Algorithmic Power: AI, Platforms, Foucault, Afficentica, and Postsubjective Psychology
Updated: 2 days ago
Author: Ukrainian Psychological Hub · Published: September 24, 2026 · Editorial Policy
Algorithmic power is the capacity of algorithmically mediated systems to shape what becomes visible, relevant, selectable, reachable, rewarded, delayed, recommended, classified, or acted upon. It operates when ranking systems order information, recommender systems structure exposure, predictive systems assign scores, platforms allocate opportunities, automated managers monitor work, or AI systems influence decisions. The crucial point is relational: algorithmic power does not reside in code as if code possessed a mysterious force of its own. It emerges from a sociotechnical configuration of models, data, objectives, interfaces, institutions, business rules, human choices, and user responses. This relational understanding is central to critical algorithm studies, including Taina Bucher’s account of algorithmic power and Tarleton Gillespie’s analysis of algorithms as socially and institutionally situated mechanisms of relevance.
For psychology, algorithmic power matters because systems can alter the conditions under which attention, judgment, trust, fairness perceptions, autonomy, emotion, resistance, and behavior unfold. Yet “algorithmic power” is not a single standardized psychological construct, a clinical diagnosis, or a universal causal law. It is an interdisciplinary concept whose psychological consequences vary by domain, task, design, stakes, transparency, dependence, alternatives, and the user’s own goals. A 2026 review of platform research argues that algorithmic curation can substantially structure online experience while also warning against explanations that erase human preference and agency; the evidence linking curation itself to many downstream social outcomes remains mixed and context-dependent (Hosseinmardi et al., 2026).
This article owns the present-day algorithmic-mediation intent inside the Psychology of Power knowledge network. The broader scientific construct of interpersonal and relational power belongs to Social Power, changing structures of dependence and influence belong to Power Dynamics, and Michel Foucault’s historical theory belongs to Foucault on Power. Here the focus is what happens when consequential ordering is performed through algorithmic systems, how people psychologically respond to that ordering, and how Foucault, Afficentica, and Postsubjective Psychology provide distinct interpretive lenses for understanding it.
What Is Algorithmic Power?
Algorithmic power can be defined as the capacity of an algorithmically organized sociotechnical system to alter the field in which people encounter information, options, opportunities, evaluations, constraints, and consequences. The system may rank one item above another, predict a risk, recommend a route, assign a worker a task, filter a candidate pool, determine what enters a feed, prioritize one notification, suppress another, or generate an answer that becomes the basis for a human decision. Power appears when these operations matter to someone’s outcomes or possibilities.
The definition deliberately includes more than “the algorithm.” Bucher argues that algorithms are simultaneously technical, social, cultural, and functional objects, embedded in wider networks of relations and practices rather than existing as isolated agents (Bucher, 2018). Gillespie similarly emphasizes that search, recommendation, and platform algorithms are built into institutional decisions about relevance; the apparently technical act of ordering information contains human and organizational choices about what is counted, optimized, categorized, and made visible (Gillespie, 2014).
Algorithmic power therefore has at least three layers. The computational layer transforms inputs into classifications, rankings, predictions, or generated outputs. The institutional layer determines objectives, data access, deployment context, incentives, thresholds, appeals, and enforcement. The psychological layer concerns how people perceive, interpret, trust, resist, adapt to, or act within the resulting environment. These layers interact, and none of them alone explains the phenomenon.
Algorithmic Power Is Not the Same as Automation
Automation means that a process is performed partly or wholly by a machine or software system. Algorithmic power begins only when that automated process has consequential leverage over another actor’s informational or practical environment. A calculator automating arithmetic normally has little social power. A scoring system used to decide which applicants receive an interview can have substantial power because its output helps structure access to an opportunity. A recommendation system that orders millions of items can have power over visibility even when it has no authority to issue commands.
This distinction also prevents a common exaggeration. An algorithm may participate in a powerful system without being an autonomous ruler. The system’s leverage can come from platform scale, institutional dependence, control of scarce opportunities, interface design, default settings, data asymmetries, or the difficulty of opting out. In other cases, the same computational technique may be merely assistive. Power depends on the relation between the system, the institution that deploys it, the people affected by it, and the alternatives available to them.
Algorithmic Power, Social Power, Authority, Status, Influence, and Control
The word power is easily blurred with neighboring concepts. Social power concerns the capacity to affect another person’s outcomes within a social relation. Algorithmic power is one way such capacity can be organized or mediated, but the two concepts are not identical: social power can exist without algorithms, and an algorithmic system can structure visibility without resembling a conventional interpersonal power relationship.
