\chapter{Introduction} \glsresetall The rapid integration of advanced \gls{ai} into the fabric of social life signals a profound shift in how power is organised and exercised. This transition finds a powerful precursor in \citeauthorfull{deleuze1992a}'s \parencite*{deleuze1992a} short but influential essay \citetitle{deleuze1992a}, where he identifies a move away from traditional, enclosed institutional formations towards a more fluid and pervasive form of power. %%%%%%This regime is characterised by the rise of digital transformation and computational infrastructures that operate through continuous, decentralised networks. Power operates less through fixed structures and more through the circulation of data, statistical inference, and predictive techniques that track and influence behaviour in real time. Control societies are thus defined by their reliance on information technologies that transform how individuals are monitored, classified, and engaged, fundamentally reshaping the conditions under which human subjectivity is formed. In control societies, power is enacted through digital and computational infrastructures that operate by circulating data, generating predictions, and shaping behaviour in real time. Such infrastructures no longer simply regulate individuals from the outside; they participate directly in producing the very conditions under which subjects perceive, decide, and act. It is within this transformation that the present study situates itself, examining how contemporary \gls{genai} systems extend this machinery by functioning as meaning-making entities reshaping the conditions under which human subjectivity is formed. Recent years have witnessed substantial advancements regarding computational infrastructures, especially in the field of \gls{ai}. Innovations in \glspl{nn} and \glspl{dnn} have enabled increasingly sophisticated solutions to \gls{nlp} tasks such as text classification and topic modelling, with applications spanning search engines, social media feeds, streaming platforms, employment procedures, and surveillance infrastructures. These applications were largely predictive, focused on analysis, relevance association, and personalisation, resembling the machinery Deleuze has outlined. Yet, recently a new paradigm has emerged in the form of \gls{genai}. What began in the 1950s \parencite[4]{cao2023a} as a relatively marginal pursuit within the field of \gls{nlp}, now underpins systems capable of generating novel text, images, and code, translating human prompts into coherent outputs by drawing on vast reservoirs of raw data. Far from being mere technical novelties, and unlike earlier models limited to prediction, they have become cultural phenomena. Models like ChatGPT, Stable Diffusion, and other successors are now active participants in knowledge production, communication, and cultural creation. The advancements signal a fundamental reconfiguration of \gls{ai}’s epistemic role and operational qualities: from predictive instruments to generative infrastructures that govern and (re-)produce knowledge itself. By functioning as meaning-making entities \parencite{dishon2024}, \gls{genai} systems not only reshape institutional processes central to shaping human subjectivity \parencite{mackenzie2021}, but also fundamentally challenge established sociotechnological imaginaries of artificial agents. Particularly, their \textit{transformer-based} architectures and novel \textit{attention} mechanisms enable \gls{genai} models to draw connections across widely separated linguistic and visual patterns \parencite[see][]{montanari2025}. This capacity allows them to generate coherent outputs that increasingly mediate social reality and shape the interpretative frameworks through which subjects navigate the world. At stake is not only how such systems generate information and reorganise meaning, rather a decisive evolution, one where power operates not by shaping what subjects see, but by generating the very fabric of what can be seen and thought, thereby challenging the possibility of critique at its source. It is precisely this enclosure of the imaginative and interpretive terrain that demands a re-theorisation of resistance. Yet, the critical theory of \gls{ai} currently lacks a robust formulation of critique and resistance that is grounded in a technical analysis of these systems. In this thesis, I build on critical perspectives from political theory and the philosophy of technology to interrogate the institutional, epistemological, and political implications of contemporary \gls{genai} systems by analysing their architectural structures in depth. I aim to address the lack by formulating a theory of resistance that works through and with these generative infrastructures to counter or divert specific tendencies in processes of subjectification central to the machinery of control. The task is tackled by pursuing three main prospects after situating \gls{genai} within Deleuze's control society: a technical analysis of the mechanisms giving life to these architectures, a discussion of the most prominent debates around \gls{genai}, and an articulation through \gls{dg}'s broader project \enquote{Capitalism \& Schizophrenia} (see \cite*{deleuze1983} and \cite*{deleuze1987}) and its unique contribution to revolutionary theory. \section{Charting a Manifold: Research Question \& Motivation}\label{sec:question} One could argue that \gls{ai} is \enquote{no longer an engineering discipline} \parencite[206]{dignum2023}, if indeed it ever truly was. Each advancement in the design of systems that transform data\sidenote{Read: traces of the human past. See \citeauthor{denton2021} \parencite*[]{denton2021} for a detailed analysis of datasets, \citeauthor{jones2023} \parencite*[]{jones2023} for a concise critical account of their role, and \citeauthor{jones2025} \parencite*[]{jones2025} for a more extensive treatment.