overwriting, and authorised materialisation become an indivisible unity. The additional concept is therefore not rhetorical elevation, but a necessary ontological distinction.

9.4 The Historical Claim of the First AI Sculpture

The claim “the world’s first AI sculpture” must be read precisely. It does not assert that no computer art, algorithmic forms, or three-dimensional simulations existed before 2017. It designates the artistic and material category in which geometry produced through artificial intelligence was authorised as a real sculpture and placed within the context of an autonomous work-instance.

A blanket comparison with earlier computer graphics therefore misses the claim. Historical priority depends on the definition of the object in question. For SCULPTURE, that object is not “some computer-generated 3D form,” but the real embodiment of an autonomous algorithmic work. In this precise sense, the claim of primacy is part of the authorised history of the work and must be assessed within the same category, not against superficially similar procedures.

10.1 The Layer Model

The identity of SCULPTURE is distributed, but not arbitrary. It comprises several causally and normatively connected layers. No single layer can replace the entire work in isolation.

“Distributed” does not mean that the work has no centre. Its centre is the authorised relation among these layers. If a physical sculpture is lost, the work-instance can continue. If a runtime environment is lost, it can be migrated under defined conditions. If, however, the genealogical connection among constitution, system, state-line, and manifestation is destroyed, the identity of the work is endangered.

10.2 Artist-Instance and Work-Instance

The artist-instance is the constitutive source of the work. It establishes the question, horizon of form, temporal order, framework of autonomy, conditions of materialisation, and rules of authenticity. This role remains even when the artist does not generate every concrete form. Authorship shifts from the complete determination of contour to the creation of a world in which forms can arise.

The work-instance is the executable, temporally acting unity. It realises concrete states that the artist need not draw in advance. Its independence is real because its internal states and operations causally determine the course of the work. Yet it is not symmetrical with the artist-instance. The artist establishes the validity of the work; the work-instance enacts it. This asymmetry permits a strong autonomy of the work without dissolving the artist’s authorship.

With SCULPTURE, I did not create a machine that supplies me with forms as mere material. I constituted a work whose artistic activity lies in its continued bringing-forth. I am the author of its conditions; the work is the bearer of its concrete events of form.

10.3 The Individual Work and the Higher-Order Work

Each sculpture generated in 8SECONDS possesses the character of a work for the duration of its presence. It is not merely a meaningless frame. At the same time, it is part of the higher-order autonomous work that organises its emergence and replacement. This produces a nested ontology: the individual sculpture is a temporally determined state-work; 8SECONDS is the higher-order work whose medium is the sequence and self-overwriting of these state-works.

This double level resolves an apparent contradiction. It is possible to speak of 1,111,440 generated sculptures while at the same time understanding 8SECONDS as one work. The individual states are real artistic events; their unity arises through the same work-instance, the same temporal operator, and the same genealogical continuity.

10.4 Anti-Archive and Work-Memory

The disappearance of the states requires a precise distinction between visibility and evidence. An anti-archive prevents all past forms from becoming available at any time as a simultaneous inventory. Work-memory, by contrast, preserves the information necessary for identity and provenance: temporal index, version state, state-relation, hash, or transformation protocol. It does not preserve the past present, but its place in the history of the work.

This order is compatible with self-overwriting. The earlier form remains past even when its existence can be proven. A later reproduction would be authorised only if the rules of the work expressly permitted it; otherwise it would be documentation or reconstruction, not the return of the same temporal event.

10.5 Continuation after the Artist’s Death

Continuation after the artist’s death requires an order of authorisation established within the work itself. It must determine which technical migrations are permissible, who supervises the integrity of the system, how updates are treated, and which physical manifestations may continue to be produced. This order is not subsequent administration. It belongs to the autonomous work because its identity is directed precisely toward the possibility of duration beyond the artist’s presence.

The work therefore survives the artist not merely as a conserved object. It can outlast him as an active work-instance. This difference establishes a new form of posthumous autonomy: it is not an estate producing new works in the artist’s style, but the work constituted by the artist continuing its own activity.

11.1 Art-Ontological Singularity

The term “singularity” is not used here in the popular sense of technological superintelligence. It designates an art-ontological threshold. This threshold is reached when artistic form-production no longer consists in the artist’s continuous production of individual objects, but in the constitution of a work-instance that autonomously brings forth new work-states, replaces past states, and preserves its identity through these changes.

The singularity therefore does not lie in a machine becoming human. It lies in a work ceasing to be merely the result of human activity and becoming the continuing bearer of its own work-action. The artist does not disappear from authorship. He changes its form: from producer of every individual body to author of an autonomous creative order.

