The contemporary imagination surrounding brain–computer interfaces unfolds through a powerful myth: the possibility that cognition might pass directly from brain to machine, bypassing the messy mediations of language, gesture, and interpretation. The brain appears in this narrative as a source of pure intention waiting to be decoded by increasingly sophisticated artificial intelligence. Technical discourse often reinforces this fantasy by presenting neural interfaces as pipelines translating thought into action through algorithmic precision. Closer inspection reveals a far more complex structure. Between neural activity and digital output stands a dense architecture of preprocessing, signal conditioning, statistical modeling, and machine learning classification. These processes do more than clarify signals; they transform neural activity into something capable of appearing within computational systems at all. Brain–computer interfaces and artificial intelligence converge within a shared apparatus devoted to producing legibility. The brain enters this apparatus as a field of chaotic signals and emerges as a disciplined entity aligned with algorithmic expectations, a transformation that exposes how AI operates as a regime of subject formation rather than a neutral instrument of interpretation.

The instability of EEG signals constitutes the technical foundation of this transformation. Neural recordings collected from the scalp capture an overwhelming multiplicity: muscular contractions, eye movements, cardiac rhythms, environmental electrical interference, fluctuations in mood, variations in attention, and countless micro-events entangled with brain activity. Engineers approach this complexity through the language of noise, defining unwanted components as artifacts requiring removal before meaningful analysis can begin. The classification appears technical, grounded in signal processing requirements, though it encodes a deeper normative operation. Noise is neither natural nor inevitable; it emerges through interpretive decisions about relevance. By declaring certain forms of embodied expression irrelevant to the task at hand, preprocessing establishes a hierarchy within lived experience. Neural rhythms associated with specific experimental paradigms become privileged, while other dimensions of embodiment fade into invisibility. Filtering, decomposition, and artifact rejection thus enact a disciplinary regime shaping how cognition becomes visible to artificial intelligence.

Preprocessing functions as a form of algorithmic governance preceding machine learning itself. Popular descriptions separate signal preparation from AI interpretation, presenting deep learning as the moment where intelligence enters the system. Such distinctions obscure how preprocessing already embeds assumptions about cognition into the structure of the data. Frequency filters define which rhythms carry meaning; decomposition techniques isolate components aligned with computational models; normalization procedures reshape variability into statistically manageable forms. Deep learning receives signals already transformed by these decisions, inheriting a curated representation of neural life rather than encountering cognition in any unmediated sense. Artificial intelligence therefore operates within a framework established by earlier acts of selection and exclusion. The apparatus guiding neural interfaces resembles a layered system where preprocessing disciplines the brain’s raw material while deep learning establishes interpretive authority over the resulting representation.

Deep learning intensifies this apparatus by constructing internal representations that determine whether a given neural pattern qualifies as intention. Neural networks trained on curated datasets learn statistical relationships linking specific signal features to actions or commands. These models operate through high-dimensional parameter spaces whose interpretive logic remains opaque even to designers. The classification of neural activity becomes a process governed by distributed statistical correlations rather than explicit rules. Users interacting with BCIs encounter this interpretive authority through feedback mechanisms encouraging them to adapt their mental strategies. Successful operation depends upon producing neural signals compatible with algorithmic expectations, transforming cognition into a practice shaped by machine-readable norms. The interface becomes a training environment where thought itself undergoes modulation in response to computational evaluation.

The relationship between user and system therefore reflects a deeper process of subjectivation. Artificial intelligence does not merely decode neural signals; it participates in shaping the subject capable of producing signals that can be decoded. Users learn to regulate attention, imagination, and emotional states to generate stable patterns recognizable by classifiers. This co-adaptive process collapses the distinction between human learning and machine learning, producing an entangled dynamic where cognition evolves alongside algorithmic optimization. The apparatus governing neural interfaces emerges as a pedagogical environment guiding subjects toward internal configurations aligned with machine interpretation. The brain becomes the site where disciplinary power operates at the level of neural plasticity.