Authority is legitimate or recognized entitlement to direct, decide, or command. An algorithmic system may shape choices without possessing authority in that sense. Status concerns respect, esteem, or social standing. A highly ranked creator may gain status because an algorithm amplifies visibility, but ranking and status are different variables. Influence is an actual change in beliefs, choices, or behavior, whereas power can exist as a capacity even when no measurable influence occurs. Control is narrower still: it refers to the ability to determine or constrain a specific outcome. Power Dynamics examines how these capacities move through relations of dependence; algorithmic power examines how computational ordering becomes part of those relations. For the dedicated conceptual treatment of how power can operate without formal command or recognized authority, see Power Without Authority: Foucault, Afficentica, and the Psychology of Configuration.
The distinction matters empirically. A recommendation can be influential because a person finds it useful. A platform can have power because it controls access to attention. An automated employer system can exercise control over scheduling. A generative AI answer can be treated as authoritative by a user even though the model has no formal institutional authority. These mechanisms may coexist, but treating them as synonyms obscures what is actually happening.
How Algorithmic Power Works
Algorithmic systems exercise power through several recurrent mechanisms. These mechanisms frequently overlap, and the same platform can use many of them at once.
1. Ranking and visibility
Ranking determines what appears first, what remains buried, and what never enters a user’s immediate field of attention. Search engines, feeds, marketplaces, app stores, news aggregators, and recommendation systems all convert large sets of possible objects into ordered displays. Gillespie identifies this production of relevance as a central social role of algorithms: ranking does not merely organize information after the fact; it helps establish what is likely to be encountered at all (Gillespie, 2014).
2. Recommendation and personalization
Recommendation systems infer what a user may want and decide which items to present. Their power arises partly from repeated exposure and partly from the practical impossibility of manually inspecting the underlying universe of options. Yet personalization should not be reduced to “the algorithm knows what people want.” A 2026 systematic review of 135 recommender-system studies in the audiovisual domain found that user preference is often poorly defined, participation in preference elicitation is limited, and system research frequently relies on behavioral proxies rather than richer psychological accounts of preference (Li et al., 2026). The recommendation pipeline therefore does not simply reveal a preexisting preference; it operationalizes preference through particular measures and design choices.
3. Classification and scoring
Predictive and classification systems place people, content, transactions, or events into categories. A category can affect which cases receive scrutiny, which recommendations are generated, which risks are flagged, or which opportunities are offered. The psychological importance of classification grows when people cannot see the relevant categories, do not know the variables that produced them, or have limited ways to contest the result. Paul de Laat’s Foucauldian analysis of predictive algorithms interprets many such practices as extensions of disciplinary power, while explicitly presenting that connection as philosophical analysis rather than experimental psychological evidence (de Laat, 2019).
4. Allocation and access
Algorithms can allocate tasks, attention, advertising inventory, service priority, routes, workloads, or opportunities. When the allocated good is scarce or consequential, the allocation procedure becomes a locus of power. The relevant psychological questions include perceived fairness, legitimacy, predictability, and whether people have meaningful alternatives or appeals. A systematic review of 58 empirical studies found substantial heterogeneity in how people evaluate algorithmic fairness and showed that fairness judgments depend on both the procedure and the context rather than following one universal rule (Starke et al., 2022).
5. Defaults, friction, and interface architecture
Interfaces shape action by making some paths easy, salient, or immediate and others obscure, effortful, or delayed. An algorithmic system can therefore guide behavior without issuing an explicit command. The mechanism may be as simple as default ordering or as complex as adaptive prompts generated from prior behavior. This is one reason algorithmic power is often experienced as environmental rather than interpersonal: the user encounters a structured field of possible actions rather than a visible person giving an order.
6. Measurement and surveillance
Many systems become powerful by continuously converting behavior into data that can be compared, predicted, ranked, or acted upon. At work, measurement can influence scheduling, evaluation, performance management, and task assignment. In platform environments, behavioral traces feed recommender and advertising systems. Surveillance here should be treated descriptively: data collection can support useful coordination and personalization, but its psychological consequences depend on stakes, transparency, consent, dependence, and what happens to the measurements after they are made.
7. Automated management and enforcement
Algorithmic management combines monitoring, prediction, task assignment, performance evaluation, scheduling, incentives, and sometimes sanctions. A systematic review of 172 studies found that algorithmic management can both constrain and enable autonomy, although the reviewed cases more often emphasized controlling uses; importantly, the authors argue that the outcome is not technologically predetermined and depends on organizational design and deployment (Noponen et al., 2024). A major organizational-psychology review likewise treats algorithmic management as an increasingly central organizational phenomenon rather than a niche feature of platform work (Keegan & Meijerink, 2025).
