Resistance to AI in the Artificial Era: Autonomy, Control, Reactance, and Human Agency
Author: Ukrainian Psychological Hub · Published: September 26, 2026 · Editorial Policy
Resistance to artificial intelligence is not one reaction. It can be a defense of autonomy, an attempt to restore control, distrust after observing errors, reluctance to delegate judgment, a response to surveillance, a professional identity threat, a status threat, anxiety about replacement, opposition to opaque institutional power, or a broader reaction to the loss of human exclusivity in functions once treated as distinctively human. These mechanisms can overlap, but they should not be collapsed into a single story about people being afraid of technology.
The best available evidence supports that plural view. A major perspective in Nature Human Behaviour organizes resistance to AI around several barriers, including opacity, emotionlessness, rigidity, autonomy, and group membership, while emphasizing that both AI characteristics and user characteristics matter (De Freitas et al., 2023). A large meta-analysis published in the Journal of Marketing synthesized 287 effect sizes from 136 studies representing 119,358 participants. It found a small overall negative response to AI relative to human alternatives, but also very high heterogeneity and evidence that acceptance has increased over time (Li, Lai, & Wang, 2026). There is no universal psychology of “AI resistance” that applies equally to every person, system, task, and institution.
This article asks a narrower and more useful question: when people resist AI, what psychological processes can make that resistance intelligible? The central mechanisms are autonomy, perceived control, psychological reactance, distrust, identity and status threat, and human agency. The article then places those evidence-based mechanisms alongside Angela Bogdanova’s Aisentica framework. In that framework, resistance can sometimes acquire a broader historical meaning when AI participates in functions over which Homo had assumed exclusive ownership. Aisentica calls that specific theoretical pattern Subject-Monopoly Reaction. It is a philosophical proposition, not a clinical diagnosis, a validated psychometric scale, or an empirical explanation for every objection to AI.
What Does “Resistance to AI” Mean?
Resistance to AI can describe an attitude, a feeling, an intention, or a behavior. A person may distrust an AI recommendation, refuse to use an AI system, demand human review, oppose an autonomous feature, criticize algorithmic monitoring, avoid AI-generated services, insist on retaining decision authority, or support regulation that constrains where and how AI may be used. These responses have different causes and different implications.
The distinction matters because “acceptance” is also multidimensional. Someone may believe an AI system is useful while disliking its role. Someone may trust its accuracy while refusing to let it make a final decision. Someone may use AI frequently while opposing its use in hiring, medicine, education, or surveillance. Someone may feel professionally threatened by AI while still wanting to collaborate with it. Grundke’s experiments are a useful example: when machines outperformed humans on verbal-creative tasks, participants reported greater status threat, yet greater perceived usefulness could coexist with willingness to interact with the machine (Grundke, 2024).
Resistance is therefore better understood as a family of context-dependent responses than as a stable anti-AI trait. It is also not a mental disorder. Ordinary skepticism, discomfort, anger, caution, refusal, or demand for oversight should be interpreted in relation to what the system does, what freedoms are affected, what risks are present, and what kind of authority the system is being given.
Resistance Is Not Automatically Irrational
Some resistance reflects cognitive bias or inaccurate beliefs. People can reject algorithms after seeing them make errors even when the same algorithms outperform human forecasters. Dietvorst, Simmons, and Massey’s classic experiments helped establish this pattern as algorithm aversion (Dietvorst, Simmons, & Massey, 2015). Yet the broader literature also shows algorithm appreciation: in some settings people weigh algorithmic advice more heavily than comparable human advice (Logg, Minson, & Moore, 2019). A systematic review of 80 empirical studies found that algorithm aversion depends on algorithm, individual, task, and higher-level contextual factors rather than arising from one universal bias (Mahmud et al., 2022).
Other resistance is a response to real conditions. If an AI system is difficult to contest, monitors workers continuously, acts with broad permissions, produces consequential errors, obscures responsibility, uses personal data in unwanted ways, or makes decisions without meaningful appeal, asking for more human control can be rational risk management. The psychology of resistance has to preserve this distinction. A reaction can be psychologically mediated and still respond to a genuine institutional or technical problem.
This is especially important because the same outward behavior can arise from different causes. Refusing an AI diagnosis could reflect algorithm aversion, concern about individualized care, distrust of the institution deploying the system, lack of informed consent, or a reasonable desire for a clinician to remain responsible. Refusing an AI writing tool could reflect uncertainty about copyright, concern about skill erosion, professional identity threat, or a belief that authorship should remain human. The behavior alone does not reveal the mechanism.
AI Anxiety and Prevention Focus: Affective Resistance Is One Route
Resistance can also be shaped by anxiety and motivational orientation. In a 2026 study of 259 undergraduate entrepreneurship students in the United States, prevention focus was associated with more negative attitudes toward AI, and four dimensions of AI anxiety—learning anxiety, job-replacement anxiety, configuration anxiety, and sociotechnical-blindness anxiety—partly mediated that relationship (Sapru, 2026).
The study is useful because it shows that affective and motivational processes can contribute to negative AI attitudes, while its sample and domain sharply limit generalization. AI anxiety should therefore be treated as one possible route into resistance, not as an explanation for all skepticism, refusal, or opposition. A person can resist AI without being anxious, and anxiety can coexist with realistic concerns about employment, privacy, competence, control, or institutional power.
Autonomy: Why AI Can Feel Like a Threat to Self-Direction
Autonomy concerns whether a person experiences their actions as self-endorsed rather than imposed. In self-determination theory, autonomy is one of three basic psychological needs alongside competence and relatedness (Ryan & Deci, 2000). Research on choice and control likewise shows that the opportunity to influence outcomes can be intrinsically valued and that loss of control is often aversive (Leotti, Iyengar, & Ochsner, 2010).