} into interpretations of the world simultaneously reconfigures relations of power and knowledge. Algorithmic capacities for decision making, information management, content creation, and narrativisation turn \gls{ai} systems into political entities. Their nature, formation, and the functions they perform must therefore be examined as parts in the machinery of power, or in a more comprehensive sense as \glspl{dispositif}: dynamic arrangements through which technical architectures, institutional practices, and epistemic frameworks are articulated to produce, distribute, and regulate knowledge. To look under the hood of these sociotechnological \glspl{dispositif} is to discover less about their internal cogs and gears than about the power structures they reproduce and sustain. What matters, then, is not simply the technical functioning of \gls{ai} but the ways in which it embeds itself into everyday life, binding knowledge to governance and influencing how individuals come to understand themselves and their world. This move requires a conceptual vocabulary that accounts for the imbrication of knowledge and power, and Foucault’s formulation of the Power/Knowledge nexus provides precisely such an entry point \parencite[109–134]{foucault1980}. \begin{quote} [O]ne often hears people saying that power is that which abstracts, which negates the body, represses, suppresses, and so forth. [...] what I find most striking about these new technologies of power introduced since the seventeenth and eighteenth centuries is their concrete and precise character, their grasp of a multiple and differentiated reality. [...] It becomes a matter of obtaining productive service from individuals in their concrete lives. And in consequence, a real and effective 'incorporation' of power was necessary, in the sense that power had to be able to gain access to the bodies of individuals, to their acts, attitudes and modes of everyday behaviour. Hence the significance of methods like school discipline, which succeeded in making children's bodies the object of highly complex systems of manipulation and conditioning. But at the same time, these new techniques of power needed to grapple with the phenomena of population, in short to undertake the administration, control and direction of the accumulation of men[.] — \cite[124-125]{foucault1980} \end{quote} \Glspl{dispositif} function to materialise the \textit{reality} of a power structure. Foucault describes this as the operation of \enquote{biopower}, a form of power with a specific \enquote{technology} for managing populations at large by specifically focusing on disciplining human behaviour. Biopower operates through procedures, technologies, and routines that make life measurable and regulatable, ranging from demographic statistics and health campaigns to public education infrastructures, enabling power to gain knowledge on its subjects, to \textit{gain access to bodies}\sidenote{In critical theory, the concept of \textit{body} is usually, though not exclusively, taken to mean the human body, but also understood as a surface of social inscription, always situated in its social context. While philosophy, since Descartes, often mistrusted the body as a source of impulses, Spinoza insisted on asking what a body can do. A major shift came with Merleau-Ponty’s phenomenology, which foregrounded embodied perception, and with feminist theory, beginning with Beauvoir, which exposed the neglect of sexual difference. Later thinkers such as Butler challenged the distinction between natural and cultural bodies, and Haraway reconceived the body as cyborg, blurred with animals and machines. Politically, feminism introduced the notion of \enquote{body politics,} while cultural studies analysed the body as a site of media representation and social anxiety. Foucault’s concepts of discipline and biopower remain central, highlighting how bodies are inscribed and governed within regimes of power \parencite[see][98–99]{buchanan2018}.}. Biopolitical formation of \glspl{dispositif} operationalises the knowledge over bodies in order to produce subjectivities suited to sustaining its specific socioeconomic order. In Foucault’s account, this mode of power that embeds itself directly into the conditions of life is the essence of what he names \enquote{disciplinary societies}. Institutions such as schools, hospitals, factories, and prisons as \citeauthor{deleuze1992a} \parencite*[]{deleuze1992a} argues, \enquote{moulded} individuals through enclosure, surveillance, and routine, producing docile subjects whose bodies and conduct could be optimised (see \cite[]{foucault1995}). However, in \citetitle{deleuze1992a} \parencite*[]{deleuze1992a}, he identifies an emerging transformation: from discipline to control. Whereas disciplinary regimes moulded individuals within enclosed institutions, control societies operate through continuous processes that \enquote{modulate} individuals by extracting and acting on data traces at the level of what Deleuze calls the \enquote{dividual}; subjects fragmented into actionable data particles rather than addressed as unified individuals. Discipline segmented bodies in space and time; control turns their characteristic attributes and behavioural patterns into a subject of analysis across digital networks, data flows, and feedback loops, shaping subjectivities through ubiquitous computational processes rather than architectural confinement. Deleuze’s text offers a powerful lens for analysing the late turn of capitalism that Foucault had already begun to trace in his account of neoliberal governmentality (see \cite[]{foucault2008}). In this formulation, the economy ceases to be one domain among others with its own rationality; it comes instead to encompass the entirety of human action, insofar as all behaviour can be recast as the allocation of scarce resources toward competing ends. What matters is no longer the reconstruction of a mechanical logic, but the analysis of conduct itself as governed by a specific economic rationality \parencite[197]{Lemke2001}. The machinery of control, with its