Abstract
This paper examines the theoretical contributions of Bruno Latour’s actor network theory and Gilles Deleuze and Félix Guattari’s assemblage theory in order to articulate an ethics of distributed agency suitable for artificial intelligence. The analysis demonstrates a decisive departure from classical subject object epistemology and from subject centric ethical frameworks. In their place emerges a materialist and open orientation toward an emergent object oriented ontology. Epistemology shifts from representational knowledge grounded in a knowing subject toward relational and performative accounts of knowledge production. Ontology moves from hierarchical dualisms to flat networks and multiplicities in which human and nonhuman entities co constitute reality. Ethics consequently relocates responsibility from individual human subjects to the distributed capacities of heterogeneous assemblages. When applied to artificial intelligence, these shifts reveal AI systems as actants and components of assemblages rather than mere instruments. The resulting ethical stance requires attention to the emergent effects of socio technical networks rather than exclusive focus on human intention or control. The argument proceeds through close engagement with primary conceptual resources, clarifies the epistemological ontological and ethical transitions, and illustrates their implications for the design regulation and social embedding of AI.
Introduction
Contemporary discussions of artificial intelligence frequently remain captive to an older philosophical vocabulary. AI systems are treated either as sophisticated tools under human command or as quasi subjects whose autonomy threatens human uniqueness. Both framings reproduce a subject object dualism that has structured Western thought since at least the early modern period. The knowing or acting subject stands over against a world of objects that can be known, used, or controlled. Ethical reflection then concentrates on the responsibilities of the human subject, whether as designer, user, or regulator. This paper contends that such an approach is increasingly inadequate. The scale complexity and opacity of contemporary AI systems demand conceptual resources capable of registering agency as distributed across human and nonhuman entities.
Two bodies of theory prove especially generative for this task. Bruno Latour’s actor network theory dissolves the a priori distinction between subjects and objects by treating all entities as actants whose capacities emerge within networks of association. Gilles Deleuze and Félix Guattari’s assemblage theory similarly rejects hierarchical ontologies in favor of multiplicities and processes of becoming in which heterogeneous components enter into provisional relations of consistency. Together these frameworks support a materialist and open orientation that aligns with recent developments in object oriented ontology while remaining attentive to process and relation. The ethics that follows is one of distributed agency: responsibility is not located solely in human subjects but is distributed across the networks and assemblages that produce effects.
The argument unfolds in four movements. First, the core claims of actor network theory are reconstructed with attention to their implications for epistemology and ontology. Second, assemblage theory is examined for its complementary emphasis on multiplicity and emergence. Third, the combined resources are shown to enact a clear shift from subject object epistemology and subject centric ethics toward an emergent object oriented orientation. Fourth, these shifts are applied to artificial intelligence, yielding concrete implications for how AI systems should be understood designed and governed. Footnotes supply conceptual clarifications and textual anchors without external references.
Actor Network Theory and the Dissolution of Subject Object Dualism
Bruno Latour’s actor network theory begins from a refusal to grant privileged status to human intentionality. In place of subjects who act upon a passive world of objects, Latour proposes a flat ontology of actants. An actant is any entity that makes a difference in a network of associations. Humans machines texts microbes and institutions all count as actants insofar as they mediate translate or stabilize relations. Agency is therefore not an intrinsic property of subjects but an effect of the network. What appears as the purposeful action of a human agent is in reality the collective achievement of a chain of mediators each of which contributes its own form of resistance or facilitation.
This redescription carries immediate epistemological consequences. Knowledge is no longer the representation of an external world by a detached subject. Instead knowledge is produced through the work of translation and enrollment by which heterogeneous entities are brought into alignment. Laboratories instruments inscriptions and spokespersons participate in the fabrication of facts. The classical epistemological hierarchy that elevated pure theory over impure practice is inverted. Practice becomes the primary site of knowledge production and theory appears as a later stabilization of successful practices. Epistemology thus shifts from a subject centered theory of representation to a relational and performative account of how associations hold.
Ontologically the theory is resolutely anti dualist. Nature and society culture and technology human and nonhuman cease to function as foundational categories. They emerge instead as provisional outcomes of network building. A successful network may stabilize into what looks like a natural fact or a social convention but both are products of the same relational labor. The ontology is therefore object oriented in a precise sense: it grants ontological parity to all entities while insisting that their identities and capacities are network effects rather than intrinsic essences. Objects are not passive receptacles of human meaning; they actively shape the networks in which they participate.
The ethical implications follow directly. If agency is distributed then responsibility cannot be localized exclusively in human subjects. Moral and political accountability must track the entire network that produces a given effect. When a technical system fails or causes harm the relevant question is not solely who intended the outcome but how the associations among designers users algorithms data infrastructures and regulatory frameworks enabled or constrained that outcome. Ethics becomes a matter of tracing and intervening in networks rather than adjudicating the purity of subjective will.