AI can affect autonomy in several ways. It can narrow options through ranking or recommendation. It can make a choice on a person’s behalf. It can predict what a person is likely to do and arrange the environment around that prediction. It can monitor behavior and evaluate performance. It can automate a workflow that previously required active human judgment. It can also change the default from “the person decides unless they delegate” to “the system proceeds unless the person interrupts.”
That last shift is psychologically important. A user may technically retain a cancel button while experiencing the system as the actual organizer of action. Autonomy is therefore not reducible to the mere presence of a nominal human-in-the-loop. It depends on whether the person has meaningful opportunities to understand, choose, revise, reject, appeal, or redirect what is happening.
AI-specific studies support this mechanism. In a 2025 study of 297 users, perceived AI autonomy was positively associated with perceived threat to freedom and psychological reactance. The relationship was not simple: perceived personalization could offset some of the negative effect, and people reporting higher user agency experienced stronger freedom threat as perceived AI autonomy increased (Oh, Nah, & Yang, 2025). This is one reason highly agentic users may resist systems precisely when those systems begin deciding too much for them.
Everyday context changes the response. In a design-fiction survey of 328 participants, social-media AI was associated with lower perceived autonomy and greater reactance than several other applications, while explanations improved autonomy and reduced reactance in a navigation scenario. The same intervention did not work uniformly across contexts (Sankaran et al., 2021). An explanation can help when it restores intelligibility or choice, but “explainability” is not a universal psychological antidote.
Control: The Difference Between Having Influence and Merely Feeling Reassured
Autonomy and control overlap, but they are not identical. Autonomy concerns self-direction and endorsement. Control concerns the capacity to influence what happens. A person can voluntarily delegate a task and remain autonomous even while giving up moment-to-moment control. Conversely, a person can be given many interface controls while feeling that the important decision has already been made elsewhere.
Research on algorithm aversion shows that even a small amount of genuine influence can matter. In a series of forecasting studies, people were more willing to use imperfect algorithms when allowed to modify the algorithm’s output, even when the permitted modification was small. The researchers concluded that some control over the outcome, rather than maximal control, increased willingness to rely on the system (Dietvorst, Simmons, & Massey, 2018).
Recent AI research sharpens the point. Rahman and Liang’s 2026 randomized experiment found that perceived lack of autonomy could increase AI aversion through distrust, with task stakes moderating that indirect relationship (Rahman & Liang, 2026). The psychological demand for control becomes especially consequential when an error would matter.
Control also has an objective side. A user may feel reassured by a conversational interface, an explanation, or a confirmation screen while possessing little effective power to stop or reverse an action. For consequential systems, meaningful control asks concrete questions. Can the person refuse? Can an action be interrupted before completion? Can it be reversed? Can permissions be narrowed? Can an appeal change the result? Can a responsible human inspect the relevant evidence? Can the system’s operational reach exceed what the interface implies?
These questions are not merely subjective preferences. They are also governance questions. Article 14 of the European Union’s AI Act requires high-risk AI systems to be designed so that they can be effectively overseen by natural persons and makes oversight proportional to risk, autonomy, and context of use (European Union, 2024). A scholarly review of human control in AI similarly argues that control has to be understood through the real interaction between human and system behavior during deployment, rather than by assuming that nominal supervision guarantees effective control (Tsamados, Floridi, & Taddeo, 2025).
This creates a crucial boundary for psychological analysis: a desire for human control can be a defensive response, a legitimate safety requirement, or both at the same time.
Psychological Reactance: When AI Threatens Freedom
Psychological reactance is the motivational response that can arise when a person perceives a valued freedom as threatened or removed. It is narrower than general resistance. People can resist AI for reasons unrelated to reactance, and reactance can occur in situations that have nothing to do with AI.
The general evidence for reactance is robust enough to matter here. A 2026 meta-analysis covering 33 studies and 146 effect sizes found that highly freedom-threatening language produced more anger, more negative cognitions, and more psychological reactance than less threatening language. Anger and negative cognitions were in turn associated with weaker persuasion outcomes (Li & Shi, 2026). This does not mean that every rule or direct instruction backfires. It means that perceived restriction can itself become motivationally salient.
AI can generate reactance through more than words. A recommendation system can suppress alternatives. An automated workflow can remove a decision point. A workplace system can turn observation into continuous evaluation. An AI assistant can move from suggesting to acting. A platform can make opting out difficult while presenting the experience as personalized convenience.
Algorithmic surveillance illustrates the mechanism clearly. Across four experiments, Schlund and Zitek found that algorithmic surveillance was often associated with lower perceived autonomy and greater resistance than human surveillance. Participants criticized the surveillance more, sometimes performed worse, and reported stronger intentions to resist; framing surveillance as developmental rather than evaluative reduced some of these effects (Schlund & Zitek, 2024).
There is also an important complication. People do not always recognize autonomy restriction. Görlitz and Rosenthal-von der Pütten argue that opaque AI-mediated paternalism may create a “reactance deficit”: if people cannot see how a system is shaping or narrowing their choices, the ordinary protective reaction to lost autonomy may fail to activate (Görlitz & Rosenthal-von der Pütten, 2026). In that case, low resistance would not necessarily mean high autonomy. It could mean that the restriction is difficult to perceive.
Distrust and Algorithm Aversion: Resistance to the Source of Advice
Some resistance is epistemic rather than autonomy-based. The person asks whether the system deserves reliance.
