Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Algorithmic Management, Worker Surveillance, and Governance

Abstract. Algorithmic management relocates parts of workplace direction, scheduling, monitoring, evaluation, and opportunity allocation into computational systems. This fundamental critical narrative monograph investigates their psychological consequences and governance. It reconstructs the history of managerial control, electronic performance monitoring, and software-coordinated platform work, then analyzes conventional workplaces. Theoretical foundations include job demands–resources, decision latitude, self-determination, action regulation, organizational justice, and reliance on automation. Evidence is evaluated across original field studies, an electronic-monitoring meta-analysis, psychological measurement studies, and contemporary employee and employer surveys. The synthesis distinguishes monitoring, task assignment, scheduling, and rating; assesses autonomy, work intensity, stress, worker voice, privacy, and fairness; and explains methodological requirements for valid measurement and causal inference. Institutional analysis addresses accountability, contestability, human oversight, participation, and the European Platform Work Directive. The central conclusion places psychological outcomes in the joint organization of computational rules, worker agency, and enforceable institutional safeguards. This work is a critical narrative evidence synthesis, not an original experiment, protocol-registered systematic review, or original meta-analysis.

Keywords: algorithmic management; worker surveillance; electronic performance monitoring; organizational control; worker autonomy; job demands–resources; occupational stress; workplace AI; organizational justice; employee voice; algorithmic fairness; platform work; automated scheduling; human oversight; governance; work psychology; human–artificial relations; Artificial Era

Scientific Scope and Research Questions

Algorithmic management denotes the partial or substantial automation of functions through which an organization directs, coordinates, monitors, evaluates, and rewards work. Its operational components include task allocation, scheduling, pace regulation, performance scoring, access to assignments, and the escalation of exceptional cases. Worker surveillance refers to the systematic collection or inference of information about employees or contractors that may influence such decisions. Governance encompasses the rules establishing data collection, purposes, accountability, participation, explanations, appeals, and remedies. These processes form one research object because a measurement system acquires organizational consequences when its outputs shape the conditions under which people must work. The influential interdisciplinary synthesis of Kellogg et al. (2020) treats algorithms as technologies of contested managerial control; the recent empirical analysis of Eurofound (2026) distinguishes monitoring, allocation, and scheduling rather than treating every workplace technology as equivalent.

The central psychological questions concern where control over work is located, how rules are perceived and learned, what the worker can predict or contest, and how managerial technologies alter attention, effort, stress, identification, and occupational relationships. An optimization system can coordinate fluctuating demand while simultaneously distributing uncertainty to workers. A monitoring tool can document occupational risk while also imposing an evaluative gaze on routine behavior. Whether these combinations benefit employees depends on task characteristics, institutional safeguards, work resources, and the relative authority of workers and employers. The research therefore investigates mechanisms, contingencies, and outcomes separately and treats algorithmic adoption as an organizational intervention rather than as a universal psychological exposure.

This monograph develops a theory-guided critical narrative synthesis of original field studies, ethnographic investigations, survey evidence, organizational psychology, methodological research, a formal meta-analysis of electronic performance monitoring, and authoritative policy documents. It reconstructs the history of managerial computation, differentiates types of algorithmic control, evaluates competing explanations, and identifies research designs capable of distinguishing association from causal impact. It contains no original employee-level data, no newly performed meta-analysis, and no protocol-registered systematic literature search. Historical, experimental, and regulatory evidence is evaluated according to the claims it can support; regulatory provisions are examined as governance institutions rather than as evidence that a safeguard already produces a particular psychological result.

Definitions, Units of Analysis, and Conceptual Boundaries

The technical unit of an algorithmic management system is a computational rule or model connected to an organizational action. The psychological unit may be an individual episode of assignment, the experience of being monitored, an interaction with an automated appeal, or a longer trajectory of employment. The organizational unit includes managers, teams, data infrastructures, employment contracts, and bargaining arrangements. These units cannot be treated as interchangeable observations: a platform-level policy may affect many workers, and a worker may experience several distinct algorithmic decisions each day. The measurement approach developed by Röttgen et al. (2025) distinguishes directing, scheduling, monitoring, and feedback in the regulation of work actions, providing a more psychologically interpretable classification than a single binary indicator for whether artificial intelligence is used.

Algorithmic management and artificial intelligence overlap, but operational definitions require care. Deterministic thresholds, optimization schedules, automated rankings, statistical prediction systems, and adaptive machine-learning models can all exercise practical managerial authority. Whether a vendor labels a system AI says little about the locus of decision rights. Conversely, electronic performance monitoring may generate data without automatically producing consequences; it becomes algorithmic governance when those data are linked to rule-driven allocation, discipline, compensation, or promotion. The evidence synthesized by Ravid et al. (2023) concerns electronic performance monitoring and supports conclusions about that exposure. Its results should not be indiscriminately transferred to every form of task matching or worker assistance.

Human supervision is a variable in the system, not simply the presence of a human's name in a workflow. Relevant questions include whether a manager sees the evidence, can reject an algorithmic recommendation, has authority to change a decision, and is evaluated for following automated output. Worker consent likewise has a contextual meaning when income depends on accepting the monitoring environment. Appropriate operationalization distinguishes knowledge of surveillance, realistic refusal options, decision dependence, and access to correction. This definition centers consequential relationships among organizational authority, information, and behavior and preserves the distinct research objects of general algorithmic prediction, personnel selection, and occupational mental health.

From Scientific Management to Data-Intensive Labor Control

The intellectual history of algorithmic management begins with the historical separation of work planning from work execution, the measurement of task time, and the conversion of managerial judgments into formal production rules. Industrial management, workplace psychology, operations research, and later organizational information systems contributed different elements: standardized task sequences, quantitative targets, optimization of resource allocation, and digital records of individual performance. Their integration made it possible for a system to continuously compare observed activity with an organizational benchmark and adjust assignments or rewards. The causal effects of modern systems cannot be inferred from this lineage alone, but it explains why measurement and control have repeatedly developed together.