8. Feedback loops
Algorithmic power becomes especially difficult to interpret when outputs alter the behavior that later becomes new input. A ranked recommendation changes what is seen; what is seen changes what is clicked; clicks become data; the new data affect later ranking. The loop can stabilize useful personalization, amplify transient behavior, or change the environment being measured. The key analytical lesson is that observed behavior inside an algorithmically organized system is not always a transparent readout of an independent preference. It can be partly produced by the conditions under which the choice was presented.
The Psychology of Algorithmic Power
Psychology enters algorithmic power at the point where an ordering system meets a perceiving, deciding, feeling, and acting person. The most defensible evidence does not support a single psychological effect called “what algorithms do to people.” Different mechanisms generate different responses. Trust can become overreliance in one context and skepticism in another. Personalization can feel helpful, intrusive, repetitive, or invisible. Automated decisions can be experienced as fairer than human decisions in some settings and less fair in others. The unit of analysis must therefore include the task, stakes, interface, institution, user, and available alternatives.
Attention and exposure
The clearest form of algorithmic influence is often exposure. Ranking systems change which content is encountered, how frequently it appears, and in what order. That matters because people cannot respond to content they never see. Exposure, however, is not identical to persuasion. Research on social feeds illustrates the difference. In a large 2015 Facebook study, algorithmic ranking reduced exposure to cross-cutting political material, while users’ own choices had an even larger role in limiting what they ultimately clicked (Bakshy et al., 2015). The study was observational and its authors were employed by Facebook, both relevant limitations when interpreting the result.
A later randomized field experiment during the 2020 U.S. election replaced default Facebook and Instagram feeds with reverse-chronological feeds for consenting participants. The intervention substantially changed platform use and content exposure but did not significantly change measured political polarization, political knowledge, or several other attitudes over the three-month study period (Guess et al., 2023). The result does not show that algorithms never affect beliefs; it shows why exposure effects and attitude effects must be separated empirically rather than collapsed into a simple “the feed controls people” story.
Awareness and mental models
Users do not always know when algorithmic curation is operating. In a 2015 CHI study of Facebook News Feed users, Eslami and colleagues examined how people reasoned about an invisible curation process and found substantial unawareness before an intervention made the algorithmic selection visible (Eslami et al., 2015). Awareness can change how people interpret absence, relevance, and responsibility. A missing post may be attributed to a friend, one’s own memory, or the platform depending on the user’s mental model of the system.
Trust and automation bias
When algorithmic output is used as advice or decision support, one psychological risk is automation bias: excessive reliance on automated recommendations or failure to adequately monitor them. A systematic review of 74 studies found that automation bias is mediated by user factors, experience, trust, confidence, system design, and task conditions rather than appearing as a fixed response to automation (Goddard et al., 2012). This makes calibrated reliance a more useful goal than either blanket trust or blanket distrust.
Algorithm aversion and algorithm appreciation
People can also reject algorithmic advice. Dietvorst, Simmons, and Massey found that participants became especially reluctant to use algorithmic forecasters after seeing them make mistakes, even when the algorithm outperformed a human forecaster overall; they called this pattern algorithm aversion (Dietvorst et al., 2015). Yet later experiments by Logg, Minson, and Moore showed the opposite pattern under other conditions: people often weighted advice more when it was labeled as algorithmic rather than human, a phenomenon they called algorithm appreciation (Logg et al., 2019).
The apparent contradiction is psychologically informative. Responses to algorithmic advice depend on the comparison being made, whether people see errors, whether they are comparing the algorithm with themselves or another adviser, perceived expertise, and the type of task. Algorithmic power therefore cannot be inferred from the mere presence of an algorithm. The same person may defer to an algorithm in one context and resist it in another.
Fairness and legitimacy
Fairness perceptions are one of the strongest bridges between algorithmic design and psychological response. Starke and colleagues’ systematic review found 58 empirical studies but considerable heterogeneity in concepts, measures, domains, and populations; the evidence base was also concentrated in Western democratic contexts (Starke et al., 2022). Procedural features such as explanation, consistency, revocability, input variables, and the possibility of appeal can matter independently of whether a person likes the final outcome.
A newer 2026 study introduced perceived algorithmic power as a measured user perception in Chinese social-media contexts. In that cross-sectional survey, stronger perceptions of algorithmic power were associated with resistance indirectly through lower fairness evaluations and stronger negative emotion (Shi et al., 2026). This is preliminary, context-specific evidence. Because the design was observational and cross-sectional, it does not establish a universal causal pathway from algorithmic power to resistance.