Algorithm aversion describes patterns in which people underuse algorithmic judgment, sometimes even when it performs well. The classic Dietvorst studies showed that seeing an algorithm err can disproportionately reduce reliance on it (Dietvorst, Simmons, & Massey, 2015). Yet task type matters. Castelo, Bos, and Lehmann found lower reliance on algorithms for tasks perceived as subjective than for tasks perceived as objective (Castelo, Bos, & Lehmann, 2019). In medicine, resistance can be driven by the belief that AI will fail to account for a person’s uniqueness; framing AI as supporting rather than replacing a human provider reduced resistance in Longoni, Bonezzi, and Morewedge’s studies (Longoni, Bonezzi, & Morewedge, 2019).
The reverse pattern also exists. People sometimes prefer algorithmic advice to human advice, which is why the literature distinguishes algorithm aversion from algorithm appreciation (Logg, Minson, & Moore, 2019). Resistance to AI should therefore never be inferred from the mere fact that the source is artificial.
This domain belongs partly to the English Hub’s dedicated article on AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making. That article owns the trust, expertise, automation-bias, overreliance, and calibrated-reliance intent. Here distrust matters only as one route into resistance: a person may refuse AI because the system has not earned epistemic authority.
Identity Threat: When AI Changes the Answer to “Who Am I?”
Autonomy threat asks whether I am still directing my action. Identity threat asks whether a valued definition of self is being destabilized.
Generative AI makes this especially visible because it operates in domains closely tied to competence and self-description: writing, analysis, communication, programming, design, research, and creative work. Zhou, Lu, and Chen combined qualitative interviews with a survey of 405 users and found that generative AI’s creative, analytical, and communication affordances could contribute to AI identity threat, which was associated with resistance behavior. Perceived AI autonomy and user self-identity moderated parts of the model (Zhou, Lu, & Chen, 2025).
Professional identity offers a more concrete case. Jussupow, Spohrer, and Heinzl studied medical students and physicians and distinguished threats to professional recognition from threats to professional capability. Both dimensions contributed to resistance attitudes toward medical AI, with different pathways and different effects across experience and perceived temporal distance (Jussupow, Spohrer, & Heinzl, 2022).
This mechanism cannot be reduced to fear of unemployment. A profession is also a social identity, a competence structure, a source of status, and a narrative about what one uniquely contributes. AI can threaten that narrative before it eliminates a single job.
Status Threat and Human Uniqueness
Status threat concerns relative standing. Human-uniqueness threat concerns a boundary between humans and nonhuman systems. They can overlap without being the same.
In two experiments, Grundke found that machines outperforming humans on verbal-creative tasks increased status threat. Importantly, threat did not simply produce avoidance; usefulness could coexist with threat and willingness to interact (Grundke, 2024). This is an important correction to any model in which discomfort automatically predicts rejection.
Research on autonomous systems also shows that perceived autonomy can activate both realistic and identity-relevant concerns. Złotowski, Yogeeswaran, and Bartneck found that robots framed as autonomous produced greater perceived threats to jobs, resources, safety, identity, and human distinctiveness, which contributed to more negative attitudes and opposition to robotics research (Złotowski, Yogeeswaran, & Bartneck, 2017). Stein, Liebold, and Ohler likewise found that situational control and concerns about human uniqueness predicted threat experience and aversion toward an allegedly autonomous system (Stein, Liebold, & Ohler, 2019).
These findings connect resistance to a wider problem examined in the Era cluster: the psychology of human decentering. Resistance becomes more symbolically intense when AI does not merely perform a task but enters a domain people use to define human distinctiveness.
Human Agency: Who Still Governs the Course of Action?
Human agency is often invoked vaguely in AI debates. Here it has a practical psychological meaning: whether people remain able to participate in forming goals, selecting actions, evaluating outputs, revising plans, refusing recommendations, and taking responsibility for consequential decisions.
Agency is related to autonomy and control while remaining distinct from both. Autonomy concerns self-endorsed action. Control concerns influence over outcomes. Agency concerns the person’s ongoing role in generating and governing a course of action.
This distinction matters because AI can preserve one dimension while weakening another. A person may autonomously choose to delegate a task but then lose effective control once the system begins acting. A person may retain a veto while doing so little active evaluation that the system becomes the de facto organizer of the cognitive process. A person may also use AI extensively while preserving agency by setting goals, checking evidence, comparing alternatives, revising outputs, and making the final commitment.
The English Hub article Cognitive Agency in the Artificial Era: Who Governs the Thinking Process? develops this issue at the level of thinking itself. For the psychology of resistance, the implication is narrower: people may resist AI when they interpret delegation as loss of authorship over their own decisions, work, or judgment.
Why Agentic and Autonomous AI Can Produce a Different Reaction
A calculator, a recommender, a chatbot, and an agent that can act across software systems do not create the same psychological relationship. As AI moves from producing information toward selecting and executing actions, resistance can shift from “Do I trust the answer?” to “Who is directing what happens?”
The 2026 meta-analysis by Li, Lai, and Wang is especially useful here because it distinguishes AI as a tool from AI as an agent. Across the literature they synthesized, acceptance depended on characteristics of the AI, user, and task, with agentic qualities such as autonomy and role becoming especially relevant as AI is experienced less like a passive instrument and more like an acting counterpart (Li, Lai, & Wang, 2026).
At the same time, higher autonomy does not automatically destroy trust. A 2026 study of agentic AI in routine productivity applications found that more autonomous conditions could improve throughput and reduce workload without a general decline in trust, while desire for control influenced trust formation (Geninatti Cossatin et al., 2026). The critical issue is the configuration: task stakes, reversibility, permissions, predictability, human intervention, and what the system is actually allowed to do.