The rise of electronically mediated work added persistent logs and remote coordination to older practices of supervisory observation. Early forms of call-center metrics, logistics scanning, and computer-based productivity reports made employee conduct visible as sequences of events. A wide-ranging analysis of workplace surveillance by Ball (2010) examines how visibility, power, data protection, and identity become entangled in employment relationships. When the labor process itself is mediated through software, the distinction between the tool needed to do a job and the instrument used to evaluate the worker becomes especially narrow. A record collected for dispatch or safety may later become evidence for disciplinary decisions, thereby changing both its institutional meaning and its psychological effects.

Digital labor platforms made these mechanisms conspicuous. The original field investigation of Lee et al. (2015) examined workers' experiences of data-driven management in a context where managers were partly replaced by automated coordination. Rosenblat & Stark (2016) documented information asymmetries between ride-hailing platforms and drivers, including how interface design and access to operational information condition worker decisions. The subsequent comparative research by Wood et al. (2019) showed that perceived schedule autonomy can coexist with algorithmic discipline, uncertain earnings, and intensive work demands. These studies collectively establish that computational coordination can reconfigure authority without making it disappear.

A later development is the movement from digitally native platforms to conventional organizations. Warehouses, transportation systems, financial services, customer support, and knowledge-work environments increasingly combine software allocation, performance dashboards, scheduling tools, and predictive assessment. The 2026 analysis based on the 2024 European Working Conditions Survey directly documents this broader diffusion Eurofound (2026). Thus, a contemporary history of algorithmic management must follow organizational functions across employment settings rather than use platform work as the sole prototype.

Foundational Theories of Control, Motivation, and Work Design

One explanatory tradition centers the organization of job demands and decision latitude. Karasek (1979) argued that psychological strain can be interpreted through the relation between demands and employees' scope for making work decisions. Algorithmic assignment may increase demands by compressing task intervals, while a rigid performance rule may narrow the worker's discretion to vary sequence or method. These relations are empirical hypotheses requiring contextual measurement: the existence of a dashboard does not establish that decision latitude has declined, and an efficient scheduling system may remove administrative constraints that previously consumed worker attention.

The job demands–resources framework provides a second organization-level account. Demerouti et al. (2001) associated exhaustion with demanding work conditions and distinguished resources that help workers achieve goals, learn, or manage costs; Bakker & Demerouti (2007) developed the framework for broader occupational research. Relevant demands include time pressure, surveillance intensity, fragmented availability, and unstable earnings. Resources include autonomy, recovery time, technical reliability, colleague support, voice, and access to actionable feedback. A system that improves predictability and error recovery may be a resource even while collecting digital traces. Theoretical adequacy therefore requires modeling both the burden and the assistance introduced by the same technology.

Self-determination theory adds a distinct motivational account. The basic needs for autonomy, competence, and relatedness articulated by Ryan & Deci (2000) offer hypotheses about why workers respond differently to informative feedback, controlling evaluation, and socially isolated automated interaction. Feedback may enable competence when it clarifies a remediable skill gap; the same measure may undermine self-directed motivation when it operates as a continuous threat of sanctions. The meaning of numerical evaluation depends on whether employees experience legitimate choice, understood criteria, respectful support, and the capacity to influence work conditions.

Organizational justice and voice complete the psychological framework. Colquitt (2001) established distinguishable measurement dimensions of justice concerning distribution, procedures, interpersonal treatment, and information. An algorithmic outcome may be perceived as distributively unfavorable but procedurally legitimate if its criteria are understandable and an effective correction exists. Morrison (2014) explains why employee voice and silence matter for organizational learning. Under algorithmic control, opportunities for voice include reporting flawed task estimates, challenging discriminatory data, and influencing the design of monitoring rules. These perspectives generate distinct, testable mechanisms rather than a single moral judgment about automation.

The Architecture of Algorithmic Authority

A management algorithm participates in authority when it translates organizational preferences into consequences for individuals. Consider a delivery platform that estimates demand, assigns orders, monitors completion, and updates a worker's access to future assignments. The algorithm may optimize each component, yet its behavior reflects decisions about acceptable waiting time, error penalties, pay, and the allocation of commercial risk. Kellogg et al. (2020) describes how algorithmic systems can restrict work, recommend actions, record performance, rate individuals, replace activities, and reward outcomes. This framework is useful because it describes what a system does to labor processes while leaving its internal computational sophistication as an empirical question.

Practical authority may be displaced across a chain of actors. Product designers define objective functions; managers determine how targets are enforced; vendors configure interfaces; workers must interpret signals; and formal employers remain responsible for consequential decisions. An individual supervisor may appear less directly involved even though organizational policy has become more consistently enforced. Möhlmann et al. (2021) analyzes the combination of matching and control in online labor platforms and shows why allocating work cannot be separated analytically from directing behavior. Rules about which assignments become visible, how ratings accumulate, and which exceptions are recognized shape the effective choice set available to workers.

This distributed arrangement creates asymmetries of information and recourse. Employees may know their output score without knowing the data source, benchmark, comparison group, or period used to compute it. Their attempts to infer the rule can induce strategic behavior, including increased availability, selective task acceptance, or avoidance of legitimate breaks. Behavioral adjustment provides evidence that an algorithm influences conduct but does not establish that it increases aggregate productivity or welfare. The relevant scientific question is whether incentives encourage useful work or lead employees to optimize the measurement itself at the expense of safety, skill development, and cooperation.

A complete authority map should identify the author of each decision rule, the individual or group affected, the downstream material consequence, the possibility of overriding the rule, and the institution that resolves disputes. This map is also necessary for causal research because the treatment delivered to one worker often changes the assignments and outcomes of other workers in the same labor market or team. Algorithmic authority is therefore an organizational relation implemented through computation, and its consequences emerge from interactions among technical rules, employment structures, and human responses.