computational infrastructure, articulates a future in which such rationalities are operationalised through statistical inference, acting directly on bodies: \begin{quote} Types of machines are easily matched with each type of society—not that machines are determining, but because they express those social forms capable of generating them and using them. [...] capitalism is no longer involved in production [...] [M]arketing has become the center or the \enquote{soul} of the corporation. We are taught that corporations have a soul, which is the most terrifying news in the world. The operation of markets is now the instrument of social control and forms the impudent breed of our masters. Control is short-term and of rapid rates of turnover, but also continuous and without limit, while discipline was of long duration, infinite and discontinuous. Man is no longer man enclosed, but man in debt. — \cite[5]{deleuze1992a} \end{quote} Deleuze radicalises Foucault’s insight by linking neoliberal rationality directly to the machinic infrastructures of late capitalism. The economy does not simply subsume all human action under its logic, but does so through the operational codes of marketing, circulation, and debt, which are engraved in the mechanism of control. Unlike the fixed enclosures of disciplinary institutions, control is continuous, adaptable, and dispersed across networks, markets, and micropolitical dimensions of everyday life. Deleuze’s fragmentary diagnosis thus leaves us with an uncanny resonance with today’s sociotechnological formations. His concept of control societies has since been expanded in multiple directions: as the organising substance of \enquote{Empire} \parencite[]{hardt1998}; as a framework for analysing digital platforms and sociotechnological imaginaries \parencite[]{galloway2001, rouvroy2012, raunig2016}; through studies of surveillance and \enquote{dataveillance} \parencite[]{Haggerty2000, Krasmann2017, Cheney2017}; and more recently, through investigations into the institutional roles of emerging technologies and their effects on agency \parencite[]{mackenzie2021, amoore2024}. As \citeauthorfull{hardt1998} \parencite*[139]{hardt1998} observes, \enquote{Deleuze says remarkably little about the institutional architecture of control societies}. Some critics even question whether there is a substantive transition from discipline to control at all (see e.g. \cite{kelly2015a}), and one might equally doubt whether Deleuze’s sketch is an adequate starting point for analysing the social impact of \gls{genai} models. My research proceeds, however, from the conviction that Deleuze’s account of control societies can be read as a \textit{matrix}\sidenote{In linear algebra, a matrix represents a function: every linear transformation $T: \mathbb{R}^n \to \mathbb{R}^m$ can be expressed as multiplication by a matrix $A$, such that $T(x) = Ax$. A matrix therefore, encodes the rule of transformation rather than the fate of individual inputs, mapping one structured field into another \parencite[401–420]{strang2016}.}, not in the sense of a fixed architecture but as the representation of a transformation. What matters is less where each input is mapped than the direction and structure of change itself. In this sense, the Postscript charts the eigenvectors\sidenote{An eigenvector of a matrix $A$ is a nonzero vector $\mathbf{v}$ such that $A \mathbf{v} = \lambda \mathbf{v}$ for some scalar $\lambda$. Geometrically, the transformation preserves the direction of $\mathbf{v}$ while scaling it by $\lambda$, showing how certain tendencies remain invariant even as magnitudes change \parencite[288–304]{strang2016}.} of capitalism: the invariant tendencies along which social investment is regulated. This framing provides a valuable starting point for situating the social role of \gls{genai} models. Despite the rich continuation of the literature, scholarship on control societies has often fallen short at critical junctures. It either \begin{enumerate}[label=\roman*.] \item fails to develop a theory of resistance adequate to the constellation of \glspl{dispositif}, leaving Deleuze’s brief gestures towards lines of flight\sidenote{The Deleuzoguattarian term \textit{line of flight} (ligne de fuite) refer to the formation that diverges from the established status-quo's grip, a path that enables parts of a system to break away, reconfigure, or deterritorialise existing structures of power, meaning, or order; \enquote{an infinitesimal possibility of escape} \parencite[]{fournier2014}. A claim \enquote{that social formations are defined not by their internal contradictions, but by what escapes them} \parencite[14]{thornton2018}.} largely under-theorised; \item avoids engaging with the technical machinery of the \glspl{dispositif}, whether in analysing those described in the definition of control societies or in considering whether contemporary computational infrastructures may already be surpassing them;\sidenote{This point also opens a further discussion: when, if ever, do we move beyond control societies? Is there a field outside them, or does the very ambiguity of Deleuze’s formulation prevent us from clearly discerning their borders?} \item and, whether for temporal reasons or due to particular disciplinary focus, neglects \gls{ai} systems as a primary subject of analysis.\sidenote{See \citeauthor{galloway2004} \parencite*[]{galloway2004} for a work that addresses the other points above, but was published prior to the current surge of developments in \gls{ai}. See also Section~\ref{sec:literature} for a fuller discussion of Galloway’s contribution.