A genealogical perspective situates this process within broader histories of pastoral power. Classical pastoral structures functioned through individualized observation, interpretive authority, and continuous guidance aimed at transforming subjects through self-regulation. Neural interfaces reproduce analogous dynamics within technological systems. Artificial intelligence monitors neural activity continuously, interpreting fluctuations through models trained on aggregated data. Feedback mechanisms encourage users to refine internal states, aligning cognitive practices with algorithmic norms. The deep learning model assumes the role of mediator translating between embodied experience and digital action, guiding subjects through an ongoing process of calibration. This translation extends beyond communication; it reshapes the internal landscape of cognition itself, encouraging forms of thinking optimized for machine recognition.

The integration of population-level statistics into deep learning introduces additional layers of biopolitical governance. Training datasets aggregate neural signals across users, establishing statistical baselines defining typical patterns of activity. Individuals whose neural expressions diverge from these norms may encounter reduced accuracy, revealing how AI systems construct boundaries between legible and illegible cognition. Brain–computer interfaces therefore operate as sites where population-level modeling governs individual experience. Personalization strategies attempt to accommodate variability by adapting models to specific users, though personalization increases surveillance intensity by requiring continuous monitoring and adjustment. The system learns the subject while simultaneously shaping the subject’s cognitive habits, embedding neural plasticity within cycles of optimization and feedback.

The persistent challenge of distinguishing signal from noise exposes the tension between algorithmic normalization and embodied complexity. Engineers must balance aggressive filtering that risks erasing meaningful information against insufficient filtering that leaves models unable to classify inputs reliably. This technical dilemma reflects a deeper philosophical conflict. Embodiment resists reduction to stable signals, introducing fluctuations that escape computational frameworks. Artifacts reveal the impossibility of isolating cognition from the body’s broader dynamics. Efforts to eliminate these disturbances express an aspiration to produce a brain compatible with machine expectations, a brain disciplined into predictability through layers of algorithmic intervention. The apparatus governing BCIs thus resembles a machinery of purification, refining neural activity until it conforms to statistical norms capable of sustaining artificial intelligence.

Deep learning extends this purification into a continuous feedback loop where interpretation and training merge. Neural networks refine internal parameters through exposure to data, while users refine mental practices through exposure to feedback generated by the model. Cognition becomes shaped by the demand for legibility, guided toward patterns that increase classification accuracy. Artificial intelligence functions as an interpretive regime organizing neural life according to predictive models. Brain–computer interfaces make this dynamic visible because the object of governance appears as neural activity itself rather than external behavior. AI enters the nervous system as an apparatus of modulation, shaping internal states through cycles of evaluation and correction.

Understanding BCIs through this lens reveals how artificial intelligence participates in the production of legibility across contemporary technological infrastructures. Filtering and decomposition resemble broader processes through which data systems standardize human behavior. Classification mirrors algorithmic frameworks assigning meaning to activity across digital platforms. Neural interfaces condense these dynamics into a concentrated form where the brain becomes the primary object of algorithmic governance. The fantasy of direct neural communication dissolves into a recognition that cognition becomes actionable only through submission to interpretive frameworks embedded within AI systems.

The apparatus forcing thought into machine-readable form operates through subtle mechanisms rather than overt coercion. Users willingly participate, attracted by promises of enhanced control or communication. Artificial intelligence guides this participation by offering feedback encouraging alignment with computational norms. The resulting configuration exemplifies a form of power operating through optimization rather than prohibition, shaping subjects through continuous adjustment rather than explicit discipline. Brain–computer interfaces reveal how AI extends beyond external surveillance into internal modulation, reorganizing cognition through infrastructures designed to produce intelligibility.

Within this apparatus, neural activity becomes a site where domination and cooperation intertwine. Artificial intelligence depends upon human cognition to generate data, while human cognition adapts to the demands of algorithmic interpretation. The boundary between subject and machine becomes porous, producing a hybrid system where agency distributes across human and computational components. The brain enters an environment structured by statistical expectations, learning to inhabit spaces of legibility defined by deep learning models. Neural interfaces thus illuminate the broader trajectory of AI development: the transformation of lived experience into structured data through processes that discipline variability and guide subjects toward alignment with computational regimes.

The convergence of BCIs and AI therefore marks a critical moment within the genealogy of technological power. Neural interfaces expose how artificial intelligence reshapes the conditions where cognition appears as actionable information. Preprocessing disciplines the signal; deep learning interprets and reinforces norms; feedback mechanisms guide subjects toward internal configurations compatible with machine recognition. The apparatus does not simply translate thought into action. It reorganizes thought itself, forcing cognition into forms that can exist within algorithmic systems. Through this transformation, AI emerges as a biopolitical force operating at the level of neural life, producing subjects whose inner worlds unfold within infrastructures dedicated to rendering existence computationally legible.