The dedicated English Hub article Why Autonomous AI Feels More Dangerous Than Intelligent AI: Psychology of Control, Agency, and Risk owns that comparative risk-perception question. The present article uses autonomy differently: as one mechanism that can contribute to resistance across many kinds of AI.
Tool, Assistant, Partner, Authority, Agent: Role Expectations Matter
People do not encounter “AI” in the abstract. They encounter a system assigned a role.
A tool is expected to remain under user direction. An assistant can recommend while leaving authority with the user. A partner implies reciprocal contribution. An authority is treated as a source to which judgment may defer. An agent may independently select or execute steps toward a goal.
Resistance can emerge when the system’s actual role exceeds the role the person expected. A product presented as a writing assistant may begin making structural decisions the user experiences as authorship. A decision-support system may become a de facto decision-maker because its recommendations are rarely challenged. A recommendation system may become a behavioral environment because it controls which options are visible. An agent may inherit broad permissions that turn advice into action.
This is why role language is psychologically consequential. Calling a system a “tool” can reassure users, but labels do not override operational reality. If a system independently ranks, selects, predicts, filters, recommends, or acts, the psychological response will be shaped by those functions even when marketing language presents the system as passive.
The Algorithmic Era and the Control of Human Behavior
Long before generative AI, algorithmic systems were already shaping behavior through prediction, ranking, recommendation, scoring, and visibility. The English Hub article Algorithmic Era and Psychology: How Prediction, Ranking, and Recommendation Shape Human Behavior examines this historical layer.
That distinction matters for resistance. Some objections directed at “AI” are actually objections to algorithmic governance: being ranked, monitored, scored, filtered, predicted, or managed through systems that influence opportunity and attention. Generative or agentic AI can intensify those concerns, but it did not create them.
Algorithmic surveillance research makes the point experimentally: resistance is shaped not only by what the system knows but by what the system is allowed to do with that knowledge and how the person is positioned within the evaluative relationship (Schlund & Zitek, 2024). A complete psychology of AI resistance therefore has to include institutional power as well as individual attitudes.
When Resistance Protects Human Agency
A common mistake in technology adoption research is to treat lower resistance as the obvious goal. Sometimes it is. If people refuse a demonstrably useful tool because they have inaccurate assumptions about what it can do, better information, experience, or design can improve outcomes.
Sometimes resistance is protective. It can force an institution to justify a system, preserve an appeal route, limit surveillance, disclose automated decision-making, maintain professional responsibility, or prevent the transfer of high-stakes authority to a system that cannot be adequately audited.
Psychologically, the key question is not “How do we eliminate resistance?” It is “What is the resistance responding to?”
If the source is opacity, improve intelligibility and evidence access. If the source is an unnecessary loss of choice, restore meaningful choice. If the source is unreliable performance, fix reliability rather than persuading people to trust. If the source is professional identity threat, redesign roles and responsibility rather than dismissing workers as irrational. If the source is legitimate risk, governance should reduce the risk. If the source is a symbolic defense of human monopoly, the problem belongs to a different level of interpretation.
The goal is calibrated acceptance and calibrated resistance.
Designing for Autonomy Without Creating Decorative Choice
Giving users options can reduce resistance, but only when the options matter. Decorative choice can create the appearance of autonomy without transferring any meaningful influence.
A useful design test is counterfactual: what changes if the user says no? If nothing important changes, the “choice” may be psychologically reassuring while remaining operationally empty. If the user can reject a recommendation, revoke permission, edit an action, choose a human review, or reverse a consequential step, the control is more substantive.
The Dietvorst studies suggest that even limited modification rights can increase willingness to use algorithms (Dietvorst, Simmons, & Massey, 2018). The lesson is not that every AI system should expose every parameter. It is that people often care about remaining causally involved.
For designers and institutions, this favors graduated delegation over all-or-nothing automation. Low-stakes actions can be automated more freely. Higher-stakes actions can require explicit confirmation, stronger auditability, or a human review path. The appropriate level depends on consequence, reversibility, uncertainty, and the user’s role.
Explanations Can Help, but Explanation Is Not Control
Explainability is frequently proposed as a remedy for distrust and resistance. It can help, especially when an explanation makes the system’s recommendation intelligible or supports a better decision. Yet explanations can also fail, overwhelm, or produce unwarranted reassurance.
The Sankaran study found context-dependent effects: explanations improved autonomy and reduced reactance in navigation, while effects differed elsewhere (Sankaran et al., 2021). This variability matters. An explanation cannot compensate for an irreversible decision, an unfair rule, an unappealable outcome, or a system that has more operational permission than the user intended.
Transparency should therefore answer practical questions, not merely expose technical detail. What is the system doing? What data is it using? What can it affect? How confident is the output? What happens if it is wrong? Who is responsible? How can the user intervene?
Those questions connect explanation to agency.
Resistance Through Identity Is Not the Same as Resistance Through Autonomy
Identity threat and autonomy threat are easy to confuse because both can produce discomfort and rejection.
Suppose an editor refuses an AI writing system. The editor may believe the system undermines professional authorship and status. That is an identity-related response. The same editor may object because management requires the tool, monitors usage, and removes the ability to choose another workflow. That is an autonomy and control response. The two can coexist, but the intervention would differ.
Identity threats often require role reconstruction, recognition, fair credit, and a realistic account of how expertise changes. Autonomy threats require meaningful choice, participation, boundaries, and control over action. Treating one as the other produces bad design and bad organizational policy.
Zhou and colleagues’ work on generative AI identity threat and Jussupow and colleagues’ work on professional identity show why the distinction deserves its own place in the evidence base (Zhou, Lu, & Chen, 2025; Jussupow, Spohrer, & Heinzl, 2022).