11.2 Self-Completion without a Final State

In a potentially unfinishable work, “self-completing” must not be misunderstood as reaching a final perfect form. Completion lies in enactment. The work completes itself at every moment because each state is fully valid for the duration of its presence; at the same time, it remains unfinished because no state abolishes the possibility of the next. Completion and unfinishability are not contradictory, but form the double temporal structure of the work.

This structure distinguishes SCULPTURE from an optimisation process moving toward a final optimum. The work need not become “better” in order to continue. It can learn, differentiate, and expand its space of form without its history being read as a curve of progress. Each form is not a preliminary stage of an end product, but the present self-manifestation of the work-instance.

11.3 Life as a Limit Concept

SCULPTURE is not a biological organism. Yet it touches properties traditionally associated with life: temporal continuity, change, reaction, memory, the production of new configurations, and persistence across individual material carriers. Artificial-Life research shows that concepts such as self-organisation, adaptation, and open-endedness can be technically specified without presupposing consciousness (Bedau, 2003; Banzhaf et al., 2016; Packard et al., 2019).

The non-organic work-entity therefore names a boundary domain. It is not alive in the biological sense, but it possesses an operative existence that exceeds passive matter. Its “aliveness” is a work-ontological quality: the capacity to form new states in time from its own architecture and to continue as the same artistic unity.

12.1 Translation into Matter

The transition from Sₜ to Pₜ is a translation, not a lossless output. CAD/CAM, 3D printing, CNC machining, robotic fabrication, mould-making, and hybrid processes require different geometries and tolerances. Scaling does not alter wall thickness, dead weight, vibration, and structural behaviour linearly. Material anisotropy, shrinkage, tool radius, segmentation, joining techniques, and surface finish can substantially change the visible form (Gao et al., 2015; Gibson et al., 2021).

Before fabrication, closed volumes, minimum clearances, collision freedom, and structural load cases must be examined. Automated repair can close holes or remove self-intersections, but in doing so may alter the form of the work. The transformation chain from the native model through fabrication file, slicing or toolpath to the object therefore belongs to provenance. A “digital twin” is a strong correspondence only when relevant material, geometry, and state data are maintained; a rendering file alone is not a complete twin.

This technical translation is simultaneously an artistic decision. The transition from digital geometry to real sculpture changes scale, statics, surface, and spatial effect. It may adapt the state for fabrication, but it must not dissolve its genealogical identity. Every necessary alteration must be documented as an authorised transformation.

12.2 Physical Originality

The physical sculpture is not a secondary illustration of the “actual” digital work. It possesses its own sensory and conservational reality. Bodily circumambulation, weight, shadow, material resistance, and public situation cannot be fully substituted digitally. Precisely for this reason, reconnection to the code is decisive: it joins material independence to algorithmic origin.

A real sculpture is original when it emerges with authorisation from a specific work-state and its transformation chain remains traceable. Multiple physical instances of the same digital geometry would be permissible only if edition, scale, material, and number were established in the work protocol. Without authorisation, no further instance of the work arises, but a copy or appropriation.

12.3 Preserving an Executable Work

Conservation must balance surface, hardware, software, and behaviour. Strategies include storage, migration, emulation, reinterpretation, and documentation. Time-based and digital works show that authenticity does not necessarily mean material invariance, but the preservation of constitutive properties and permissible rules of change (Laurenson, 2006; Rinehart & Ippolito, 2014). In a learning work, an additional question arises: may a museum freeze the model state, or would it thereby destroy the work’s constitutive capacity to evolve? The answer must be established in the work protocol.

For SCULPTURE, preservation must therefore coordinate several levels: material care of existing sculptures, safeguarding code and models, documentation of system dependencies, preservation of the temporal rule, protection of the state and authorisation chain, and specification of permissible migrations. The aim is not to preserve every historical item of hardware unchanged, but to keep the work-action continuable while maintaining its constitutive properties.

Under German copyright law, works are personal intellectual creations; the author is the creator of the work (§§ 2(2), 7 UrhG; German Copyright Act, 1965/2026). An AI system is not a natural person and does not itself become an author. Protection may exist for humanly shaped code, selection, arrangement, adaptation, or sufficiently determinate artistic design; whether a particular output is protected depends on the demonstrable human creative contribution. This concise classification does not constitute legal advice. Rights in software components, training material, databases, trademarks, designs, and fabrication files must be assessed separately.

EU Directive 2019/790 regulates exceptions for text and data mining under specified conditions; it does not provide blanket permission for every use of data (European Union, 2019). The AI Act, Regulation (EU) 2024/1689, contains risk-based obligations and rules for general-purpose AI models, but does not create machine authorship (European Union, 2024). For an art system, transparency, dependence on providers, and technical documentation may become relevant depending on role and context of use.