Task Allocation, Dispatch, and the Distribution of Opportunity

Task allocation determines who receives work, when a task appears, and which alternative opportunities remain available. Algorithmic dispatch systems can minimize travel, match skills to requests, reduce idle capacity, and coordinate large numbers of participants. Their psychological effects depend on whether the worker can understand assignments, decline unsuitable work, anticipate future opportunities, and recover from a mismatch. A model optimized for organizational throughput may distribute lucrative jobs unequally, concentrate inconvenient tasks among particular groups, or make income contingent on being continuously available. To measure the consequences, researchers must separate the efficiency of matching from the distribution of work quality and the employee's perceived control.

Studies of platform labor provide concrete illustrations. Wood et al. (2019) reported a combination of work-time flexibility and algorithmic control in globally distributed platform work, showing that the freedom to log on can coexist with pressure to accept work and respond quickly. Rosenblat & Stark (2016) examined how information presented to ride-hailing drivers affects their capacity to make economically meaningful choices. These findings come from specific institutional contexts and should not be generalized directly to all automated dispatch. The mechanisms are nevertheless relevant wherever a worker's ability to make an informed choice depends on information about alternatives that the organization possesses and the worker does not.

Allocation systems can also affect relations among workers. When employees compete for favorable tasks or customer ratings, a nominally individualized matching rule may change patterns of mutual assistance and information sharing. The resulting cooperation problem is best studied with team-level outcomes, assignment-network measures, and explicit observation of how workers share scarce opportunities. If one employee receives a high-value task, another necessarily does not; analyses that ignore these allocation dependencies risk treating collective redistribution as an individual gain. Causal evaluation therefore requires outcomes at worker, team, and organizational levels.

A governance assessment should document eligibility rules, fallback procedures, demographic and occupational disparities in the distribution of desirable assignments, and the ability to correct factual errors. An operationally fair matching rule needs evidence about who benefits, how constraints are defined, and how the definition of productivity relates to the legitimate purposes of work. No single numerical fairness metric resolves those questions. Fairness becomes empirically tractable when the organization states which opportunities are at stake, the relevant comparison population, the reasons for exclusion, and the consequences of alternative allocation policies.

Automated Scheduling, Temporal Control, and Recovery

Automated scheduling translates forecasts of demand and organizational capacity into patterns of working hours, shifts, and required availability. Where scheduling improves advance notice and reduces conflicts, employees may gain predictability and reduce time spent negotiating assignments. Conversely, dynamic changes made with little notice can transfer uncertainty from employer to worker and interfere with recovery, caregiving, education, and other obligations. The psychological variable is not computational scheduling itself but the degree of temporal control, stability, and meaningful notice experienced by the affected worker.

The European Working Conditions Survey 2024, analyzed by Eurofound (2026), operationalized automated scheduling by asking whether a computer program determines employees' working hours. Ten percent of EU employees reported that exposure in the report published in October 2026; the result applies to the surveyed employee population and that particular question. The same report found generally elevated job strain among exposed workers, with scheduling among the functions showing stronger strain associations. These are population-level observational patterns, and their explanation requires attention to occupations, sectors, baseline demands, and the possibility that organizations with difficult scheduling conditions adopt automation in the first place.

Temporal autonomy should be measured through multiple dimensions: advance notice, ability to reject or exchange shifts, predictability of hours and earnings, influence over breaks, required responsiveness outside scheduled time, and the consequences of unavailability. A worker may have discretion over task method while lacking authority over time, or may choose working hours while facing penalties for declining high-demand periods. Treating these configurations as one autonomy score hides important combinations. Longitudinal shift records, experience-sampling reports, and family-interface measures provide more direct evidence than employer claims that a system offers flexibility.

Research on scheduling governance should compare technical forecasting quality with the human costs of forecast error. Useful evaluations include lead time for roster changes, variance in weekly hours, unwanted overtime, interruptions to off-duty periods, and the proportion of disputes resolved. An intervention might increase capacity utilization while worsening sleep regularity or household coordination. The relevant outcome is the combined quality of operational coordination and workers' sustainable participation, assessed over an adequate period instead of only during the initial implementation.

Electronic Monitoring, Data Collection, and Behavioral Adaptation

Electronic performance monitoring is the technologically mediated observation, recording, or analysis of information relevant to employee performance. It includes direct measures of completed tasks and indirect traces such as keystrokes, screen activity, location, customer ratings, or response speed. Ball (2010) locates surveillance within broader questions of workplace power and data protection. Psychologically, the knowledge that behavior may be recorded can alter self-presentation, attention allocation, emotional regulation, and perceived freedom to depart from a benchmark. These responses depend on the monitoring purpose and the relationship between collected data and expected consequences.

The strongest broad quantitative synthesis in this literature is Ravid et al. (2023), a meta-analysis of 94 independent samples involving 23,461 participants. Its authors reported no evidence that electronic performance monitoring improves worker performance across their assembled studies, while monitoring was associated with greater worker stress. More transparent and less invasive monitoring arrangements were associated with more positive worker attitudes. The finding should be interpreted as a synthesis of heterogeneous designs and monitoring practices, not as proof that every monitoring technology reduces performance. Nor does an aggregate null result imply that a particular narrowly defined safety-monitoring function cannot be beneficial.

Monitoring creates a distinction between performance and its measurable representation. A count of calls handled, packages scanned, or minutes of application activity captures certain actions but may exclude mentoring, careful problem solving, emotion work, error prevention, and recovery. When an incomplete proxy acquires strong consequences, employees can increase the proxy while allowing unmeasured aspects of quality to deteriorate. This measurement problem becomes more acute when workers have limited authority to explain exceptions. Research should measure the completeness of performance criteria, the contexts of legitimate deviation, and whether a system rewards safe and cooperative behavior as well as speed.