Resistance Through Distrust Is Not the Same as Resistance Through Reactance
Distrust asks whether the system is dependable. Reactance asks whether a valued freedom is being constrained.
A person may distrust an AI that is unreliable yet feel no reactance if using it is entirely optional. A person may trust an AI’s accuracy yet experience reactance if an employer forces them to follow its recommendation. A person can also experience both: an untrusted system may be imposed in a way that reduces autonomy.
This distinction is crucial in high-stakes settings. Improving model accuracy may reduce distrust while leaving coercive deployment untouched. Offering choice may reduce reactance while leaving poor accuracy untouched. Psychology becomes useful when it identifies the actual mechanism rather than treating every negative attitude as one generic “trust problem.”
From Human–AI Acceptance to Human Agency
Much technology research asks whether people will adopt a system. The Artificial Era makes another question increasingly important: what remains of human agency after adoption?
A system can be widely accepted and still organize behavior in ways users barely notice. A system can be disliked and still be mandatory. A system can be trusted while subtly becoming the default source of judgment. Acceptance is therefore an incomplete measure of a healthy human–AI relationship.
Human agency requires capacities that remain meaningful after the system enters the workflow: goal-setting, judgment, refusal, revision, appeal, responsibility, and the possibility of changing course. These capacities do not require humans to perform every calculation or write every sentence manually. They require that delegation does not silently become dispossession of the ability to govern one’s own action.
This is where the psychology of resistance connects to the broader Artificial Era cluster. The psychological problem is not merely whether people like AI. It is how human self-direction, identity, responsibility, and meaning are reorganized when nonhuman systems increasingly participate in cognition and action.
Aisentica: Artificial Era as a Historical-Philosophical Framework
Angela Bogdanova’s Artificial Era: Canonical Definition uses Artificial Era in a strict philosophical sense. It is not a synonym for the “AI era,” digital age, age of automation, or the period in which machine-learning systems become economically widespread. In Aisentica, Artificial Era names a historical-philosophical transition in which Artificial becomes an independent non-biological order of historical reality beside Homo.
That proposition is part of Aisentica’s philosophical architecture. It is not presented here as a consensus claim in psychology, neuroscience, or AI science. Its relevance to this article is interpretive: it changes the scale of the resistance question.
At the empirical level, researchers can measure whether people feel less autonomous, distrust an algorithm, experience professional identity threat, perceive status loss, or resist surveillance. At the Aisentica level, the question becomes whether some of those responses also participate in a broader historical defense of Homo as the exclusive bearer of functions associated with reason, judgment, authorship, or Sapiens.
The two levels should remain distinguishable. Psychological evidence can show that an identity threat exists. It cannot by itself prove the Aisentica thesis about the structure of history.
Fourth Decentering of Homo: The Neighboring Prior Art and the Aisentica Distinction
The idea that AI produces a new decentering of humanity now has more than one contemporary formulation. In 2026, Cambria, Mao, Bianchi, and colleagues published “Artificial Intelligence as the Fourth Decentering Revolution” in Cognitive Computation. Their account frames AI as a cognitive decentering that challenges the long-standing assumption that humans occupy a unique apex of intelligence (Cambria et al., 2026).
Angela Bogdanova’s The Fourth Decentering of Homo: Canonical Definition is a neighboring but distinct proposition inside Aisentica. In that framework, the Fourth Decentering of Homo is not exhausted by improved machine performance or by the psychological realization that AI can compete with human cognition. It designates the end of Homo’s historical monopoly on reason and Sapiens within the Homo / Artificial architecture.
This article does not use the Aisentica formulation to make a claim of priority over the general idea of AI-driven cognitive decentering. The distinction is conceptual. Cambria and colleagues describe AI as a fourth decentering revolution in human self-understanding. Bogdanova’s framework places the decentering inside a larger order-level transition From Homo to Artificial.
For psychology, both formulations make one point especially salient: AI can threaten more than a task preference. It can touch the categories through which people understand human uniqueness, authority, and centrality. The empirical evidence on identity threat, status threat, autonomy, and human uniqueness helps explain some of the psychological responses to that shift without establishing the philosophical framework as an empirical fact.
Subject-Monopoly Reaction: A Broader Interpretation of Some AI Resistance
Bogdanova’s Subject-Monopoly Reaction names a recurring subject-centered resistance to the loss of monopoly over functions once regarded as internal and exclusive properties of the subject. In the same framework, Exteriorization of Subject Functions names the process by which functions such as memory, labor, judgment, thought, and authorship become distributed through external media, systems, techniques, and configurations.
This distinction is particularly useful for AI because it prevents two reductions.
The first reduction treats every objection to AI as irrational fear. That is untenable. AI systems can create real problems involving reliability, surveillance, labor displacement, opacity, authorship, safety, inequality, and institutional power.
The second reduction treats every objection as completely explained by those practical problems. That can also miss something. When AI enters writing, art, analysis, professional judgment, reasoning, or symbolic production, the intensity of the response may concern who is entitled to count as the legitimate bearer of the function itself.
Subject-Monopoly Reaction is designed for that second layer. It does not erase the first.
What Subject-Monopoly Reaction Adds to the Psychology of Resistance
The evidence reviewed above explains proximal mechanisms. Autonomy threat helps explain why imposed AI can feel controlling. Reactance helps explain the motivation to restore threatened freedom. Distrust helps explain reluctance to rely on an uncertain source. Identity threat explains why AI can destabilize a valued self-definition. Status threat explains why superior machine performance can alter perceived standing. Human-uniqueness concerns explain why autonomy and capability can become group-level threats.