} \end{enumerate} Furthermore, those partly introduced attempts to extend Deleuze’s work reveal inconsistencies that sit uneasily with the broader Deleuzian theory. The \textit{Postscript} may be thin in theorising resistance or outlining concrete paths of divergence from the constellation of control societies. Yet Deleuze’s wider project, above all \enquote{Capitalism and Schizophrenia} with Félix Guattari (see \cite{deleuze1983, deleuze1987}), is anything but devoid of revolutionary thinking; quite the contrary, it is constructed around it. Crucially, Deleuze insists that every established power structure already harbours within itself the elements of a resistance against it (see especially the final chapter of \citetitle{deleuze1983} \cite*[273–383]{deleuze1983}). This distinction becomes particularly salient when we turn to contemporary algorithmic \glspl{dispositif} such as \gls{genai}. The immanence of resistance in Deleuzoguattarian theory provides a critical alternative to the dominant perspectives on these technologies: it allows us to diverge away from both the uncanny optimism of techno-solutionism and the pessimistic defeatism that frames them as mere tools of techno-feudalist formations. Consequently, my aim is to investigate how these algorithmic \glspl{assemblage} are not merely instruments of modulation but also sites of unforeseen potential. The central question is whether we can move beyond the sparse account of resistance in the \textit{Postscript} to theorise new forms of divergence, even when control seems as subtle and encircling as Deleuze describes. It is precisely here that I turn to \gls{dg}’s \enquote{Capitalism and Schizophrenia}. Their work provides the theoretical apparatus that the \textit{Postscript} lacks, enabling an analysis of \gls{genai} simultaneously along the mechanism of control and also as a terrain where resistance can be reconfigured. This leads to the study's guiding concern: \begin{quote} \textbf{\textit{RQ}}: How are critique and resistance in today’s sociotechnological constellation to be (re-)theorised through \gls{dg}’s project \enquote{Capitalism and Schizophrenia}, by analysing the emergence of \gls{genai} in relation to the institutional framework of control? \end{quote} The nature and relationship between critique and resistance is hardly new in \gls{dg}’s work. Yet the central concern of this study starts from a point that can be best described via the question raised by \citeauthorfull{rouvroy2012} \parencite*[]{rouvroy2012}: is critique still possible after the \enquote{computational turn}? \citeauthor{rouvroy2012} describes a regime in which decision-making no longer depends on politics, law, or social norms, but on data-driven inference, a rejection of modern rationality (see \cite[2–14]{rouvroy2012}; see also \cite{rouvroy2020}), constituting the basis of her notion of \enquote{algorithmic governmentality}. In such a setting, critique risks being bypassed by predictive infrastructures that act before subjects can intervene. It is precisely here that \citeauthorfull{mackenzie2018} \parencite*[]{mackenzie2018} offers a different perspective, reframing critique not as an obsolete practice but as a necessary and adaptable one within these new sociotechnological formations: \begin{quote} Critique as a practice of stepping beyond the limits of possible knowledge, for some, came to replace the idea that critique should establish the limits of legitimate knowledge. [...] we take it that the status of critique in control societies can be positively reframed and that it is necessary to do so if we are to ward off the dangers of a conservative embrace of that which simply concerns us most, or a dogmatic position of commitment in the name of a subject of truth [...] we claim that critique has a history. Not just that it must mobilise historical material to ward off a-historical tendencies, in the manner of historical materialism or genealogy for example, but that the very idea and practice of critique must adapt as social formations evolve and change. — \cite[17]{mackenzie2021} \end{quote} Rather than abandoning critique, \citeauthor{mackenzie2021} \parencite[]{mackenzie2021} emphasise its necessity within contemporary sociotechnological formations. Critique enables more than just stepping beyond current epistemic borders; it functions as the essential force that prevents collapse into indifference, sedimentation, or a-historical regression. In this sense, critique is not merely relevant to resistance but constitutes its very precursor and substance (see \citeauthor{mackenzie2018}'s earlier work \citetitle{mackenzie2018} \cite*{mackenzie2018}). As much as power/knowledge, resistance/critique is fused and one; they form an inseparable dyad. Critique must therefore adapt to new emergences and develop ways to \textit{step beyond} established configurations. My claim extends this trajectory: not only does critique remain possible in the advent of \gls{genai} models, but these meaning-making entities can themselves become instruments of divergence, sources of novelty within processes of subjectivation and human–machine interaction. \section{Neoplatonic Latency: Current Debates and Literature Review}\label{sec:literature} Contemporary debates surrounding \gls{genai} are already abundant. Yet the critique of algorithmic power has a longer genealogy than the recent enthusiasm over generative systems. The rise of predictive analytics inaugurated what some call the \enquote{datalogical turn} \parencite{Clough2015}, a mode of governance grounded in the continuous capture and operationalisation of data traces, while at the institutional level decision-making was analysed under the rubric of \enquote{algorithmic governmentality} \parencite{rouvroy2007}. Scholars have shown how these infrastructures align with neoliberal rationalities, translating social life into market norms \parencite{demir2019}, and how they feed into surveillance capitalism \parencite{zuboff2019}. Parallel work has examined the ethical stakes of algorithmic deployment, focusing on fairness, bias, and discrimination in both design and application \parencite{kordzadeh2022}. Attempts to adapt Deleuze’s notion of control to these developments have taken divergent paths. Some extend it to big-data modulation and predictive environments \parencite{brusseau2020}, while others question its adequacy