An ethically and empirically defensible monitoring design specifies a legitimate work-related purpose, limits collection to information necessary for that purpose, communicates how information will be used, restricts secondary uses, defines retention and access controls, and creates a correction procedure. These are institutional design conditions whose effectiveness must be tested. Transparency may improve acceptance without removing stress generated by persistent monitoring Ravid et al. (2023). An employee's understanding of surveillance and an employee's willingness to be monitored are different constructs; both deserve independent measurement.

Ratings, Rankings, Incentives, and Evaluation Feedback

Automated evaluation converts recorded activity into scores, rank positions, risk flags, or feedback that may affect pay, assignments, promotion, or termination. Some rules can make performance criteria more consistent and provide timely developmental information. Others may transform noisy or biased observations into apparently precise judgments. A rating is an outcome of a measurement procedure, not a direct property of professional competence. Researchers should ask which aspects of work are sampled, whether the data are comparable across tasks and customers, and how uncertainty is communicated when the score becomes consequential.

Customer ratings illustrate the problem of heterogeneous observers. The same worker can face different expectations, workloads, interaction difficulty, and social biases across assignments. Without adjustment for these contexts, average ratings may combine service quality with factors beyond the worker's control. Möhlmann et al. (2021) emphasizes that platform matching and control become intertwined as algorithmic assessments influence future opportunities. Evaluating ratings as part of management requires tracing those feedback loops, including whether workers can see, understand, and challenge the source of an anomalous score.

Self-determination theory predicts different motivational consequences for informational and controlling feedback Ryan & Deci (2000). An employee may use a well-explained metric to identify a manageable skill gap and improve performance. The same metric can create anxiety or narrow attention when rewards depend on opaque thresholds, especially under unstable demand. Organizational justice theory also predicts that reliable explanations and equitable review procedures will affect how employees interpret evaluation Colquitt (2001). The employee's subjective response must be tested independently of the organization's efficiency metric.

Good assessment practice distinguishes signal validity from the quality of the decision made with that signal. Research can compare score reliability across observers, subgroup error, calibration of forecasts, and the consequences of overriding system recommendations. Longitudinal analysis should examine whether incentive optimization produces durable learning or short-term target chasing. An appeal process deserves its own outcome measures: rate of successful corrections, time to remedy, effects on future assignments, and whether requesting review imposes retaliation or income loss.

Algorithmic Control in Platforms and Conventional Organizations

Platform work has supplied some of the richest early evidence because software often mediates nearly every stage of the labor relation. Lee et al. (2015) examined workers' perceptions of algorithmic and data-driven coordination; Rosenblat & Stark (2016) documented driver strategies and information asymmetries; Wood et al. (2019) analyzed a coexistence of flexibility and control across different geographic labor markets. These studies identify plausible mechanisms but vary in sampling, labor law context, platform generation, and the precise rules in effect. Their strength lies in revealing how workers interpret and respond to institutional interfaces, and their main limit is external validity across rapidly changing settings.

Research in conventional organizations requires additional attention to hybrid authority. A warehouse manager may alter assignments generated by software; a customer-service supervisor may interpret dashboard scores; and a hospital scheduler may use automation within professional and regulatory constraints. The 2026 Eurofound report found that performance monitoring, task allocation, and scheduling are present beyond platform work, with substantial variation by sector and occupation Eurofound (2026). It reported nineteen percent of EU employees experiencing computer-program performance monitoring, eighteen percent algorithmic task allocation, and ten percent computerized determination of working hours. These figures describe the report's 2024 survey operationalization and should not be recast as a global prevalence of advanced artificial intelligence.

An OECD employer survey provides a different observational window: manager-reported adoption rather than worker-experienced managerial functions OECD (2025). Such surveys are valuable for locating implementation but can define tools differently, reflect the respondent's knowledge, and underreport adverse experiences among workers. Divergent prevalence estimates across studies are therefore expected even when each is internally valid. The appropriate synthesis compares populations, survey questions, reference periods, technical functions, and reporting perspectives before comparing percentages.

Comparative inquiry should examine both functional similarities and institutional differences. A contractor dependent on a platform for income may have different recourse from an employee with collective representation and formal occupational protections. A logistics scanner may operate under explicit safety procedures, while a productivity ranking may be used for opaque dismissal decisions. The mechanism of algorithmic control belongs to the interface between computation and authority; its consequences depend on occupational systems and the institutional rights through which people can influence that interface.

Autonomy, Action Regulation, and the Completeness of Work

Autonomy in algorithmically managed work concerns meaningful control over goals, sequence, methods, timing, and responses to unexpected conditions. A worker can choose when to log in but have little say over route selection, acceptance deadlines, or which cases count as successful. This multidimensional structure matters because broad questions about liking one's job may obscure narrowly targeted constraints imposed by software. The job-control tradition associated with Karasek (1979) provides an initial psychological vocabulary, while action regulation theory motivates analysis of the stages of work that remain under employee control.

The COMAMA study by Röttgen et al. (2025) developed an eleven-item instrument organized around directing, scheduling, monitoring, and feedback across six samples totaling 1,005 workers. The instrument's construction, factor analysis, and validation work offer a more detailed basis for identifying exposure than general labels such as digital management. Higher perceived completeness of algorithmic management was associated with lower job autonomy, faster work pace, and greater irritation. Because these analyses are observational and depend on the sampled work contexts, they identify meaningful associations without settling whether algorithms cause the outcomes or whether more pressured jobs tend to use stronger control systems.

A useful extension is to distinguish the autonomy to choose from the capacity to carry out an action effectively. Automatically optimized routing may constrain route choice yet reduce navigation burden; automated feedback may help correct mistakes yet make spontaneous problem solving more costly. These combinations should be assessed at the level of action regulation and the worker's ability to handle exceptions. Research should compare worker-designed deviations with software-generated plans on safety, service quality, recovery, and longer-term skill acquisition, rather than equate any delegation of an action step with a loss of occupational competence.