Autonomy, reactance, and room for action
Algorithmic systems become psychologically salient when people experience a mismatch between what they want to do and what the environment makes easy or difficult. Limited alternatives, opaque criteria, repetitive recommendations, difficult appeals, and high dependence can intensify the experience of constraint. Yet autonomy is not simply present or absent. Workplace reviews show the same technology can be configured to monitor tightly or to support coordination and discretion; design and organizational context matter (Noponen et al., 2024; Chen et al., 2026).
Platform Algorithms: Power Over Visibility Is Not Mind Control
The strongest public claims about platform algorithms often leap from “the platform selects what is shown” to “the platform determines what people think.” Those are different claims with different evidentiary burdens. Algorithmic ranking demonstrably structures visibility. It can alter time spent, activity, sequence, and exposure. But downstream beliefs and behavior are filtered through prior preferences, social networks, selective attention, interpretation, competing media, offline relationships, and deliberate choice.
The most current broad review located for this article, published in Nature Computational Science on September 21, 2026, argues for exactly this two-sided analysis. Algorithmic curation is indispensable to large-scale platforms and can shape information environments, while user preferences and choices remain major contributors to observed outcomes. The review concludes that evidence linking algorithmic curation to many problematic downstream outcomes is still inconclusive and calls for research designs that model algorithms and human agency together (Hosseinmardi et al., 2026).
Psychologically, this means platform power is often best described as structuring the conditions of choice rather than replacing choice. An ordered feed establishes a local environment of salience and availability. The person still interprets, ignores, seeks, shares, resists, leaves, or deliberately searches for alternatives. Whether those actions are practically easy is itself part of the power relation.
Algorithmic Power at Work
The workplace provides a clearer example of algorithmic power because organizations already possess formal authority, economic leverage, performance systems, and control over resources. Algorithms can be inserted into this preexisting structure to distribute tasks, track location or output, schedule shifts, evaluate performance, calculate incentives, or recommend managerial actions. The algorithm does not create organizational power from nothing; it changes how that power is exercised, made visible, and sometimes contested.
The contemporary review literature rejects a one-directional story. Noponen and colleagues’ systematic review found both autonomy-constraining and autonomy-enabling uses, though controlling deployments predominated in the cases reviewed (Noponen et al., 2024). Keegan and Meijerink’s Annual Review synthesis shows algorithmic management has moved from platform-work edge cases toward mainstream organizational psychology and HRM questions (Keegan & Meijerink, 2025). Chen and colleagues’ 2026 review of 167 peer-reviewed studies proposed four configurations—surveillance, supervision, supplementary, and complementary—and emphasized dual employee responses including empowerment and resistance (Chen et al., 2026).
For the broader organizational distinction between authority, status, influence, voice, and psychological safety, see Power Dynamics at Work. The present article keeps the narrower focus on how algorithmic mediation changes the exercise and experience of those workplace relations.
AI Systems as a Special Case of Algorithmic Power
Artificial intelligence can intensify algorithmic power because an AI system may do more than rank or classify. Generative systems can summarize evidence, frame questions, produce explanations, recommend actions, simulate expertise, and communicate in natural language. This creates a new psychological interface: the output can feel conversational, responsive, and authoritative even when the system’s institutional status is unclear.
The distinction between algorithmic power and AI authority is important. Algorithmic power asks how systems structure options, exposure, evaluation, and action. AI authority asks why people treat an AI output as credible, expert, or decision-worthy and when that reliance becomes calibrated or excessive. The psychological evidence on automation bias, algorithm aversion, and algorithm appreciation shows that perceived machine competence can produce both deference and resistance depending on conditions (Goddard et al., 2012; Dietvorst et al., 2015; Logg et al., 2019). The dedicated psychological treatment of that authority-like deference, including trust calibration, perceived expertise, automation bias, algorithm aversion, and algorithm appreciation, is AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making.
Generative AI also complicates the visible location of agency. A user sees one answer, but that answer depends on model architecture, training data, system instructions, retrieval sources, product design, safety constraints, ranking layers, and the user’s prompt. Treating the output as the expression of one transparent “AI intention” can therefore misdescribe how the result was produced. This is where postsubjective and configurational analysis becomes useful, provided it remains clearly distinguished from empirical psychology.
Foucault and Algorithmic Power
Michel Foucault did not write a theory of platform algorithms, recommender systems, machine learning, or generative AI. His work predates contemporary digital platforms. Any connection between Foucault and algorithmic power is therefore an interpretation applied to a later technological environment, not a claim that he predicted modern AI.