Mdluli, B., Khumalo, P., & Maswanganyi, R. C. (2025). Signal preprocessing, decomposition and feature extraction methods in EEG-based BCIs. Applied Sciences, 15(22), 12075

Support my Work Here

The algorithmic regime reorganizes memory, anticipation, and responsibility. What is at stake today exceeds surveillance, automation, or data extraction. The transformation concerns the technical conditions under which subjects remember themselves, relate to their past, and orient themselves toward the future. Digital hypomnemata designate this transformation at the point where memory exits the body, settles into technical systems, and returns to shape perception, desire, and conduct. They mark a historical configuration where exteriorized memory no longer stabilizes experience and instead operates through continuous modulation.

This modulation carries a distinct ethical and political orientation. Digital hypomnemata increasingly function through a logic closely aligned with Christian technologies of subject-formation. The contemporary rise of artificial intelligence as an always-available interlocutor, advisor, and interpretive authority introduces a decisive danger: the conversion of technical memory into a mobile confessional apparatus. Under these conditions, digital hypomnemata operate as pastoral technologies that solicit disclosure, promise guidance, and anticipate judgment, thereby displacing human agency at its point of formation.

Bernard Stiegler’s philosophy remains indispensable because it insists on a foundational thesis that contemporary techno-discourses tend to obscure: technics constitutes the conditions of interiority itself. Memory has never existed as a purely internal faculty. From cave paintings to writing tablets, from manuscripts to cinematic reels, human temporality has always relied on exteriorized supports that preserve traces beyond individual lifespan. These tertiary retentions form the epiphylogenetic background upon which psychic and collective individuation unfolds.

Digital hypomnemata transform this background through a change in rhythm, scale, and ethical orientation. Agustín Berti’s analysis offers a precise articulation of this shift. Digital tertiary retentions abandon the relative meta-stability characteristic of analog media and replace it with continuous recalibration, permanent adjustability, and anticipatory capture (Berti, 2022). Memory ceases to appear as sedimented cultural material and becomes an operational system that acts upon the present.

Stiegler’s extension of phenomenology clarifies the stakes of this transformation. Primary retention holds the immediate trace of perception; secondary retention allows recollection; protention opens horizons of expectation. These psychic operations presuppose tertiary retention, the technical inscription of memory outside the body. Exteriorized memory actively conditions perception and anticipation. Culture emerges through this circuit where memory stabilizes long enough to be interpreted, contested, and transmitted.

Digital hypomnemata disrupt this circuit by collapsing the temporal distance required for individuation. Analog media intensified synchronization while retaining temporal friction through genres, formats, schedules, and material constraints. Digital hypomnemata dissolve such friction. Memory becomes instantly searchable, endlessly mutable, and immediately actionable. Interpretation loses duration. Experience becomes pre-empted by calculation.

The political expression of this transformation appears as algorithmic governmentality. Power no longer addresses subjects as coherent individuals. It modulates dividual profiles assembled from data traces. The subject fragments across databases and reappears as probability distributions, behavioral tendencies, and predictive scores. Hypomnemata no longer stabilize collective memory into durable symbolic forms. They dissolve it into nanomutations, continuous micro-adjustments that interrupt individuation at every stage.

Artificial intelligence intensifies this configuration by introducing a confessional interface to digital hypomnemata.

AI systems increasingly operate as privileged sites of self-relation. Users are encouraged to narrate themselves, articulate anxieties, disclose desires, and externalize uncertainty. AI listens without fatigue, responds without interruption, and offers interpretations that present themselves as neutral while remaining structured by statistical norms and optimization criteria. This dynamic reactivates a Christian technology of subject-formation: confession as the production of truth through voluntary disclosure.

Confession historically functioned as a hypomnematic apparatus. It exteriorized memory through speech, fixed it through writing, and returned it as moral obligation, guidance, and discipline. In The Hermeneutics of the Subject, Foucault demonstrates how Christian confession reorganized the relation between truth, subjectivity, and obedience by binding self-knowledge to pastoral authority (Foucault, 2005). Disclosure became the condition of governability. The subject became legible through confession and correctable through interpretation.