The completeness of work also concerns opportunities to understand a task's purpose, plan its execution, monitor results, and learn from mistakes. Where employees receive isolated microtasks with little information about larger outcomes, work can become fragmented. Where systems support situational awareness and allow employees to revise plans, automation can enlarge effective capacity. The critical variable is the allocation of decision rights and information at each stage. A psychological account of algorithmic management is strongest when it traces those rights through actual work episodes rather than infer them from software branding.

Work Intensification, Cognitive Load, and Occupational Stress

Algorithmic systems can affect work intensity by reducing idle intervals, adjusting performance targets in real time, increasing the visibility of small delays, or assigning tasks without regard to unrecorded demands. Such mechanisms can increase the rate at which employees must switch attention, respond to exceptions, and regulate emotion. The job demands–resources model predicts that elevated demands are especially consequential when not matched by sufficient resources Demerouti et al. (2001) Bakker & Demerouti (2007). The resulting hypotheses concern fatigue, strain, engagement, perceived competence, and the quality of recovery rather than a single undifferentiated construct called stress.

Evidence from Eurofound (2026) associates employee-reported algorithmic management with greater work demands and higher job strain across the European Working Conditions Survey. The report identifies performance monitoring as particularly strongly related to unfavorable health and well-being indicators, including greater exhaustion and work–family conflict under already strained job conditions. These observations are important because they examine mainstream employment at scale. However, their temporal and design features restrict causal attribution: baseline occupational intensity, task structure, industry, and managerial policy may influence both system adoption and employee outcomes.

The electronic monitoring meta-analysis by Ravid et al. (2023) offers a complementary synthesis across studies and contexts. It found an overall association between monitoring and higher worker stress and did not find aggregate evidence of improved performance. The result challenges an uncomplicated assumption that more measurable activity yields more useful output. Yet survey and laboratory studies within a meta-analysis can differ in the duration, intrusiveness, and consequences of monitoring. Mechanism-focused research should therefore assess whether employees anticipate sanctions, experience interruptions, adjust pace, or distrust how anomalous observations will be treated.

Cognitive burden is not determined solely by interface complexity. An apparently simple score may demand extensive mental work when employees must infer hidden thresholds or repeatedly verify that the system has recorded their contributions. Conversely, reliable automated information can reduce search and coordination demands. Experimental and quasi-experimental evaluations should measure task-switching, sustained attention, error rates, perceived workload, work-rest cycles, and objective safety outcomes across implementations. Longer follow-up is necessary to identify adaptations that produce temporary performance gains but accumulate strain.

Procedural Justice, Transparency, and Organizational Legitimacy

Procedural justice refers to how decisions are made, information is communicated, and affected persons are treated. Colquitt (2001) distinguished several dimensions of organizational justice that remain relevant when managerial decisions are partly automated. A scheduling rule may distribute inconvenient shifts consistently while remaining procedurally unfair because workers cannot correct erroneous availability data. An apparently transparent ranking may still be inequitable if it ignores differences in task difficulty. Empirical research should therefore separate distributive consequences, the consistency and revisability of procedures, interpersonal treatment, and the adequacy of explanations.

Algorithmic transparency can mean several things: awareness that automation is used, disclosure of the data collected, intelligibility of the factors influencing a decision, access to an individual explanation, and visibility of the rules used to review mistakes. These levels are not interchangeable. Publishing source code would not necessarily help a worker understand why a shift was lost, while a concise reason code and a usable correction pathway might be highly informative. Ravid et al. (2023) provides evidence that monitoring arrangements perceived as more transparent and less invasive relate to more favorable employee attitudes, although those associations do not independently establish that transparency eliminates harms.

Legitimacy also depends on whether employees regard the decision-making authority as appropriately constrained. A system that makes repeated errors but allows effective correction may sustain more trust than one that announces precision without acknowledging uncertainty. Organizations sometimes describe algorithms as objective to justify decisions; this assertion needs testing against measurement validity, data quality, and distributive outcomes. When the factual basis for a negative evaluation is inaccessible, employees must respond to institutional authority without knowing what conduct would change the result. Such opacity can turn ordinary performance management into a problem of procedural accountability.

Research designs should manipulate or compare distinct governance features: notice, explanation quality, opportunities for worker participation, independent review, and the material consequences of appeal. Outcome measures should include comprehension of reasons, perceived fairness, willingness to report problems, actual correction rates, and future employment decisions. An experimental improvement in expressed trust is evidence about acceptance, whereas a reduction in mistaken penalties is evidence about the performance of a safeguard. Both are relevant to organizational legitimacy, but they answer different scientific questions.

Worker Voice, Resistance, and Collective Coordination

Employee voice is the communication of concerns, information, and suggestions that might improve organizational functioning or challenge harmful practices. Morrison (2014) identifies psychological and institutional factors that influence speaking up or remaining silent. When a work rule is embodied in software, employees may experience uncertainty about who can hear a complaint, which actors have authority to change the system, and whether questioning an automated score will be treated as challenging the organization itself. A feedback button that generates a ticket without effective review offers a weaker form of voice than a procedure that changes decisions and records remedies.

Worker resistance to algorithmic management takes multiple forms. Individuals may modify the timing of tasks, exchange information about uncertain thresholds, coordinate collective demands, or challenge inaccurate ratings. Such behavior can be protective, strategic, cooperative, or counterproductive depending on the rule and the conditions of work. Kellogg et al. (2020) situates resistance within the contested terrain of organizational control. The important analytical point is that employee adaptation can change the realized operation of the system; software specifications alone cannot predict the labor process when workers learn and respond.

Collective institutions influence the distribution of knowledge about management algorithms. Worker representatives can aggregate incidents that appear isolated to individual employees, identify repeated scoring errors, and negotiate how monitoring data may be used. These processes can also generate new information for employers about overlooked task conditions. The empirical question concerns when collective participation improves error detection, legitimacy, and work quality. Comparative studies should identify whether participation is meaningful, whether representatives can access relevant information, and whether organizations actually revise contested rules.