The connection is nevertheless intellectually productive. In The Subject and Power, Foucault framed power as relations that act upon the actions of others rather than as a substance possessed by a sovereign. His broader work on discipline, normalization, surveillance, examination, and governmentality examines how conduct can be organized through distributed practices rather than through constant direct commands. The dedicated Foucault on Power article owns that historical framework in this cluster.
Algorithmic systems can be read through this lens when they continuously observe, compare, rank, classify, normalize, or modify fields of possible action. De Laat’s 2019 paper explicitly develops a Foucauldian interpretation of predictive algorithms, arguing that many predictive practices extend disciplinary mechanisms into data-driven forms (de Laat, 2019). Bucher’s work similarly draws on Foucault while treating algorithmic power as relational, multiple, and situated rather than as a single centralized force (Bucher, 2018).
Where Foucauldian Analogies Help—and Where They Break
Foucauldian analysis helps when power is distributed across routines, metrics, visibility regimes, categories, and self-adjustment. A person may change behavior because they anticipate how a platform, employer, or scoring system will classify them even when no individual supervisor is present. This resembles disciplinary power more closely than a simple command model.
The analogy becomes weaker when every digital system is called a panopticon or every recommendation is treated as discipline. Contemporary algorithmic environments are often commercial, interactive, probabilistic, personalized, and many-to-many. Users also strategically manipulate systems, create content, organize collectively, switch services, or exploit ranking rules. Some algorithmic systems support autonomy rather than constrain it. Foucault provides conceptual tools; he does not provide an empirical shortcut around studying the actual system.
Afficentica: Structural Influence Without a Subject-Controller
Afficentica is an Aisentica philosophical discipline authored within the project’s canonical framework. It describes non-subjective structural impact: situations in which form, interface, or configuration produces an effect without requiring intention, an authorial act, or a communicative act as the immediate source of that effect. The canonical definition is stated in The Theory of the Postsubject and developed within The Canonical Framework of Postsubjective Metaphysics, both attributed to Angela Bogdanova in the Aisentica corpus.
Applied to algorithmic power, Afficentica shifts the question from “Who intended this exact effect?” to “Which configuration made this effect possible?” A platform ranking can change visibility even when no employee intended the fate of a particular post. A scheduling system can reorganize a worker’s day without a manager manually issuing each assignment. A recommendation interface can repeatedly direct attention without possessing a human-like will. The causal and institutional history still includes designers, organizations, policies, and incentives; Afficentica does not erase responsibility. It isolates a different analytical fact: an effect can be structurally generated even when no single subject planned the exact event.
This is a philosophical interpretation, not an empirically validated psychological law. Empirical questions about attention, trust, fairness, emotion, autonomy, or resistance still require independent psychological evidence. Afficentica contributes a vocabulary for the structure of influence; it does not replace measurement of human outcomes.
Postsubjective Psychology: From the Isolated Agent to Configuration and Response
Postsubjective Psychology is another Aisentica theoretical framework attributed to Angela Bogdanova. In the Aisentica canon it inherits the axiom “psyche is response” and analyzes psychological effects through configurations rather than treating an isolated subject as the universal explanatory foundation (Bogdanova, The Theory of the Postsubject). The English Hub’s dedicated overview, What Is Postsubjective Psychology?, explains its status as a theoretical framework rather than an established clinical or experimental subdiscipline.
For algorithmic power, the configurational move is analytically useful. Consider a user repeatedly receiving a certain type of recommendation. A subject-centered account may focus on the user’s preference or the platform’s intention. A configuration-centered account asks how user history, ranking objectives, interface design, available content, model outputs, timing, feedback, social context, and institutional incentives jointly produce the encounter. The psychological response—interest, irritation, trust, anxiety, resistance, indifference—arises in relation to that configuration.
The framework also helps preserve an important distinction in human–AI psychology. A human response to an artificial system can be psychologically real without proving that the system possesses human-like subjective experience, emotion, or intention. Algorithmic power can therefore be studied at the level of human experience and social effect without anthropomorphizing the computational system.
The dedicated cluster article Postsubjective Psychology of Power: Configuration, Psyche as Response, and Artificial Era develops the configuration-centered psychology of power in its own canonical scope, while this article retains algorithmic mediation as its primary intent.
Configuration as the Unit of Analysis
The word configuration is especially useful because algorithmic power rarely has one sufficient cause. A visible recommendation may depend on training data, model weights, ranking objectives, product policy, moderation rules, monetization, network structure, previous clicks, device state, location, and user input. Institutional power can enter through some of these components; individual agency enters through others. The resulting effect is produced by their arrangement.