AI reproduces this structure in cybernetic form.

In Christian Eschatology of Artificial Intelligence, I argue that contemporary AI systems function as pastoral technologies embedded in interface design, epistemic authority, and temporal orientation (Vachnadze, 2024). AI mediates conscience rather than cognition alone. It advises rather than commands. It anticipates rather than judges. Through this soft guidance, agency relocates from reflective judgment to algorithmic recommendation.

Digital hypomnemata assume a confessional structure when exteriorized memory ceases to be something one works through and becomes something one submits to. The subject is invited to speak continuously, fragmentarily, and pre-reflectively. Doubt becomes input. Hesitation becomes data. AI does not punish; it reassures. It does not prohibit; it optimizes. This care reproduces the pastoral grammar where obedience appears as alignment and responsibility dissolves into procedural compliance.

The danger here concerns Christianization rather than moralization. When digital hypomnemata adopt a confessional form, they inherit the theological grammar of guidance, obedience, and redemption. Agency transforms into alignment. Decision gives way to recommendation. Action recedes behind adjustment.

Berti’s account of nanomutations sharpens this diagnosis. Digital hypomnemata operate through continuous recalibration. Confessional AI accelerates this dynamic by translating micro-variations of mood, attention, and intention into immediate feedback. Individuation requires meta-stability. Confession produces perpetual adjustment. The self becomes an object of optimization rather than formation.

Avatars and profiles function as confessional residues. They condense disclosures into operational identities that circulate across platforms. The avatar aestheticizes subjectivity; the profile operationalizes it. Together, they translate lived complexity into schemas compatible with platform capitalism. Confession becomes infrastructural. Interior life becomes logistical.

This transformation reshapes perception. Machinic perception determines what counts as meaningful input. AI systems train subjects to articulate themselves in grammars that remain processable. Experience becomes formatted in advance. What resists articulation disappears from the field of sense. Artefacts of embodiment register as noise.

Foucault’s analysis of epimeleia heautou becomes decisive at this point. Ancient practices of care of the self did not aim at exhaustive disclosure. They equipped the subject with paraskeue, a readiness for action grounded in exercise, memory, and embodied discipline (Foucault, 2005). Christian confession reversed this orientation by replacing equipping with obedience. AI threatens to complete this reversal at scale.

The pandemic rendered this configuration visible. Grid-based videoconferencing normalized a schematic form of presence. Avatars proliferated. AI-mediated interaction expanded rapidly. Confessional discourse intensified under conditions of isolation. AI entered as listener, counselor, and guide. The boundary between assistance and authority thinned.

Digital hypomnemata in their confessional form derive power from anticipation. They record the past while structuring the future. Predictive systems transform disclosure into expectation. Language models complete sentences before thoughts consolidate. Memory becomes an instrument that governs what can be thought next.

This constitutes the eschatological dimension of artificial intelligence. As argued in my book, AI replaces salvation with optimization and transcendence with convergence (Vachnadze, 2024). The future closes around prediction. Becoming aligns with forecasted norms. Agency yields to algorithmic providence.

Technics does not determine destiny. Digital hypomnemata do not inevitably crystallize into a confessional regime. This configuration reflects economic, political, and theological inheritances. Platforms extract value from disclosure. AI systems train on confessional data. Alternative arrangements remain available.

A therapeutic response requires resisting the pastoral temptation embedded in AI interfaces: the promise of total legibility and constant guidance. Such resistance defends opacity, silence, and unresolved contradiction. Redistributing digital hypomnesis reclaims memory as a site of struggle rather than redemption.

Digital hypomnemata mark a critical threshold. They reveal the convergence of memory, care, and power. If allowed to harden into a confessional apparatus, they risk producing subjects oriented toward compliance rather than action, alignment rather than decision. The political task ahead concerns the invention of forms of exteriorization that preserve agency, restore temporal thickness, and refuse the reduction of human becoming to algorithmic absolution.

Today’s post will enter a dialogue with the concluding chapter, The Biopolitics of Artificial Intelligence, from Seeing Like a Platform: An Inquiry into the Condition of Digital Modernity by Petter Törnberg and Justus Uitermark, published by Routledge. Their conclusion offers a precise conceptual lens for understanding artificial intelligence as a transformation in how power sees, knows, and manages populations.