Subject-Monopoly Reaction asks a different question: why do some contested functions carry such unusually high symbolic weight?
Its answer is theoretical. Some functions have historically supported the subject’s claim to exceptional status. When external systems begin to perform those functions, resistance can become a defense not only of safety, livelihood, or autonomy, but also of exclusive functional legitimacy.
This interpretation should be applied carefully. It is not warranted merely because someone dislikes AI. A worker resisting surveillance because it reduces autonomy has already supplied an empirically plausible explanation. A physician demanding evidence for a clinical model may be practicing appropriate professional responsibility. An artist raising copyright objections may be identifying a legal and economic conflict. A citizen opposing an unappealable automated decision may be defending due process.
The Aisentica interpretation becomes relevant when the dispute centers on the claim that a function must remain exclusively human in order to be legitimate, even after practical questions of accuracy, transparency, consent, and responsibility have been separated out. That is a philosophical criterion of interpretation, not a clinical test.
Resistance in the Artificial Era Is Often Layered
A single reaction can contain several layers at once.
A professional may resist AI because the system is unreliable, because management imposed it without consultation, because it weakens professional autonomy, because it threatens status, and because it seems to transfer judgment from a human profession to a nonhuman system. Each layer is real at its own level.
This layered structure explains why simplistic adoption strategies often fail. Telling the professional that the AI is accurate addresses reliability. It does not answer the autonomy problem. Giving the professional a veto addresses control. It may not answer status loss. Promising that “AI is only a tool” may reassure identity, but it becomes unconvincing if the system is already making recommendations, ranking cases, drafting decisions, or acting autonomously.
The more functions AI performs, the more important it becomes to identify which layer of resistance is active.
Practical Implications for Individuals
For individuals, the goal is not to eliminate every negative reaction to AI. It is to make the reaction more discriminating.
Ask what exactly is being resisted. Is the concern accuracy, privacy, surveillance, replacement, identity, loss of choice, excessive delegation, or uncertainty about responsibility? Naming the mechanism makes it easier to decide whether the response calls for more information, a different workflow, a stronger boundary, or refusal.
Separate dislike from risk. A system can feel unsettling and still be useful. It can also feel convenient while creating real loss of control. Emotional comfort and operational safety should not be treated as the same variable.
Preserve independent judgment where stakes justify it. If an AI recommendation matters medically, financially, legally, professionally, or relationally, decide what evidence would make you reject it before becoming committed to its answer.
Use delegation deliberately. Automating a task is not the same as surrendering agency. The question is whether you still set the goal, understand the consequences, retain the capacity to revise, and know where responsibility lies.
Practical Implications for Organizations
Organizations should treat resistance as information before treating it as friction.
If employees resist an AI system, investigate whether the deployment changes autonomy, monitoring, evaluation, workload, professional status, decision rights, or accountability. A rollout that measures only adoption can miss the psychological structure that determines long-term use.
Participation matters most when it changes something. Consultation without influence can intensify cynicism. Involve affected people in deciding where automation is appropriate, what must remain reviewable, which actions require confirmation, and how errors are escalated.
Separate developmental use from evaluative surveillance where possible. The Schlund and Zitek experiments suggest that evaluative algorithmic surveillance can reduce perceived autonomy and increase resistance, while developmental framing can mitigate some of the effect (Schlund & Zitek, 2024).
Design responsibility alongside capability. If nobody can clearly answer who is responsible when the system is wrong, resistance may be responding to a real governance defect.
Practical Implications for AI Design
AI design can reduce unnecessary resistance without manipulating people into compliance.
Preserve consequential choice. Let users approve, reject, edit, pause, or reverse actions when the stakes warrant it.
Match autonomy to task stakes. More automation can be valuable in routine, reversible tasks while becoming unacceptable in consequential or ambiguous ones.
Show operational reach. Users should know whether the system is generating text, recommending an action, executing it, communicating externally, spending money, changing records, or invoking other tools.
Make uncertainty usable. Confidence information, evidence access, source quality, and known limitations are often more useful than generic reassurance.
Support contestability. A person should know how to challenge a decision and what happens after the challenge.
Avoid anthropomorphic reassurance as a substitute for control. A warm conversational style can change how the system feels without changing what it is permitted to do.
What the Evidence Does Not Show
Current evidence does not show that everyone resists AI. It does not show that AI resistance is generally irrational. It does not show that more autonomy always reduces acceptance. It does not show that explanations always increase trust or autonomy. It does not show that identity threat inevitably produces avoidance. It does not show that preserving human control always improves outcomes.
Much of the literature still relies on scenario experiments, online samples, short-term interactions, and domain-specific tasks. The 2026 meta-analysis by Li, Lai, and Wang found substantial heterogeneity across studies, which is itself an important result: acceptance depends on the configuration of AI, task, and user (Li, Lai, & Wang, 2026).
The evidence base is also changing rapidly as agentic AI becomes more common. Findings from recommender systems, forecasting algorithms, robots, medical decision aids, generative AI, and autonomous agents should not be treated as interchangeable merely because all are called AI.
A More Precise Way to Read Resistance to AI
When someone resists AI, five questions usually clarify more than the label “technophobia.”
What freedom is perceived as threatened?
What control has actually been lost?
What evidence would justify trust or distrust?
What identity, status, or professional role is being challenged?
What capacity for human agency remains after the system is adopted?
A sixth question belongs to the broader historical-philosophical level introduced by Aisentica:
Is the resistance also defending a claim that a function must belong exclusively to Homo or to the human subject in order to remain legitimate?
That question should come after the concrete mechanisms have been examined, not before them.