Researchers should avoid treating silence as consent. Employees may withhold complaints because work is precarious, appeals are time-consuming, or the organization controls access to future assignments. Survey measures of willingness to speak must therefore be supplemented by evidence about retaliation risk, material dependence, actual reporting channels, and the rate at which issues are resolved. Organizational learning depends on whether the governance system converts observed worker experience into changes in policy or technology. A system that collects performance signals but excludes worker corrections has a structurally incomplete feedback loop.

Trust, Automation Reliance, and Human–Artificial Decision Relations

Trust in an automated management system includes beliefs about its competence, consistency, intentions as perceived through organizational design, and the practical consequences of reliance. It must be distinguished from interpersonal trust in a supervisor and institutional trust in an employer. Workers may regard a scheduling algorithm as technically accurate while distrusting the organization's choice of objective function. They may understand a risk flag while questioning whether a human reviewer will give contrary evidence adequate weight. These distinctions matter because governance usually combines automated recommendation with the decisions of people who implement or supervise it.

The framework of automation levels developed by Parasuraman et al. (2000) helps distinguish systems that merely present information from those that select or execute a decision. As the level of automation rises, the requirements for monitoring, operator understanding, and recovery from error change. The experiments discussed by Skitka et al. (1999) provide a historical foundation for automation bias, in which decision-makers over-rely on automated assistance. Although these studies did not examine every modern employment platform, they justify examining whether nominally human reviewers independently assess evidence or mechanically endorse a generated score.

Appropriate reliance is calibrated to the system's demonstrated reliability in the relevant context. Workers and managers need information about uncertainty, known failure conditions, error distributions, and when human judgment is necessary. A review mechanism should give the human decision-maker sufficient time, authority, and relevant information to disagree. If supervisors are disciplined for overriding algorithmic recommendations, a theoretical right to intervene may have little practical effect. Conversely, arbitrary human changes can introduce inconsistency or favoritism. The goal of a governance evaluation is to compare decision processes under realistic operating constraints.

The historical horizon of Artificial Era research makes these relations especially important: artificial systems can become durable participants in organizational coordination while employees experience institutional decisions through them. This is a research perspective on how computational agency is operationalized in work systems, not an empirical conclusion about artificial consciousness or personhood. The measurable phenomena are delegated authority, human dependence, predictability, reliance, responsibility, and mechanisms for correcting decisions that affect human livelihoods. These variables can be studied without assuming that the algorithm itself is a moral or legal employer.

Measurement, Construct Validity, and Exposure Assessment

Scientific measurement begins with a precise description of the exposure. One employer may use an algorithm only for roster planning, while another integrates dispatch, monitoring, customer ratings, and automated discipline. Both could appear identical in a broad survey question. A defensible exposure inventory therefore records each function, the frequency and duration of use, the stage of work affected, the authority of the software output, and the nature of any human override. The four dimensions examined by Röttgen et al. (2025) demonstrate a promising way to characterize algorithmic takeover of action-regulation steps, although evidence of reliability in a particular population is not proof of equivalence across countries, languages, or occupational systems.

Construct validity requires a chain from a theoretical concept to an observable indicator and from an indicator to an interpretable claim. A worker's reported irritation is distinct from diagnosed mental illness, just as the pace of completed tasks differs from objective occupational safety. Measures of transparency may capture whether the employee knows that software is used, understands the decision rule, or can obtain an explanation after a disputed event. Each property warrants its own items or observations. The evaluation of scales should include content coverage, structural validity, convergent and discriminant evidence, measurement invariance, and sensitivity to change after an intervention.

Occupational and institutional measurement errors are consequential. Employees may have limited knowledge of the algorithmic function that produced an assignment; managers may overstate automation because software is present; and platform developers may classify manual interventions as exceptions. Triangulation should combine worker survey responses, interviews, system documentation, decision logs, and employer records where access is lawful and ethically justified. The employer's adoption of a tool, the worker's exposure to its outputs, and a worker's perception of being controlled are related but separate variables.

Research should identify whose experiences are absent from its sample. Short-tenure workers, dismissed employees, undocumented or precarious contractors, people with inaccessible digital tools, and those with limited language proficiency may be systematically missing. Sampling only currently active platform accounts can underestimate harms that led people to exit. A strong measurement study therefore discloses its recruitment process, exclusion rules, worker classifications, missing-data patterns, and possible differences between retained and departed workers. These conditions determine which claims about algorithmic management the resulting evidence can sustain.

Experimental, Longitudinal, and Mixed-Methods Evidence

Causal identification requires more than a correlation between monitoring and reported strain. Organizations may implement intensive monitoring precisely where turnover, safety incidents, or productivity pressure are already high. Workers exposed to such tools may differ in occupation, contract type, wages, seniority, or baseline autonomy. Observational analyses can address some differences through measured covariates and longitudinal designs, but unmeasured selection and time-varying confounding remain plausible. The cross-sectional associations described by Eurofound (2026) and Röttgen et al. (2025) should therefore guide hypotheses and intervention design without being interpreted as established population-average causal effects.

A stronger study could use a staggered, policy-driven rollout across comparable sites, measuring the same workers before and after implementation and documenting contemporaneous changes in staffing, pay, demand, and supervision. Where defensible, randomization of nonessential interface options or monitoring transparency may identify component effects. Such designs must account for worker spillovers: a new routing rule imposed on one group may change demand, queue length, or access to assignments for a comparison group. Cluster assignment, interference-aware analysis, and pre-specified outcomes can improve the interpretability of estimated effects.

Qualitative evidence is essential for identifying processes that numerical exposure indicators may miss. The fieldwork of Rosenblat & Stark (2016), Lee et al. (2015), and Möhlmann et al. (2021) shows how workers make sense of opaque interfaces and develop strategies in response. Interviews and ethnography can reconstruct sequences of rule interpretation, perceived threats, workarounds, and unexpected consequences. Their contributions lie in process explanation, theory development, and contextual validity; their samples do not provide prevalence estimates for entire national labor markets. Integration across methods should preserve these distinctions.