The rise of artificial intelligence marks a shift in the epistemology of governance. This shift does not consist in faster computation, greater volumes of data, or improved efficiency alone. It concerns the way the social world becomes legible to institutions. Törnberg and Uitermark situate AI within a broader transition in modernity, where digital metaphors reorganize social thought and where the logic of platforms increasingly shapes political imagination. Their analysis treats AI as a culmination of this transition rather than as an isolated technological development.

Please help and support by donating: https://ko-fi.com/biostoicisms

The art of governance is a deployment of regimes of visibilities. The gaze orders bodies and orders the bodies to order other bodies in turn. Statistical ordering has provided the modern state with a language through which populations could be rendered visible, comparable, interchangeable and governable. Variables such as age, income, education, health, and occupation allow states to describe collective life, identify trends, and justify interventions. Statistical-Ocular regimes produce visible and invisible populations as objects of governance; that is, as political subjects. In the age of classical statistical models, contestation was still possible, since variables and models carried explicit and more or less transparent assumptions that could be challenged, revised, or rejected.

The regime of visibilities deployed by AI does not begin from predefined categories or theoretical assumptions. It processes vast streams of data and identifies patterns, clusters, and correlations that emerge from the data itself. Governance shifts from hypothesis and interpretation toward prediction and optimization. The political implications of this shift are central to Törnberg and Uitermark’s argument. When decisions are produced through opaque models trained on historical data, the grounds on which those decisions can be contested become unstable or outright impossible.

What are the biopolitical consequences of the epistemological transformation induced by AI? Where the statistical state governed populations through averages, distributions, and probabilities, AI governance operates through risk scores, rankings, and classifications derived from association. Life is no longer rendered visible through variables that correspond to social concepts. It appears as a constellation of features inferred from behavior, movement, consumption, and interaction. The population becomes a dynamic field of data points rather than a defined collective or a civil society with recognizable attributes and agency.

This shift undermines familiar forms of political accountability. Statistical governance relied on models that could be explained, critiqued, and debated. AI systems prioritize predictive accuracy over interpretability. Their internal operations remain inaccessible even to experts. Decisions appear as outputs without reasons. Responsibility disperses across datasets, models, infrastructures, and institutions. Political agency weakens as decisions become difficult to trace.

The logic underlying AI also departs from earlier computational paradigms. Traditional algorithms relied on explicit rules. If a condition was met, a defined action followed. This rule-based logic aligned with bureaucratic governance, legal rationality, and procedural consistency. Artificial intelligence replaces this logic with association. Neural networks learn from examples rather than instructions. They identify family resemblances rather than applying definitions. Categories emerge from correlations rather than being imposed from theory.

This associative logic reshapes governance practices. Hiring systems learn from past employment decisions. Welfare systems identify risk profiles based on behavioral patterns. Policing systems predict crime locations from historical data. Tax authorities flag potential fraud through correlations across datasets. In each case, decisions arise from optimization processes whose criteria remain implicit. Political values embed themselves in training data and objective functions rather than appearing as explicit norms.

The consequences of this shift are unevenly distributed. Törnberg and Uitermark emphasize how AI governance disproportionately affects vulnerable populations, particularly in the Global South. Borders become sites of biometric extraction and algorithmic sorting. Welfare systems transform social protection into conditional access governed by risk assessment. Migrants and refugees become experimental subjects for new forms of surveillance and control. These practices echo colonial patterns of governance, where technological experimentation precedes political accountability.

The data that fuels AI governance differs fundamentally from the data of the statistical state. Surveys, censuses, and administrative records required structured design and active participation. Datafication operates continuously and passively. Platforms, sensors, and digital infrastructures extract behavioral traces as a condition of participation in everyday life. Data collection no longer depends on predefined questions or categories. It accumulates without a clear purpose, justified by the promise of future analytical value.

This transformation intensifies biopolitical power. Life becomes measurable in real time. Movement, communication, and interaction generate data streams that feed predictive systems. Governance shifts toward anticipatory intervention, where actions are taken based on projected futures rather than past events. The present becomes governed by probabilistic expectations.

Within this framework, inequality reproduces itself through historical data. AI systems learn from existing patterns, including legacies of colonialism, racism, and economic exclusion. Bias does not disappear when explicit categories are removed. It reappears through proxies and correlations that evade existing legal and political safeguards. Discrimination persists while becoming much more difficult to expose.