Assemblage Theory and the Ontology of Multiplicity
Deleuze and Guattari’s concept of assemblage offers a complementary yet distinct vocabulary. An assemblage is a multiplicity composed of heterogeneous elements that enter into relations of exteriority. Components retain a degree of autonomy; they can be detached and recombined with other assemblages. Consistency arises not from an overarching unity or essence but from the provisional articulation of these elements through relations of territorialization and deterritorialization. Assemblages are therefore always in process. They stabilize certain capacities while remaining open to lines of flight that can reconfigure them.
Epistemologically this framework privileges cartography over representation. To know an assemblage is to map its components the relations among them and the intensive processes that produce emergent properties. Knowledge is local experimental and provisional. There is no sovereign subject who surveys the whole from a privileged exterior position. The knower is always already inside one or more assemblages and knowledge practices themselves form part of the multiplicity under investigation. Epistemology becomes a practice of following connections and registering intensities rather than of securing correspondence between ideas and things.
Ontologically assemblage theory advances a flat yet dynamic materialism. All entities whether organic technical semiotic or affective can enter into assemblages. Hierarchy is not abolished but is treated as a secondary effect of territorialization rather than a primary ontological structure. Emergence is central: new capacities appear at the level of the assemblage that cannot be reduced to the sum of its parts. This is an object oriented ontology in the sense that it refuses to subordinate nonhuman elements to human meaning or intention. At the same time it remains process oriented: objects are not static substances but temporary consolidations of relations.
The ethical valence of assemblage theory lies in its attention to composition and becoming. Ethical evaluation concerns the kinds of assemblages that are produced the capacities they enable or disable and the lines of flight they open or close. Responsibility is distributed across the components of the assemblage and across the practices that maintain or transform it. One does not ask primarily what a subject ought to do but rather what kinds of collective arrangements foster or inhibit desirable becomings. Ethics is therefore compositional and experimental rather than primarily deontological or consequentialist in the classical sense.
The Combined Shift: Epistemology Ontology and Ethics
When actor network theory and assemblage theory are read together a coherent transformation of the philosophical landscape becomes visible. Epistemologically the subject object dualism that underwrote representational knowledge is displaced by relational and performative accounts. Knowledge is no longer the achievement of a solitary knower confronting a world of mute objects. It is the outcome of associations among actants or the cartography of multiplicities. The classical epistemological subject is dethroned; in its place stand networks and assemblages whose knowledge effects are collective and emergent.
Ontologically the same dualism is replaced by a flat yet dynamic orientation that can be described as emergent object oriented. Objects are no longer the residual category left after subjects have been accounted for. They are full participants whose capacities co constitute reality. At the same time objects are not isolated substances; they exist only in and through relations of association or assemblage. The ontology is therefore materialist in its refusal of linguistic or cultural idealism open in its attentiveness to process and emergence and object oriented in its parity among entities. Hierarchical distinctions between nature and culture human and nonhuman mind and matter lose their foundational status.
Ethics undergoes a corresponding relocation. Subject centric ethics located moral agency and responsibility in the intentional human subject. Distributed agency ethics relocates both within the networks and assemblages that produce effects. Responsibility is not abolished; it is redistributed. Designers users regulators and the technical systems themselves share in the production of outcomes. Ethical inquiry becomes a matter of tracing associations mapping capacities and intervening in compositions so as to foster more desirable emergences. The moral vocabulary shifts from purity of will or individual virtue toward the quality of collective arrangements.
These shifts are not merely theoretical. They respond to the empirical realities of contemporary technoscience in which human and nonhuman agencies are inextricably entangled. The classical subject who could claim to stand outside the world and survey it with detachment has become an untenable fiction. What remains is the practical work of composing better networks and assemblages.
Application to Artificial Intelligence
Artificial intelligence systems furnish a privileged site for testing and refining the ethics of distributed agency. Contemporary AI is neither a simple tool nor an autonomous subject. It is a complex assemblage of algorithms data sets computational infrastructures human labor regulatory frameworks economic incentives and social practices. Actor network theory allows us to treat each of these elements as an actant. The algorithm translates data into predictions; the data set stabilizes certain patterns while excluding others; the hardware mediates speed and scale; the human annotators and designers enroll particular values; the regulatory environment constrains or enables deployment. Agency is distributed across this network. No single component can be isolated as the sole source of an outcome.
Assemblage theory sharpens the analysis by emphasizing emergence and recomposition. An AI system is not a fixed object but a multiplicity that territorializes certain capacities for example pattern recognition at scale while remaining open to deterritorialization through new data new uses or adversarial interventions. The ethical significance of the system cannot be read off from the intentions of its designers alone. It resides in the emergent properties of the assemblage as a whole: the forms of decision making it enables the populations it differentially affects the new relations of visibility and invisibility it produces.