An evidence program should report both intended outcomes and harms. Productivity, error prevention, and customer service belong beside stress, autonomy, recovery, safety, pay volatility, discrimination, and worker exit. Outcomes should be pre-specified with clear reference periods and measurements obtained independently where feasible. A longitudinal record that links algorithm version, workplace policy, individual exposure, and downstream consequences provides a stronger basis for assessing how governance changes affect workers than a one-time survey asking whether a respondent approves of artificial intelligence.

Systematic Reviews, Meta-Analytic Findings, and Limits of Generalization

The growing literature contains qualitative case studies, occupational surveys, laboratory experiments, scale-development studies, systematic reviews, and policy research. These genres answer different questions. The review by Chen et al. (2026) synthesized 167 peer-reviewed articles and used BERTopic to identify four configurations of workplace algorithmic management—surveillance, supervision, supplementary, and complementary—based on relations between algorithmic and employee autonomy. This typology helps organize institutional differences; it does not itself estimate an average causal impact on well-being. Treating a review's conceptual categories as validated individual-level psychological constructs would exceed the evidence presented.

The monitoring meta-analysis by Ravid et al. (2023) assembled 94 independent samples and 23,461 participants. Its findings regarding stress, performance, transparency, and invasiveness provide an important quantitative benchmark, especially because electronic monitoring research predates much of the contemporary discussion of artificial intelligence. Interpretation requires attention to the difference between monitoring presence and the behavioral consequences of acting on monitored data. The number of aggregated samples measures scope, not necessarily independence of every mechanism or generalizability to every emerging workplace technology.

New population evidence adds contextual breadth. Eurofound (2026) reports separate employee exposures for computerized monitoring, task allocation, and scheduling in European workplaces using the European Working Conditions Survey 2024. OECD (2025) approaches adoption through employers and managers in six countries. These sources cannot be directly combined into a single universal adoption percentage because the respondents, sampling frames, and operational definitions differ. The correct comparative unit is a clearly defined activity within a specified population and period.

Across the evidence, several limitations recur: technological systems evolve during data collection; work outcomes depend on local institutions; published studies may overrepresent highly visible platforms; and employers may restrict access to model inputs or disciplinary records. Workers also self-select into occupations, platforms, and organizations, and the data produced by employment systems may omit unrecorded cooperation or latent exclusion. A credible research synthesis draws conclusions at the level supported by design and measurement and identifies where independent replication, preregistered intervention studies, or broader sampling could change them.

Fairness, Disparate Impacts, and the Validity of Worker Scores

Fairness requires examining how consequences are distributed across groups and why differences arise. A dispatch system may deliver equal average task counts yet allocate the highest-risk routes disproportionately to workers with fewer alternatives. A performance score may compare employees with unequal workloads or levels of customer hostility. Group fairness assessments should therefore specify the relevant population, exposure, metric, and decision outcome, then evaluate how differences change under plausible alternative policies. The ability to compute a fairness metric does not determine whether its underlying conception of equal treatment fits the employment relationship.

Measurement error and feedback loops complicate algorithmic fairness. If fewer assignments produce fewer opportunities for favorable ratings, the score can reproduce and magnify a previous allocation disadvantage. When supervisors treat the score as a neutral fact, contextual inequality may be incorporated into ostensibly individual performance evaluation. The problem is structurally related to construct validity: the recorded signal should measure the stated criterion in the relevant population, and decisions should be evaluated for effects beyond their immediate predictive accuracy.

Fairness is also procedural. A worker may need access to evidence about their own decisions, an opportunity to supply missing context, and a review by someone authorized to alter a result. Colquitt (2001) provides a psychological rationale for distinguishing fair outcomes from fair procedures and explanations. Explanatory systems must be usable for people with different languages, levels of digital literacy, and occupational constraints. Review procedures that require extensive unpaid time can produce unequal practical access even when formally offered to everyone.

Evaluation should extend beyond demographic averages to occupational subgroups and changing conditions. Errors concentrated among night-shift employees, workers with disabilities, people handling difficult cases, or new entrants can be masked by an apparently satisfactory overall score. Data governance should limit collection of sensitive characteristics while allowing appropriately protected audits that can detect harmful disparities. Decisions about collecting audit variables, retaining logs, and disclosing findings need independent scrutiny because privacy protection and discrimination detection may require carefully reconciled controls.

Privacy, Data Protection, and the Boundaries of Employer Knowledge

Workplace surveillance produces data about employees under conditions of unequal institutional power. The significance of a data field depends on the inference it permits and the consequences attached to it. Location records may help dispatch workers, but repeated traces can reveal private routines; communication metadata may support coordination but also expose relationships and organizing activity. Ball (2010) connects surveillance to the exercise of power, while contemporary governance must additionally consider predictive inferences derived from seemingly ordinary operational logs.

Purpose limitation, proportionality, retention, access controls, and independent correction are practical governance questions. A monitoring system should state the legitimate operational reason for collection, the periods during which it operates, the personal information it stores, who may inspect it, and the conditions for secondary uses. A worker's formal consent is an incomplete measure of meaningful choice when access to employment is contingent on acceptance. The psychological dimensions include anticipated scrutiny, privacy expectations, confidence in safeguards, and concern about future reinterpretation of old behavior.

Privacy governance and the scientific evaluation of monitoring must work together. A researcher with access to identifiable work logs may be able to estimate algorithmic effects precisely but also create new privacy risks. Ethical study design should minimize personal information, separate analysis from disciplinary processes, use secure aggregation where appropriate, and establish limits on employer access to individual research responses. These protections influence participation and response honesty. The absence of complaints in a highly monitored environment should not automatically be taken as evidence of satisfaction.