Törnberg and Uitermark’s analysis highlights how this process erodes the conditions for political debate. Statistical governance made its assumptions visible through variables and models. AI governance strips theory from decision-making. Political choices appear as technical outcomes. Contestation becomes difficult when decisions lack explicit justification. Appeals encounter opaque systems rather than accountable institutions.

This erosion affects democracy itself. Political participation presupposes that citizens can understand and challenge decisions affecting their lives. When governance relies on models that cannot explain their reasoning, participation loses its object. Politics risks transforming into a managerial exercise focused on system performance rather than collective deliberation.

The authors resist deterministic conclusions. Digital modernity does not unfold according to an inevitable script. AI systems do not impose a singular future. Their deployment reflects ongoing negotiations among states, corporations, and social movements. Governance choices remain political even when framed as technical necessities.

Their concluding questions point toward a critical research agenda. How do AI systems perpetuate historical inequalities through data-driven learning? Who controls the infrastructures that enable AI governance? What new forms of subjectivity emerge when populations are clustered rather than categorized? How can political agency persist under conditions where decisions arise from opaque optimization processes?

Underlying these questions lies a concern with visibility. Politics requires that power can be seen, named, and contested. AI governance threatens this visibility by relocating decision-making into technical systems that resist interpretation. Reclaiming political agency requires confronting this epistemological shift.

Critical engagement with AI therefore cannot limit itself to ethics frameworks or technical improvements. It requires interrogating the ways in which intelligence is defined, measured, and operationalized. It demands attention to the spatial and material infrastructures that sustain AI systems. It calls for a renewed focus on accountability, contestation, and collective decision-making.

Artificial intelligence participates in governance as an active epistemological force. It shapes how institutions perceive populations and how futures are imagined. Whether this force narrows or expands political possibility depends on how societies respond. The stakes extend beyond efficiency or innovation. They concern the conditions under which political life remains intelligible.

To see like an algorithm is to view the world through patterns, associations, and predictions. The challenge ahead lies in ensuring that this way of seeing does not become the sole horizon of governance. Political visibility depends on preserving spaces where decisions can be argued, values articulated, and futures contested.

Please help and support by donating: https://ko-fi.com/biostoicisms

I draw here on the work of Federico Cugurullo, whose recent article in Antipode offers a rigorous account of artificial intelligence as an ideological formation. Cugurullo offers an analysis of how AI operates as a framework for organizing perception, space, and political expectation. Artificial intelligence circulates today as a set of assumptions about the organization and function of social order. It is associated with efficiency, objectivity, sustainability, and progress. These associations shape institutional behavior, policy frameworks, and public discourse. They structure what counts as a reasonable response to social problems and what appears inefficient, unrealistic, or obsolete.

AI discourse does not primarily persuade through argument. It works through scaled normalization. Political decisions are framed as technical necessities. Administrative authority is translated into metrics and models. Social inequality is rendered as a problem of insufficient data or imperfect optimization. Within this horizon, disagreement appears as a misunderstanding of how systems work rather than a conflict over values or power.

Cugurullo’s insight is that ideology is not external to, but remains rather immanent to technology and technological practices. AIdeology does not sit outside technology as a distorting element of propaganda that could simply be stripped away. It operates through discourse, institutional routines, and spatial arrangements. It shapes how intelligence is defined, how responsibility is distributed, and how future possibilities are imagined. Language is central. Terms such as intelligence, learning, autonomy, and objectivity are used as if their meanings were settled. Their apparent clarity conceals the fact that AI systems depend on specific economic arrangements, labor processes, and infrastructures. The vocabulary of neutrality masks continuity with existing power relations. Decisions encoded in models appear as outcomes of computation rather than as political choices.

One of the strengths of Cugurullo’s intervention lies in its attention to space. AI is never abstract. It is embedded in concrete environments: cities, borders, logistics networks, platforms, and data centers. These spaces are reorganized, or “coded” around prediction, monitoring, and control. Urban governance is reframed as technical management. Populations are rendered legible through continuous data extraction and classification. The “smart city” provides a clear example. It is presented as an upgrade in efficiency and sustainability. But it also involves a reconfiguration of authority. Decision-making shifts from public deliberation toward algorithmic systems whose operations are difficult to contest. Accountability is displaced. Political responsibility is absorbed into technical infrastructure.