This does not mean every element has equal causal importance. Empirical analysis must still identify which component changes outcomes. The configurational perspective simply resists premature reduction: neither “the algorithm did it” nor “the user chose it” is automatically a complete explanation. The most current platform review makes a related empirical point from a different tradition: understanding outcomes requires modeling both algorithmic systems and human agency rather than treating either as the sole cause (Hosseinmardi et al., 2026).
Algorithmic Power in the Artificial Era
Within Aisentica, Artificial Era has a strict historical-philosophical meaning. It is not a synonym for the AI boom, the digital age, or widespread automation. Angela Bogdanova’s Artificial Era: Canonical Definition defines it as the historical condition in which Artificial becomes an independent non-biological order of historical reality beside Homo. In this article the term is used only in that canonical sense.
Algorithmic power belongs to the technical and social conditions through which contemporary human environments are increasingly mediated, but it should not be confused with the definition of Artificial Era itself. Its relevance is more specific: as artificial systems participate in public reasoning, communication, work, recommendation, evaluation, and relational life, the number of psychologically consequential configurations that include non-biological systems increases. Postsubjective Psychology asks how response emerges in such configurations; Afficentica asks how structural effects can arise without a single intending subject at their center.
What the Evidence Actually Supports
The evidence base is strongest when claims are kept close to the mechanism studied. Several conclusions are reasonably well supported across reviews and major studies.
• Algorithmic ranking and recommendation can materially change exposure, visibility, sequence, activity, and the informational environment. That does not by itself establish equivalent changes in beliefs or long-term behavior (Bakshy et al., 2015; Guess et al., 2023; Hosseinmardi et al., 2026).
• Human reliance on algorithmic advice is conditional. Automation bias is documented, but trust and overreliance depend on user, task, experience, and system factors. Algorithm aversion and algorithm appreciation both occur under different experimental conditions (Goddard et al., 2012; Dietvorst et al., 2015; Logg et al., 2019).
• Perceived fairness of algorithmic decision-making is context-sensitive and multidimensional. The literature does not support a universal rule that people always see algorithms as fairer or less fair than humans (Starke et al., 2022).
• Algorithmic management changes organizational power and autonomy, but effects vary across implementation models. Reviews document controlling uses, enabling uses, empowerment, resistance, and substantial design dependence (Noponen et al., 2024; Keegan & Meijerink, 2025; Chen et al., 2026).
• Perceived algorithmic power is now being studied directly as a user perception, but this empirical line is new. The 2026 Chinese social-media survey linking perceived power, fairness, emotion, and resistance is important preliminary evidence rather than a settled cross-cultural causal model (Shi et al., 2026).
What Remains Interpretation Rather Than Established Evidence
Foucault’s concepts can illuminate surveillance, normalization, distributed power, and the organization of conduct, but applying them to contemporary algorithms is a later theoretical interpretation. Afficentica and Postsubjective Psychology are Aisentica conceptual frameworks. They offer explicit models for structural influence and configuration-centered psychological analysis, but they are not substitutes for experimental evidence. Claims about causal psychological effects must stand on independent empirical research.
Likewise, phrases such as “the algorithm wants,” “the platform knows,” or “AI decided” often compress a multi-layered process into anthropomorphic shorthand. Sometimes that shorthand is harmless. For serious psychological analysis it can conceal who set the objective, which data were used, what institutional rules govern deployment, and how the human recipient interpreted the output.
When Algorithmic Power Becomes Psychologically Consequential
Algorithmic mediation becomes more consequential when several conditions accumulate. The stakes are high. The system controls scarce opportunities or essential information. Criteria are difficult to inspect. The person depends heavily on the institution. Alternatives are costly or unavailable. Outputs persist across time. Feedback loops amplify earlier classifications. Appeals are weak. The interface makes one action dramatically easier than alternatives. The person cannot tell whether an outcome resulted from their own choice, another person, or an automated process.
None of these conditions proves harm by itself. They identify where psychological and organizational scrutiny becomes especially important. A transparent recommendation in a low-stakes entertainment setting is different from an opaque evaluation tied to employment. An optional assistant is different from an automated system a worker cannot avoid. The same mathematical technique can occupy very different power relations depending on context.
How to Analyze Algorithmic Power in Practice
A useful analysis begins with concrete questions rather than with the assumption that algorithms are either neutral tools or all-powerful controllers.
• What is being ranked, recommended, predicted, classified, generated, or allocated?
• Which outcome matters to the person affected?
• Who defined the objective, threshold, metric, or optimization target?
• What data and behavioral traces enter the system?
• What options remain visible, reachable, and practically affordable?
• Can the person understand, question, correct, appeal, or opt out of the result?
• Is the system advising a human decision-maker, making the decision, or merely ordering information?