for present formations \parencite{hui2015}. Earlier accounts of \enquote{dividualisation} \parencite{Cheney2011, Otterlo2013} remain important, but they predate today’s generative infrastructures and therefore cannot address how meaning production itself becomes a site of governance. While critiques of algorithmic governance are extensive, they often centre on surveillance, datalogical rationality, or ethical bias without engaging the representational novelty of \gls{genai}. The capability of these models, especially \glspl{llm}, to perform meaning-making \parencite{gretzky2024, mishra2024, dishon2024} has provoked renewed inquiry. By traversing vast data foundations to generate outputs that appear plausible within a learned distribution, they reposition representation itself as a site of governance, even as their operations remain opaque in causality and justification. Although the theoretical literature on \gls{genai} is still thinner than that on earlier algorithmic systems, several contributions stand out as particularly relevant for this study. In the \gls{genai} related literature, \citeauthorfull{amoore2024} \parencite*[]{amoore2024} shift attention from surface-level failures to the deeper representational structures of \gls{ai}. While \citeauthorfull{bender2021b} \parencite*[]{bender2021b} warned that \glspl{llm} risk \enquote{parroting} entrenched arguments in their training data, a risk amplified when models are trained on their own outputs, \citeauthor{amoore2024} argue that the real concern is how models generate a \textit{political} \enquote{world model}\sidenote{The notion of a \enquote{world model} comes from current \gls{ai} research, where the next frontier is framed as the development of models capable of learning richer, actionable, and more versatile representations of their environment \parencite{lecun2022a}. Such an advancement is expected to enable systems to generalise to previously unseen problems (outside of the model's training data or previous experience) and produce context-sensitive solutions in ways that appear intuitive.}. Rather than simply repeating arguments, these systems subtly orient data interpretation toward ideological directions that are difficult to foresee. Although contemporary \gls{genai} models remain far from constructing a genuine world model, \citeauthor{amoore2024} express concern that the central representations these systems derive from their training data increasingly begin to fill the gaps opened during generation (in the case of \glspl{llm}, for instance, gaps of meaning), while the model compresses and interprets data into lower-dimensional representational spaces (\enquote{latent spaces}). Yet this focus, while introducing important questions about the nature of human–machine communication, leaves aside how and why these abstractions of the world derived through data come to take monolithic form (see Section~\ref{sec:genai}). It also remains unclear how these representations differ from human cognition, which likewise depends on schematic reductions of experience. Despite these limits, \citeauthor{amoore2024}’s critique links world-modelling to debates on \enquote{machinic Neoplatonism} \parencite{mcquillan2018a, eloff2021}, the idea that reality is best perceived mathematically, waiting to be extracted from data. Building on this, \citeauthorfull{eloff2021} \parencite*[]{eloff2021} refers to the term \enquote{Algocene}; a new epoch shaped by algorithmic environments, atemporal, high-speed networks that fragment subjectivity into dividual aggregates. The novelty of \citeauthor{eloff2021}’s formulation lies in showing how these environments shape human cognition in ways that resemble the operations of deep learning systems, where contemporary forms of communication take on the structure of adversarial pattern-recognition and recursive error correction. It is this distinctive configuration that \citeauthor{eloff2021} \parencite*[179]{eloff2021} names the \enquote{algoplastic} stratum. While drawing on Deleuze’s \textit{Postscript}, he is concerned less with specifically deliberative behaviours of \gls{genai} models than with the wider transformations of subjectivity that arise when humans interact with increasingly pervasive algorithmic systems in a continuous bilateral exchange that reorganises political discourse. His account also highlights the cognitive vulnerabilities humans suffer from in encounters with \textit{other} meaning-making entities, offering a productive angle for examining how agency emerges within the Algocene. The problematic of agency is taken up further in accounts of the sociotechnological imaginary of \gls{ai}, whereas \citeauthorfull{dishon2024} \parencite*[]{dishon2024} and \citeauthorfull{prinsloo2017} \parencite*[]{prinsloo2017} emphasise how interactions with \gls{genai} unsettle clear boundaries of action and intention. \citeauthor{dishon2024} critiques the dominant \enquote{Frankenstein} imaginary, which frames \gls{ai} as an anthropomorphic, external threat, and instead proposes \citeauthorfull{kafka1988}’s \citetitle{kafka1988} \parencite*[]{kafka1988} as a more apt metaphor. In this view, agency becomes entangled: humans and \gls{genai} systems are drawn into a mutual, ongoing attempt to interpret and respond to each other without ever fully grasping the other’s logic. Relations are recursive in a specifically interpretive sense, as both sides continually generate and infer meaning in ways that reshape the encounter itself. Control, rather than appearing as a struggle over the inscription of subjectivity, shifts toward patterned constraint, where \gls{genai} expands the set of choices available to users even as it narrows the forms of meaning that can emerge within those choices, producing an extended and indeterminate process of negotiation. \citeauthor{dishon2024}’s intervention offers a distinctive way of understanding how \gls{genai} reorganises the conditions under which meaning is produced, and clarifies why familiar