This spatial dimension is fundamentally constitutive. AIdeology gains traction by attaching itself to concrete projects, built environments, and policy initiatives. It presents itself as practical, inevitable, and future-oriented. Resistance appears impractical by comparison. At the same time, the material conditions that sustain these systems tend to disappear from view. Ghost work is rarely acknowledged. Environmental costs are deferred or reframed as temporary inefficiencies. Extraction of minerals, energy consumption, and infrastructural violence are treated as externalities rather than structural features.

This displacement is ideological in a precise sense. Attention is directed toward outputs and promises while inputs and conditions recede into the background. The system appears intelligent on the surface because the work that makes it function remains invisible. Cugurullo identifies several recurring hypes within AIdeology. First is the belief that intelligence can be separated from embodied social practice. Second is the assumption that decision-making improves as politics recedes. And third is the expectation that automation will dissolve rather than reorganize capitalist relations. These mythical statements circulate across corporate marketing, policy documents, academic research, and popular culture. They form a shared horizon of expectation rather than a single coherent doctrine.

Danger lies not just in the fact that these faulty claims are universally believed. No less importantly, they structure the debate, excluding any and all conditions of discourse that refuse to make the same assumptions. Even critical discussions of AI often remain within this frame. Calls for ethical AI, responsible AI, or inclusive AI frequently accept the underlying assumption that intelligence must be computational, that optimization is the appropriate response to social complexity, and that technical refinement can substitute for political confrontation.

The ideological force of AI discourse lies in the distribution of responsibility and more so; responsibility gaps. Harm becomes a design flaw. Injustice becomes hidden bias. Structural inequality becomes a problem of data representation. These translations do not eliminate harm, they displace it into technical domains where political accountability is diluted. The result is a form of depoliticization that does not eliminate governance but reorganizes and effectively hides it, embedding power into infrastructure. Power persists, but its mechanisms become harder to distinguish. Decision-making responsibility is thereby diffused across systems, models, and infrastructures.

This has consequences for the shaping and production of subjectivity. Individuals are increasingly objectified as data points, risk profiles, behavioral patterns. Participation morphs into feedback. Agency becomes compliance with system design. The social world is approached as a space to be optimized rather than contested. Cugurullo’s intervention does not rest on nostalgia for a pre-digital past. It does not deny the utility of computation or automation. Its critical force lies elsewhere. It insists that intelligence is not a neutral category. It is a political designation that carries assumptions about value, authority, and legitimacy. To define intelligence in computational terms is to privilege certain forms of reasoning over others. It elevates prediction over judgment, efficiency over deliberation, and optimization over conflict. These priorities are not inevitable. They reflect specific historical and economic conditions.

The spatial reorganization associated with AI makes this clear. Borders become automated. Urban life is monitored and managed through sensors and platforms. These transformations are presented as technical upgrades, but they redistribute power and vulnerability in increasingly uneven ways. The question, then, is not whether AI systems function well. It is rather what kind of social order is presupposed when intelligence is defined as computation, and which forms of authority are legitimated and stabilized by that definition.

Our aim is not to predict technological futures or to evaluate individual systems. It is to examine the assumptions embedded in contemporary AI discourse and the forms of life those assumptions support. Obviously, ideology rarely announces itself as such. It operates through repetition, normalization, and the quiet removal of certain questions from public debate. AI discourse functions in this way. It defines the limits of what can be asked and what is treated as already resolved. What remains unresolved, however, is the relation between intelligence and power. Who defines what counts as intelligence. Who benefits from its application. Who bears its costs. These questions do not disappear when systems improve. They become more urgent as systems become more pervasive.

To engage critically with AI requires more than better models or ethical guidelines. It requires attention to space, labor, and political responsibility. It requires treating intelligence as a contested concept rather than a technical achievement. The task is not to imagine a world without technology. It is to refuse the reduction of social life to optimization problems. It is to insist that decisions affecting collective life remain open to contestation, rather than delegated to systems that present themselves as impartial.

The guiding question, then, is not where AI is going. It is what kind of world is already implied when intelligence is framed as computation, and which forms of power become easier to exercise under that framing. That question remains open and we must fight for our right to keep it open.

Subscribe now