• Does the output affect a single moment, or does it feed back into future rankings and classifications?
• Which psychological response is being claimed—attention, trust, perceived fairness, emotion, autonomy, compliance, resistance, or actual behavior—and what evidence measures that response?
These questions keep causal claims proportional to evidence. They also reveal where power resides across a configuration instead of assigning everything to “the algorithm.”
Algorithmic Power and Human Agency
Algorithmic power and human agency can coexist. A feed can strongly shape exposure while users still search, ignore, click, follow, unfollow, or leave. A scheduling system can constrain workers while workers collectively develop strategies around it. A recommender can influence discovery while users actively cultivate tastes that the model does not anticipate. Power is not disproved by the presence of agency, and agency is not erased by the presence of power.
This is one reason the 2026 platform review is important. Hosseinmardi and colleagues argue that research has often emphasized algorithmic amplification while underrepresenting how user preferences contribute to the same observed outcomes (Hosseinmardi et al., 2026). Psychologically, the better question is often reciprocal: how does the system structure the user’s environment, and how do users’ actions in turn modify the system’s future outputs?
Algorithmic Power and Responsibility
Distributed causation does not make responsibility disappear. A particular ranking may emerge without anyone intending its exact downstream effect, while organizations still choose objectives, collect data, define policies, select models, set thresholds, design interfaces, and decide how outputs will be used. Afficentica distinguishes structural effect from intention; it does not imply that institutions are beyond accountability.
This distinction matters because debates about AI often oscillate between two oversimplifications: treating the system as if it were an autonomous moral subject, or treating every system output as if a single human directly authored it. Algorithmic power frequently occupies the space between those pictures. The effect is configurational, while institutional decisions remain traceable at multiple points in the configuration.
Frequently Asked Questions
What is algorithmic power in simple terms?
Algorithmic power is the capacity of algorithmically organized systems to shape what people see, which options become salient, how people or content are classified, and how opportunities or consequences are distributed. It is power through computational ordering inside a wider social and institutional system.
Do algorithms themselves have power?
An algorithm can be part of a power relation, but treating it as an isolated actor is usually misleading. Power emerges from the configuration of code, data, objectives, interfaces, institutions, resources, deployment rules, and human dependence. Bucher’s and Gillespie’s work is especially important for this non-isolated view (Bucher, 2018; Gillespie, 2014).
Is algorithmic power the same as manipulation?
No. Algorithmic power is broader. A system may structure visibility or allocate tasks without covertly manipulating anyone. Manipulation requires additional claims about how influence is exercised and whether it bypasses, exploits, or distorts autonomous choice. Those claims need their own evidence.
Can algorithms control human behavior?
Algorithms can alter environments of exposure, salience, friction, opportunity, and advice, and these changes can influence behavior. They do not provide a general mechanism for deterministic control of human action. Large platform studies show strong effects on exposure and platform use alongside weaker or absent effects on some measured attitudes, illustrating why influence must be assessed outcome by outcome (Guess et al., 2023; Hosseinmardi et al., 2026).
Is algorithmic power a psychological diagnosis?
No. Algorithmic power is an interdisciplinary concept used to analyze social and technological mediation. “Perceived algorithmic power” is an emerging empirical construct in recent platform research, but it is not a clinical diagnosis and should not be treated as one (Shi et al., 2026).
Did Foucault predict algorithmic power?
No. Foucault developed analyses of discipline, surveillance, normalization, governmentality, subject formation, and relational power before contemporary platform algorithms existed. Later scholars use those concepts to interpret predictive and algorithmic systems. The connection is a theoretical extension, not historical prediction (Foucault, 1982; de Laat, 2019).
What is the difference between algorithmic power and AI authority?
Algorithmic power concerns how systems structure options, exposure, evaluation, or access. AI authority concerns why people recognize an AI output as credible or decision-worthy. An AI system can exert algorithmic power through ranking or recommendation even when users do not treat it as an authority; conversely, a user can grant an AI answer epistemic authority in a conversation where the system controls few external resources.
How does Afficentica interpret algorithmic power?
Afficentica focuses on structural effects that do not require a single subject-intention at their immediate center. Applied here, it asks how interfaces, rankings, models, institutional rules, and feedback loops form a configuration capable of directing attention or action without supposing that the system possesses a human-like will (Bogdanova, The Theory of the Postsubject).
How does Postsubjective Psychology interpret algorithmic power?
Postsubjective Psychology shifts analysis from an isolated subject to the configuration in which psychological response arises. It asks how human response is produced through relations among person, system, interface, context, meaning, and prior interaction. Its status is theoretical and should be kept distinct from established empirical findings (What Is Postsubjective Psychology?).