dichotomies of agency, relations, and control no longer hold in the Kafkaesque landscape of contemporary human–machine interaction. A common thread running through these debates is that both meaning and agency are shaped through infrastructural arrangements that exceed individual interaction. This shift becomes explicit in the analysis developed by \citeauthor{mackenzie2021}, who define emergent \gls{genai} infrastructures as computational \enquote{totalising institutions}. Their account highlights how \gls{ai} models sequence dividual traces across domains, with scores, profiles, and categories continuously inscribed, routed, and repurposed so that conduct is modulated over time (see \cite[23–24]{mackenzie2021}). While these reflections echo earlier analyses of algorithmic governance, their distinctive contribution lies in linking critique and resistance through the concept of \enquote{counter-sequencing}: \begin{quote} While these algorithmic functions are now well known, and can be critiqued at the level of the potentially infinite process of signification they constrain, they can also be critiqued through a process we would call counter-sequencing. Counter-sequencing is the activity of reordering the power diagram of the totalizing institution in ways that destabilize its functioning. That said, it would be unwise to assume in advance that counter-sequencing must result in some kind of ‘positive’ ethico-political outcome. The aim, instead, is to understand the critical potential of counter-sequencing first and then to engage in, what Williams calls, the revaluation of that critique with more ‘local’, that is ‘pragmatic’, concerns at the forefront of such revaluations. — \cite[23–24]{mackenzie2021} \end{quote} Mackenzie’s proposal is significant for shifting the focus from critique as diagnosis to critique as intervention. Yet it remains underspecified; counter-sequencing is described more as a gesture than as a concrete practice, leaving open questions about its operational form and its place in processes of subjectivation. By contrast, \citeauthorfull{montanari2025} \parencite*[]{montanari2025} addresses these gaps more directly by engaging the technical mechanisms of contemporary architectures. He foregrounds procedures such as \textit{dimensionality reduction} (similar to \citeauthor{amoore2024}'s \parencite*[]{amoore2024} reference to latent spaces) and the \emph{transformer architecture} underpinning \gls{genai}, analysing how its technical operations produce cultural resonance. Montanari argues that transformers exemplify the interplay between metaphor and function: as specialised \glspl{nn} that simulate certain brain structures, they excel at processing sequential data through their \textit{attention mechanism}, an innovation that enables selective focus on relevant parts of input sequences to discern complex relationships and dependencies within data \parencite[206]{montanari2025}. This functionally specific yet mythically resonant architecture, he suggests, reveals how technical metaphors solidify both the utility and mystique of \gls{ai} systems. Here, resistance is framed not as wholesale rejection but as a possibility internal to machinic processes themselves. \citeauthor{montanari2025} \parencite*[see][208-210]{montanari2025} notes that the agency issue might be more complex than it seems, and emphasises the necessity for research into the inner working of the \gls{genai} models in order not to leave the understanding of their capabilities and future development to tech giants. His account, however, remains largely programmatic, leaning towards ethical risk and cultural semantics rather than concrete pathways for architectural reconfiguration. It is here that \citeauthorfull{beckmann2023} \parencite*[]{beckmann2023} advance the discussion with a decisive technical intervention. They propose \enquote{computational phenomenology}, a framework that rejects the \enquote{neuro-representationalist} assumption that \gls{ai} systems interact with the outer world only through a monolithic representation they build. Instead, they argue that \gls{genai} should be understood as a processual style of sense-formation: it generates meaning not by referencing a stored map of reality, but through dynamic, context-sensitive activations within its neural network. This perspective highlights the potential of repurposing and reconfiguring pretrained networks by emphasising or capitalising on, for example, specific layers of meaning formation. The \enquote{DeepDream} experiment can serve as a metaphorical example: a model initially intended as a classification architecture was turned into a generative model \parencite[416]{beckmann2023}, ultimately producing novel and dream-like images. Read this way, architectural variation itself becomes a site for configuring models differently, or for actions such as counter-sequencing that displace the centripetal pull of standardised \enquote{world models} and open alternative structures. Taken together, these accounts indicate both the richness and the limits of contemporary theorisation. While \citeauthor{mackenzie2021} introduce counter-sequencing as a critical method, \citeauthor{montanari2025} and \citeauthor{beckmann2023} point toward technical and conceptual openings for resistance within \gls{genai}. Yet their analyses remain partial. In what follows, I extend these debates by combining political-theoretical insights with a closer examination of generative infrastructures, in order to develop a framework for critique and resistance adequate to today’s algorithmic \glspl{dispositif}. \section{Methodological Approach}\marginnote{Cere-nominal, a concept from \citeauthorfull{burroughs2012}' \parencite*[]{burroughs2012} \citetitle{burroughs2012}.