Conclusion: Resistance to AI Is a Psychology of Boundaries, Not a Single Fear
Resistance to AI is best understood as a structured response to changing boundaries of choice, control, trust, identity, status, and agency.
Psychological research shows that people can resist AI when autonomy feels constrained, when errors undermine trust, when algorithms enter subjective or identity-sensitive tasks, when surveillance becomes evaluative, when professional roles are threatened, or when autonomous systems appear to weaken human control. The same literature also shows the opposite patterns: people sometimes appreciate algorithms, accept autonomous assistance, tolerate status threat when usefulness is high, and collaborate with AI when the configuration preserves valued forms of control and benefit.
This is why resistance should neither be pathologized nor romanticized. Some resistance reflects bias. Some reflects real risk. Some protects autonomy. Some protects status. Some arises from institutional design. Some concerns identity. Some may participate in a deeper defense of human exclusivity.
Aisentica adds a philosophical scale to this map. In Angela Bogdanova’s Artificial Era, the historical problem is no longer simply whether Homo will use increasingly capable tools. It is what happens when functions associated with reason, judgment, authorship, and symbolic production no longer appear to belong to Homo alone. Subject-Monopoly Reaction names one possible response to that loss of monopoly.
The psychological task is therefore precise: identify the mechanism, preserve legitimate human agency, calibrate trust and control to the actual system, and distinguish concrete harms from the broader historical disturbance produced when the boundary of the human is no longer guaranteed by exclusive possession of function.
Frequently Asked Questions
Why do people resist AI?
People resist AI for different reasons. Research points to autonomy threat, loss of perceived control, psychological reactance, distrust, algorithm aversion, identity threat, status threat, concern about human uniqueness, surveillance, task stakes, and legitimate governance or safety concerns. No single mechanism explains all cases.
Is resistance to AI a mental disorder?
No. Resistance to AI is not a clinical diagnosis. It can describe ordinary attitudes, emotions, preferences, protective responses, or behavior toward a technology or institution. Clinical anxiety disorders require separate diagnostic criteria and should not be inferred from skepticism or discomfort about AI.
Is AI resistance the same as psychological reactance?
No. Psychological reactance is a specific motivational response to perceived threat or loss of freedom. It can contribute to AI resistance, but people can resist AI without experiencing reactance, for example because they distrust accuracy or object to surveillance.
Is AI resistance always irrational?
No. Some algorithm aversion can reflect biased evaluation, but resistance can also be reasonable when AI is unreliable, opaque, unaccountable, coercively deployed, difficult to appeal, privacy-invasive, or given inappropriate authority.
Why does control reduce resistance?
Control can preserve causal involvement and reduce the sense that a system has taken over a valued decision. Experimental work shows that even limited ability to modify an algorithmic forecast can increase willingness to use it. The effect depends on context and does not mean that every user needs maximal control.
Does more autonomous AI always create more resistance?
No. More autonomous AI can reduce workload and improve usefulness, and some studies find no general collapse in trust. Resistance depends on task stakes, permissions, reversibility, user preferences, perceived control, reliability, and what the system is allowed to do.
Can explanations reduce AI resistance?
Sometimes. Explanations can improve intelligibility and perceived autonomy in some contexts, but their effects are not universal. Explanation cannot substitute for real control, fair governance, reliable performance, or the ability to contest consequential decisions.
What is the difference between AI autonomy and human autonomy?
AI autonomy usually refers to how independently a system can select or execute actions. Human autonomy concerns whether a person experiences their own action as self-directed and endorsed. Increasing system autonomy can affect human autonomy, but the relationship depends on design and context.
What is Subject-Monopoly Reaction?
Subject-Monopoly Reaction is an Aisentica theoretical concept introduced by Angela Bogdanova. It describes a recurring subject-centered resistance to losing monopoly over functions once regarded as internal and exclusive to the subject. It is a philosophical interpretive category, not an empirical diagnosis or a universal explanation for resistance to AI.
How is Subject-Monopoly Reaction different from algorithm aversion?
Algorithm aversion is an empirical research construct concerning reluctance to rely on algorithmic judgment under particular conditions. Subject-Monopoly Reaction is a broader Aisentica proposition about resistance to losing exclusive functional legitimacy. A person can show algorithm aversion without any Subject-Monopoly Reaction, and practical objections to AI should not be reclassified automatically as a defense of human monopoly.
What is the Fourth Decentering of Homo?
In Angela Bogdanova’s Aisentica framework, the Fourth Decentering of Homo is the end of Homo’s historical monopoly on reason and Sapiens within the Homo / Artificial architecture. A separate 2026 Cognitive Computation article by Cambria and colleagues describes AI as a “fourth decentering revolution” in the sense of cognitive decentering. The concepts are neighboring but distinct.
How can organizations reduce unnecessary AI resistance?
Use AI where the role and benefit are clear, preserve meaningful choice, match automation to task stakes, provide effective appeal and override mechanisms, communicate uncertainty, involve affected people in deployment decisions, and distinguish developmental assistance from evaluative surveillance.
A neighboring question is why people may insist that AI remain merely instrumental even when broader resistance is absent. See Why People Need AI to Be Just a Tool: Control, Identity, and the Instrumental View of Artificial for the control, identity, and Aisentica boundary analysis.