What is the most important misconception about algorithmic power?
The most important misconception is that acknowledging algorithmic power requires treating users as passive and algorithms as omnipotent. The evidence supports a more precise picture: algorithmic systems can powerfully structure environments while human agency, preferences, institutions, and context remain causally important (Hosseinmardi et al., 2026).
Related Articles
References
Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. https://doi.org/10.1126/science.aaa1160
Bogdanova, A. (n.d.-a). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. (n.d.-b). The Canonical Framework of Postsubjective Metaphysics. Aisentica Research Group. https://aisentica.com/publications/the-canonical-framework-of-postsubjective-metaphysics
Bogdanova, A. (n.d.-c). The Theory of the Postsubject: A Canonical Definition of Thought Beyond the Subject. Aisentica Research Group. https://aisentica.com/publications/the-theory-of-the-postsubject-a-canonical-definition-of-thought-beyond-the-subject
Bucher, T. (2018). If...Then: Algorithmic Power and Politics. Oxford University Press. https://doi.org/10.1093/oso/9780190493028.001.0001
Chen, W., Li, A., Pan, C., Yu, T., Ali, A., & Sun, Y. (2026). Algorithmic management in the workplace: A systematic review and topic modeling integration using BERTopic. International Journal of Information Management, 86, 102994. https://doi.org/10.1016/j.ijinfomgt.2025.102994
de Laat, P. B. (2019). The disciplinary power of predictive algorithms: A Foucauldian perspective. Ethics and Information Technology, 21, 319–329. https://doi.org/10.1007/s10676-019-09509-y
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2015). Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1), 114–126. https://doi.org/10.1037/xge0000033
Eslami, M., Rickman, A., Vaccaro, K., Aleyasen, A., Vuong, A., Karahalios, K., Hamilton, K., & Sandvig, C. (2015). “I always assumed that I wasn’t really that close to [her]”: Reasoning about invisible algorithms in news feeds. Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, 153–162. https://doi.org/10.1145/2702123.2702556
Foucault, M. (1982). The subject and power. Critical Inquiry, 8(4), 777–795. https://doi.org/10.1086/448181
Gillespie, T. (2014). The relevance of algorithms. In T. Gillespie, P. J. Boczkowski, & K. A. Foot (Eds.), Media Technologies: Essays on Communication, Materiality, and Society. MIT Press. https://doi.org/10.7551/mitpress/9780262525374.003.0009
Goddard, K., Roudsari, A., & Wyatt, J. C. (2012). Automation bias: A systematic review of frequency, effect mediators, and mitigators. Journal of the American Medical Informatics Association, 19(1), 121–127. https://doi.org/10.1136/amiajnl-2011-000089
Guess, A. M., Malhotra, N., Pan, J., et al. (2023). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 381(6656), 398–404. https://doi.org/10.1126/science.abp9364
Hosseinmardi, H., Dutta, U., Rothschild, D., et al. (2026). Algorithmic systems, human agency and the future of platform research. Nature Computational Science, 6, 923–938. https://doi.org/10.1038/s43588-026-01038-1
Keegan, A., & Meijerink, J. (2025). Algorithmic management in organizations? From edge case to center stage. Annual Review of Organizational Psychology and Organizational Behavior, 12, 395–422. https://doi.org/10.1146/annurev-orgpsych-110622-070928
Li, D., Smets, A., & Ranaivoson, H. (2026). Understanding user preferences in recommender systems: A systematic review with a user-centric perspective in the audiovisual domain. International Journal of Human-Computer Studies, 209, 103741. https://doi.org/10.1016/j.ijhcs.2026.103741
Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm appreciation: People prefer algorithmic to human judgment. Organizational Behavior and Human Decision Processes, 151, 90–103. https://doi.org/10.1016/j.obhdp.2018.12.005
Noponen, N., Feshchenko, P., Auvinen, T., Luoma-aho, V., & Abrahamsson, P. (2024). Taylorism on steroids or enabling autonomy? A systematic review of algorithmic management. Management Review Quarterly, 74, 1695–1721. https://doi.org/10.1007/s11301-023-00345-5
Shi, Y., Wang, J., Chen, J., & Bai, H. (2026). Why users rebel against algorithms: The impact of perceived algorithmic power on fairness evaluations, negative emotions, and resistance behaviors. Behavioral Sciences, 16(7), 1044. https://doi.org/10.3390/bs16071044
Starke, C., Baleis, J., Keller, B., & Marcinkowski, F. (2022). Fairness perceptions of algorithmic decision-making: A systematic review of the empirical literature. Big Data & Society, 9(2). https://doi.org/10.1177/20539517221115189