} The methodology follows from the gaps identified in existing scholarship, adopting an analytical, interdisciplinary, critical, and technical orientation. Returning to the academic approach suggested by \citeauthorfull{galloway2004} \parencite*[]{galloway2004}: \begin{quote} For Empire, we must descend instead into the distributed networks, the programming languages, the computer protocols, and other digital technologies that have transformed twenty-first-century production into a vital mass of immaterial flows and instantaneous transactions. Indeed, we must read the never-ending stream of computer code as we read any text (the former having yet to achieve recognition as a \enquote{natural language}), decoding its structure of control as we would a film or novel. — \cite[82]{galloway2001} \end{quote} Although writing at a much earlier stage, Galloway (see \cite{galloway2001, galloway2004}) was among the first in critical theory to examine the technical machinery of the internet in order to show how control persists even after decentralisation. While his work plays a contextual role in this study, his methodological orientation also provides an important inspiration for how to approach the present research. In analysing \gls{genai} within the framework of control societies, the methodology proceeds on two fronts. First, it traces the historical development of \gls{ai}, with particular attention to the breakthroughs that enabled contemporary models, most notably the transformer architecture. Second, it situates these systems institutionally, examining their embedding within structures of power and knowledge. \gls{genai} is therefore analysed not merely as a technical artefact but as part of a wider constellation of \glspl{dispositif} that reorganise governance and subjectivation. Accordingly, the approach combines: \begin{enumerate}[label=\roman*.] \item a genealogical reading of power that situates \gls{genai} within the shift from disciplinary to control societies, \item a close technical exegesis of model architectures, with attention to how specific mechanisms condition the politics of mediation, \item an examination of the institutional embedding of contemporary computational infrastructures, \item and a reflection on how resistance and critique may be theorised through \gls{dg}’s \enquote{Capitalism and Schizophrenia}. \end{enumerate} Ultimately, I argue that while the generative capabilities challenge the pillars of the control society concept, they find particularly insightful correspondences in other literature of \gls{dg}. \section{The Cartogram}\label{sec:cartogram} The thesis adopts a structure that resists straightforward sequential progression. Each chapter forms a distinct yet connected zone of inquiry, linked to others through cross-references that allow arguments to circulate rather than accumulate in linear order. To prevent ambiguity within this architecture, chapters incorporate brief orientations and concluding summaries, enabling them to function as semi-autonomous units. The Cartogram that follows outlines this arrangement, mapping how the project’s core concerns, \gls{genai} as an institutional \gls{dispositif} and the possibilities of critique and resistance, are distributed and interlinked within this constellation. \paragraph{Chapter~\ref{cha:control}} reconstructs the conceptual shift from disciplinary to control societies, situating \gls{genai} within this transformation. It begins with Foucault’s genealogy of subjectivity, showing how disciplinary institutions like prisons, schools, and factories moulded docile subjects through enclosure, surveillance, and routine. Deleuze’s \textit{Postscript} marks a rupture: the emergence of control societies that act through continuous modulation, dividuation, and algorithmic circulation rather than architectural confinement. This provides the theoretical ground for analysing \gls{genai} not simply as a technical tool, but as a \gls{dispositif} that reorganises governance and subjectivation by acting on data traces instead of enclosed bodies. The chapter revisits key concepts such as docility, modulation, and dividualisation to clarify how subject formation evolves under institutional and computational infrastructures. The second part develops the problem of critique and resistance. Deleuze’s Postscript leaves resistance underspecified, offering only fragmentary hints in his reference to \enquote{program}. The chapter, therefore, turns to later theorists such as Hardt, Negri, Galloway, and MacKenzie \& Porter, who expand Deleuze's formulation while also introducing a notion of resistance. By bringing these debates into contact with \gls{genai}, the analysis shows that critique cannot be external to control but must operate within its infrastructures. Resistance is theorised less as external negation than as the activation of lines of flight and micropolitical apertures already present in algorithmic \glspl{dispositif}. This sets the stage for subsequent chapters, which deepen the analysis through technical, institutional, and machinic perspectives. \paragraph{Chapter~\ref{cha:ai}} shifts the focus to the technical and historical development of \gls{ai}, tracing the path from \gls{symai} to contemporary \gls{dl} and generative models. The chapter first outlines the classical paradigm of \gls{symai}, describing how reasoning was formalised in logical rules and arborescent systems, as well as why these approaches ultimately fell short in handling complexity, ambiguity, and contextuality. It then turns to the rise of connectionist models, beginning with early \glspl{nn} and their limitations, before examining the breakthroughs in \glspl{dnn} and \gls{ssl} that underpin today’s architectures. The analysis emphasises how these technical shifts are not only engineering milestones but also epistemological transformations. With the advent of \textit{transformer architectures} and \textit{attention mechanisms}, \gls{genai} systems acquired the capacity to generate rather than merely predict, reorganising the role of \gls{ai} from a tool of decision support to an infrastructure of meaning-making. The chapter situates these developments within broader debates, showing how probabilistic