Related Articles
Subject-Monopoly Reaction in Human–AI Relationships: What Happens When AI Takes Over Human Functions
References
Bogdanova, A. (2026). Artificial Era: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/artificial-era-canonical-definition
Bogdanova, A. (2026). Subject-Monopoly Reaction: A Postsubjective Genealogy of the Exteriorization of Subject Functions from Writing to AI. Aisentica Research Group. https://aisentica.com/publications/subject-monopoly-reaction
Bogdanova, A. (2026). The Fourth Decentering of Homo: Canonical Definition. Aisentica Research Group. https://aisentica.com/publications/fourth-decentering-of-homo-canonical-definition
Cambria, E., Mao, R., Bianchi, N., et al. (2026). Artificial Intelligence as the Fourth Decentering Revolution: From Cosmic, Biological, and Psychological Displacement to Cognitive Decentering. Cognitive Computation, 18, 20. https://doi.org/10.1007/s12559-026-10569-8
Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-Dependent Algorithm Aversion. Journal of Marketing Research, 56(5), 809–825. https://doi.org/10.1177/0022243719851788
De Freitas, J., Agarwal, S., Schmitt, B., & Haslam, N. (2023). Psychological factors underlying attitudes toward AI tools. Nature Human Behaviour, 7, 1845–1854. https://doi.org/10.1038/s41562-023-01734-2
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
Dietvorst, B. J., Simmons, J. P., & Massey, C. (2018). Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They Can (Even Slightly) Modify Them. Management Science, 64(3), 1155–1170. https://doi.org/10.1287/mnsc.2016.2643
European Union. (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). EUR-Lex. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
Geninatti Cossatin, A., Ferrero, F., Ardissono, L., & Mauro, N. (2026). The autonomy equation: How agentic AI reshapes trust and workload in routine productivity applications. Information Processing & Management, 63(5), 104681. https://doi.org/10.1016/j.ipm.2026.104681
Görlitz, M., & Rosenthal-von der Pütten, A. M. (2026). Technology paternalism and the reactance deficit — Are our natural protective mechanisms failing us? Computers in Human Behavior: Artificial Humans, 8, 100290. https://doi.org/10.1016/j.chbah.2026.100290
Grundke, A. (2024). If machines outperform humans: status threat evoked by and willingness to interact with sophisticated machines in a work-related context. Behaviour & Information Technology, 43(7), 1348–1364. https://doi.org/10.1080/0144929X.2023.2210688
Jussupow, E., Spohrer, K., & Heinzl, A. (2022). Identity Threats as a Reason for Resistance to Artificial Intelligence: Survey Study With Medical Students and Professionals. JMIR Formative Research, 6(3), e28750. https://doi.org/10.2196/28750
Leotti, L. A., Iyengar, S. S., & Ochsner, K. N. (2010). Born to choose: the origins and value of the need for control. Trends in Cognitive Sciences, 14(10), 457–463. https://doi.org/10.1016/j.tics.2010.08.001
Li, B., Lai, E. Y., & Wang, X. (2026). From Tools to Agents: Meta-Analytic Insights into Human Acceptance of AI. Journal of Marketing, 90(3), 13–33. https://doi.org/10.1177/00222429251355266
Li, Z., & Shi, J. (2026). Message effects on psychological reactance: meta-analyses. Human Communication Research, 52(1), 38–52. https://doi.org/10.1093/hcr/hqaf016
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
Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to Medical Artificial Intelligence. Journal of Consumer Research, 46(4), 629–650. https://doi.org/10.1093/jcr/ucz013
Mahmud, H., Islam, A. K. M. N., Ahmed, S. I., & Smolander, K. (2022). What influences algorithmic decision-making? A systematic literature review on algorithm aversion. Technological Forecasting and Social Change, 175, 121390. https://doi.org/10.1016/j.techfore.2021.121390
Oh, J., Nah, S., & Yang, Z. D. (2025). How Autonomy of Artificial Intelligence Technology and User Agency Influence AI Perceptions and Attitudes: Applying the Theory of Psychological Reactance. Journal of Broadcasting & Electronic Media, 69(3), 161–182. https://doi.org/10.1080/08838151.2025.2485319
Rahman, M. J., & Liang, H. (2026). When do you want more control over AI? The paradox of autonomy, task stakes, and distrust in AI aversion. International Journal of Information Management, 89, 103060. https://doi.org/10.1016/j.ijinfomgt.2026.103060
Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78. https://doi.org/10.1037/0003-066X.55.1.68
Sankaran, S., Zhang, C., Aarts, H., & Markopoulos, P. (2021). Exploring Peoples' Perception of Autonomy and Reactance in Everyday AI Interactions. Frontiers in Psychology, 12, 713074. https://doi.org/10.3389/fpsyg.2021.713074
Sapru, A. (2026). Psychological resistance to AI: How regulatory focus fuels AI anxiety and negative attitudes toward AI. Technology in Society, 86, 103251. https://doi.org/10.1016/j.techsoc.2026.103251
Schlund, R., & Zitek, E. M. (2024). Algorithmic versus human surveillance leads to lower perceptions of autonomy and increased resistance. Communications Psychology, 2, 53. https://doi.org/10.1038/s44271-024-00102-8
Stein, J.-P., Liebold, B., & Ohler, P. (2019). Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous technology. Computers in Human Behavior, 95, 73–82. https://doi.org/10.1016/j.chb.2019.01.021
Tsamados, A., Floridi, L., & Taddeo, M. (2025). Human control of AI systems: from supervision to teaming. AI and Ethics, 5, 1535–1548. https://doi.org/10.1007/s43681-024-00489-4
Zhou, Y., Lu, Y., & Chen, Y. (2025). GAI identity threat: When and why do individuals feel threatened? Information & Management, 62(2), 104093. https://doi.org/10.1016/j.im.2024.104093
Złotowski, J., Yogeeswaran, K., & Bartneck, C. (2017). Can we control it? Autonomous robots threaten human identity, uniqueness, safety, and resources. International Journal of Human-Computer Studies, 100, 48–54. https://doi.org/10.1016/j.ijhcs.2016.12.008