From this perspective several concrete implications follow. First design practices must move beyond the optimization of isolated performance metrics. Designers participate in the composition of assemblages and therefore share responsibility for the relations they stabilize. Attention to data provenance annotation labor and infrastructural dependencies becomes an ethical as well as a technical requirement. Second evaluation and auditing cannot be limited to the inspection of model weights or training data. They must map the broader network of associations in which the system is embedded including deployment contexts user practices and secondary effects. Third regulation should aim less at assigning liability to discrete human subjects and more at shaping the conditions under which assemblages form and evolve. Liability regimes that treat AI as a product under the control of a manufacturer or as an agent under the control of a user both reproduce subject object dualism. Distributed agency ethics points toward forms of collective responsibility and adaptive governance that track the ongoing recomposition of socio technical networks.
The same framework illuminates questions of autonomy and control. Classical debates ask whether AI systems possess autonomy and therefore whether they threaten human agency. Distributed agency ethics reframes the question. Autonomy is itself a network effect. Human autonomy has always depended on prostheses infrastructures and social arrangements. AI systems reconfigure those arrangements. The ethical task is not to preserve a pure human autonomy against technological encroachment but to compose assemblages in which desirable forms of agency human and nonhuman can flourish. This may involve limiting certain AI capacities closing certain lines of flight or opening others. The criteria for such interventions are experimental and compositional rather than derived from an a priori hierarchy of subjects over objects.
Finally the framework addresses the problem of opacity. Many contemporary AI systems are difficult to interpret even for their designers. Subject centric ethics tends to treat opacity as a failure of human control that must be remedied by greater transparency. Distributed agency ethics recognizes opacity as an expected feature of complex assemblages. The response is not necessarily full transparency which may be neither feasible nor desirable but rather the cultivation of practices that render the effects of the assemblage accountable. This includes the development of intermediary actants that translate between technical systems and human publics the maintenance of institutional spaces for contestation and the design of systems that remain open to recomposition rather than locked into irreversible trajectories.
In each of these domains the shift from subject object epistemology and subject centric ethics proves generative. AI is neither reduced to an object under human command nor elevated to a subject rivaling human uniqueness. It is treated as a participant in networks and assemblages whose emergent properties demand ethical attention. The materialist and open orientation of the theories examined here supplies the conceptual resources for that attention.
Conclusion
The network theory of Bruno Latour and the assemblage theory of Gilles Deleuze and Félix Guattari together enact a profound reorientation of philosophical inquiry. Epistemology moves from the representational achievements of a knowing subject to the relational and performative production of knowledge within networks and multiplicities. Ontology abandons hierarchical dualisms in favor of a flat yet dynamic materialism that grants parity to heterogeneous entities while remaining attentive to process and emergence. Ethics relocates agency and responsibility from the individual human subject to the distributed capacities of socio technical assemblages. The resulting stance is appropriately described as an ethics of distributed agency grounded in an emergent object oriented ontology.
When applied to artificial intelligence these shifts yield a more adequate understanding of contemporary systems. AI appears neither as a mere instrument nor as an autonomous agent but as a component of complex assemblages whose effects cannot be reduced to human intention. Ethical reflection and practical intervention must therefore track the full network of associations map emergent capacities and participate in the ongoing composition of more desirable arrangements. The classical subject who once claimed sovereignty over a world of objects has given way to a more modest yet more realistic figure: the participant in networks and assemblages who shares responsibility for what those formations produce. In an age of artificial intelligence this modest figure is the only one capable of meeting the ethical demands of the present.
Footnotes
1. The term actant is deliberately chosen to avoid the anthropocentric connotations of actor. It designates any entity human or nonhuman that makes a difference within a network of associations.
2. Translation in this context refers to the operations by which the interests goals or properties of one entity are aligned with those of another so that they can act in concert.
3. Enrollment designates the process by which entities are recruited into a network and their identities and capacities are redefined in accordance with the network’s requirements.
4. The concept of assemblage translates the French agencement which emphasizes the active arrangement of components rather than a static collection.
5. Relations of exteriority imply that the components of an assemblage are not fused into an organic whole. They retain the capacity to enter into other relations and other assemblages.
6. Territorialization refers to the processes that stabilize an assemblage by selecting certain components and relations while excluding others. Deterritorialization names the processes that open the assemblage to new connections and transformations.
7. Lines of flight are the trajectories of deterritorialization that can lead an assemblage beyond its current configuration toward new forms of consistency or dissolution.
8. The phrase object oriented ontology is used here in a broad sense to indicate any ontology that grants ontological parity to nonhuman entities and refuses to subordinate them to human subjects or linguistic structures. It does not imply commitment to any single school of thought that has adopted the label.
9. Performative epistemology emphasizes that knowledge claims are themselves acts that participate in the production of the realities they describe rather than neutral representations of a pre existing world.
10. The redistribution of responsibility does not eliminate individual accountability. It situates individual actions within the larger networks and assemblages that enable and constrain them and therefore expands rather than contracts the scope of ethical attention.