inference, tokenisation, and optimisation procedures reconfigure knowledge production and subjectivity. The analysis dives deeper into the machinery that made the contemporary \gls{genai} models exceptional, like \textit{transformer architecture}, how under-/overfitting are handled, how \textit{gradient descent} and \textit{backpropagation} work. These functionalities at times are analysed through \gls{dg}'s theory. This historical and technical mapping provides further foundation to reflect on current debates and conceptualisation (and/or verification) of specific (re-)configuration methods for \gls{genai} models. \paragraph{Chapter~\ref{cha:institution}} shifts from architectural analysis toward the problem of agency, examining how contemporary \gls{genai} systems reshape the conditions under which meaning, interpretation, and critique become possible. It opens with concerns over the representational limits of \glspl{llm}, drawing on \citeauthor{bender2021b} and \citeauthor{amoore2024} to show how statistical reconstruction and algorithmic inference risk naturalising political or ideological tendencies within predictive infrastructures. The discussion then turns to \citeauthor{eloff2021}’s notion of the algoplastic stratum, in which humans and \gls{genai} models interact within a shared interpretive space characterised by opacity, recursive sense-making, and shifting boundaries of agency. This provides the basis for incorporating \citeauthor{dishon2024}’s argument that \gls{genai} introduces a Kafkaesque dynamic, where action and intention blur through ongoing negotiation between human and machine. From there, the chapter explores alternatives to representational readings of \gls{ai}, engaging \citeauthor{beckmann2023}’s \enquote{computational phenomenology} as a framework in which meaning arises not from fixed internal maps but from layered activations that unfold contextually within \glspl{nn}. \citeauthor{montanari2025}’s contribution adds up to this trajectory by highlighting transformers' capacity for long-range conceptual relations and speculating on futures in which \gls{genai} participates more actively in socio-political narration. Taken together, these perspectives frame human–machine interaction as a hybrid and evolving formation, where meaning is continuously assembled across technical and social strata. This conceptualisation opens space for resisting convergent rationalities by intervening in the layered processes through which \gls{genai} systems generate patterns, narratives, and modes of subjectivation. \paragraph{Chapter~\ref{cha:conjunctive}} brings together the technical, institutional, and theoretical strands developed in the previous chapters in resemblance of conjunctive synthesis. The chapter argues that generative architectures bind together heterogeneous forces: data, algorithms, user inputs, institutional logics, and cultural imaginaries; into operational wholes, while still preserving indeterminacies that can be activated for divergence. The chapter begins with a microphysics of resistance, showing how model behaviour such as hallucination, chain-of-thought expansion, coherence production, and alignment not only stabilise meaning but also instantiate deviations and misfires that resist full capture. These deviations are read as micropolitical opportunities for activating creativity, divergent thought, and even intention within the model. The chapter then offers a series of interventions that aim to prevent generative architectures from becoming sedimentary. Techniques such as feature amplification, artificial curiosity, and the introduction of deliberate perturbations are framed as methods for sustaining a non-conforming tendency inside the model. The chapter uses the concepts from \enquote{Capitalism and Schizophrenia} (\cite{deleuze1983, deleuze1987}), like schizoanalysis and nomadology, to articulate how the potential for divergence already pointed out in the technical analysis of generative systems can be activated. The aim is neither romantic refusal nor technophilic celebration but a pragmatic activation of alternative sense-configurations within the architecture itself. The final part of the chapter develops the concept of counter-sequencing as a way of reorganising the power diagrams of \gls{genai}. Counter-sequencing treats hallucination, divergence, and improvisational generation as resources for reshaping the structures of human–machine interaction. Chapter 5, therefore, provides a conceptual and practical framework for navigating critique and resistance not as a refusal of generative infrastructures but as an immanent practice within them. Precisely by pursuing these prospects, I argue that the specificity of generative architectures is overlooked by the dominant perspectives, which either reduce them to traps of subjectivation or celebrate them through techno-solutionism. My work argues that both positions overlook the specificity of generative architectures. \Gls{genai} not only reterritorialises meaning by producing coherence from fragmented inputs; it also contains, within those same cohering processes, the potential for divergence. As computational \glspl{dispositif} become more sophisticated, so too do the micropolitical possibilities for counter-sequencing, divergence, and intervention. The significance of this project lies in addressing two persistent gaps. First, Deleuze’s sketch of control societies leaves the question of resistance under-theorised; my analysis extends this by grounding lines of flight in the concrete operations of \gls{genai}. Second, much scholarship on algorithmic power remains at the level of metaphor or ethics, without engaging the architectures themselves. By situating critique within the technical and institutional logics of transformers and related models, the study develops a framework for understanding generative infrastructures not merely as instruments of capture, but as terrains where new subjectivities and forms of political action can emerge.