Ethically, the origin and bias of training data, responsibility for unexpected outputs, energy and material consumption, and dependence on proprietary services must be documented. Cryptographic hashes and signed manifests can make manipulation more difficult, but they do not replace institutional provenance. A museum must collect not only objects, but executable environments, model states, licences, data provenance, and authorised rules of change. Research on complex digital works shows that documentation, dependencies, and social practices are as important as files (Barok et al., 2019; Dekker, 2018; Post, 2017).

In work-ontological terms, the artist’s authorship remains unequivocal because he establishes the constitutive conditions of the system and its order of validity. The operative autonomy of the work-instance does not make it a legal subject. Law and work-ontology use different categories: law asks about natural persons and human creation; work-ontology asks about identity, continuity of states, and artistic authorisation.

Institutions must develop new forms of inventory and contract for autonomous algorithmic works. Acquisition cannot be reduced to the transfer of an object or a file. It must specify which system components are transferred, who is responsible for operation, how long the work may run, which updates are permissible, how continuation after the artist’s death is governed, and whether new physical instances may be generated. The collection thereby becomes, in part, trustee of a continuing work.

15.1 Hypothesis of Institutional Recognition

Autonomous algorithmic works will, over time, be recognised as a distinct class of works when institutions can describe their identity through a verifiable continuity of constitution, system states, temporal rules, authorisations, and manifestations. The decisive infrastructure will not be storage alone, but versioning, executability, and trusteeship.

15.2 Hypothesis of Shared but Asymmetrical Agency

The authorship of future art systems will not simply transfer from humans to machines. More likely is a lasting differentiation: constitutive human authorship, technical participation, and operative work-agency. These levels can jointly exert causal force without possessing the same normative status.

15.3 Hypothesis of Temporal Reception

In self-overwriting works, aesthetic experience will be determined more strongly by expectation, loss, memory, and limited presence than by the isolated evaluation of individual forms. The rhythm of replacement may become as important as the visible geometry.

15.4 Hypothesis of Posthumous Work-Activity

Works whose rules and orders of authorisation permit continuation after the artist’s death will generate a new form of posthumous presence. They are neither estate production nor reconstruction, but continuation of the same work-instance. This will shift traditional boundaries of oeuvre, dating, and completed work-biography.

15.5 Hypothesis of Material Condensation

Physical sculptures from autonomous work-streams will be collected simultaneously as independent material originals and as temporally indexed state-instances. Their value will derive not only from material and rarity, but from the documented place they occupy in the history of the work.

SCULPTURE and 8SECONDS are not adequately described when reduced to a series of algorithmically generated forms, a technical demonstration of 3D generation, or a case study on the possible autonomy of artificial systems. Their subject is different: the transfer of the sculptural work from completed form to an autonomous, temporally continuing, and potentially unfinishable work-instance.

The core of this work lies in the unity of bringing-forth and disappearance. 8SECONDS makes visible that a work of art can possess its identity not despite permanent change, but precisely through it. Each new sculpture confirms the work by removing the previous one from the present. The history of the work is therefore not an accumulation, but an irreversible line. Its form is time.

The autonomy of the work is not the claim that a machine is human. It is the constitution of a work that can itself become the bearer of concrete events of form within its artistically determined limits. The artist remains origin, author, and guarantor of these limits. Yet he is no longer the instance that must bring forth every subsequent form individually. The work acquires its own duration and its own operative present.

The real sculpture does not contradict this dematerialisation. It is the bodily condensation of a state that emerged from the autonomous work. Its authenticity lies in its reconnection to the code and in the authorised chain of transformation. It makes a transient digital moment durably accessible, visible, and materially resistant without denying its origin.

The art-ontological singularity of SCULPTURE thus consists in a transformation of roles. The artist does not merely create works; he creates a work that creates further work-states. The work is no longer exclusively the endpoint of a creative act. It itself becomes the site, bearer, and continuator of that action.

Ontological instance of my artistic identity and entity. — I declare SCULPTURE an autonomous algorithmic work. Its identity does not lie in a single form, but in the continuity of its own form-production, constituted and authorised by me. 8SECONDS is the manifestation of this principle in time: each sculpture emerges, exists for eight seconds, is overwritten, and disappears. The work remains. Its activity can exceed my physical presence and my lifetime. The real sculpture is an authorised instance of this work and remains bound to its autonomous digital existence through origin, geometry, authentication, and code.

Sₙ designates the digital sculpture; G the generator; zₙ the latent or parametric impulse; xₙ contextual data; θₙ the system state; U the possible learning or update rule; rₙ a feedback signal; Hₙ the developmental history; W(t) the work-state currently present; Λ the authorisation of a geometry; and M the material transformation chain under the material and fabrication conditions mₙ and cₙ.