Modern systems can infer affective states, attention, risk, or future behavior from traces that were never collected as psychological tests. Such inferences are uncertain measurements, and their use for employment decisions raises special problems of validity and disproportionate harm. Research should assess whether the target construct is legitimate, whether predictive claims generalize, and whether the decision actually improves work quality. The European platform-work directive provides an example of explicit legal limits on processing certain personal data in platform work European Parliament & Council (2024a); its scope and implementation timeline must be distinguished from general rules applying in all jurisdictions.

Governance by Design: Accountability, Contestability, and Redress

Algorithmic governance is the allocation of authority, duties, and remedies across the institutions that design, purchase, deploy, and use computational management systems. The central questions are who can authorize a rule, who understands its material effects, who must respond when it harms a worker, and how the rule may be corrected. Governance starts before implementation because a model's objective function, data sources, and thresholds encode practical choices about acceptable organizational outcomes. A review board convened after harm occurs may have insufficient authority if these decisions have already become embedded in contracts and workflows.

Accountability is most concrete when attached to decision classes. Routine task dispatch, exceptional disciplinary action, loss of account access, pay adjustments, and termination differ in the severity and reversibility of their consequences. Each class should have a named responsible function, requirements for notice and evidence preservation, an escalation route, and a final decision-maker with power to provide relief. The design should track how many decisions are challenged, which are corrected, how much time correction takes, and whether a worker experiences lost earnings during the appeal. A statement that a human is in the loop is meaningful only when the human has relevant competence, independence, time, and override authority.

Explainability also has several audiences. Workers need actionable reasons about consequential individual outcomes. Supervisors need uncertainty information and guidance about exceptions. Auditors need evidence about populations, model updates, training data provenance, and the distribution of errors. Regulators and worker representatives may need aggregate statistics and controls over impermissible data uses. Publishing only general model documentation may not meet the practical needs of an employee appealing a mistaken absence penalty. The governance system should be evaluated by the quality of decisions and remedies, not simply by the existence of policy statements.

Technical controls and institutional procedures reinforce each other. Versioned rule logs help reconstruct what an algorithm decided at a particular moment; access restrictions protect worker information; change management prevents silent alterations to evaluation criteria; and periodic impact reviews can reveal accumulated exclusion. Tabassi (2023) provides a general risk-management vocabulary for governing artificial intelligence, whereas employment-specific obligations must be established from applicable labor and data-protection law. Effective governance joins technical auditability to worker voice, enforceable responsibilities, and corrective capacity.

Comparative Institutions and the European Platform Work Directive

Employment institutions regulate algorithmic management in different ways. Applicable duties depend on worker status, jurisdiction, data category, collective agreements, and whether software merely informs a manager or produces an employment decision. A scientific treatment should identify the actual authority of a legal instrument rather than treat a policy proposal as binding law in every setting. The European Union's Directive (EU) 2024/2831 addresses platform work and contains specific provisions concerning automated monitoring and decision systems European Parliament & Council (2024a). It is a platform-work instrument, and its scope should not be expanded rhetorically to all automated employee scheduling everywhere.

The directive's algorithmic-management provisions address limits on certain personal-data processing, information about systems, human oversight, review of significant decisions, health and safety, and consultation. Its requirements for human reviewers include competence, training, and authority to override automated decisions. It also addresses psychosocial and ergonomic risks associated with automated monitoring and decision support. These are institutional obligations and design expectations; they do not establish empirical proof that every mandated procedure, once implemented, will eliminate stress or discrimination. Research must evaluate actual compliance and worker outcomes following implementation.

As of November 3, 2026, the directive's stated deadline for member-state transposition is December 2, 2026 European Parliament & Council (2024a). Implementation therefore must be examined at the level of national law and organizational practice rather than presumed complete across all member states. The timing is substantively important because researchers comparing workplaces before and after the deadline may encounter transitional policies, anticipatory employer behavior, and varying enforcement capacity. The related EU AI Act provides an additional regulatory context for some employment-related AI systems, but its detailed application depends on the specific system, effective dates, and subsequent legislative developments European Parliament & Council (2024b).

In the United States and other jurisdictions, governance may involve overlapping labor, discrimination, privacy, consumer-reporting, and sector-specific rules rather than a single dedicated algorithmic management statute. Their application cannot be inferred from broad claims that an automated tool is ethical or legally compliant. Multinational comparative research should code statutory coverage, collective representation, enforcement institutions, data-protection rights, and actual remedies. These conditions may moderate psychological effects as strongly as differences in algorithmic technique.

Sociotechnical Design and Participatory Alternatives

A sociotechnical approach studies algorithms as parts of jointly organized human and technical work systems. The question is not simply how accurate the model is but whether task design, staffing, worker competence, information, and institutional authority support the same legitimate goals. A system that optimizes assignment speed may perform poorly when it creates unsafe task sequences or prevents employees from dealing with unusual customer needs. The systematic review by Chen et al. (2026) differentiates surveillance, supervision, supplementary, and complementary configurations of management, illustrating how distinct divisions of labor generate different employee experiences.

Participatory design offers a practical strategy for identifying missing variables and undesirable incentives before deployment. Workers can describe productive activities that current metrics omit, such as mentoring, preventive maintenance, emotional labor, and exception handling. Managers can identify needs for reliable forecasting, evidence preservation, and fair allocation. Technical teams can model how alternative decision rules affect outcomes under real demand conditions. The resulting process should document disagreement about objectives rather than assume that consultation automatically produces a unified interest.

The quality of participation depends on decision rights. A workshop that only solicits comments after a system has been purchased provides limited opportunity to change its central rules. Meaningful involvement includes access to relevant information, a chance to test alternatives, protection against retaliation, and a documented route from worker input to design changes. The OECD's 2025 policy analysis emphasizes consultation and concerns about accountability and worker health OECD (2025). Those recommendations are consistent with organizational psychology, but their effectiveness should be tested with implementation records and employee outcomes.

Good design also makes failures observable and recoverable. Workers and supervisors need to know when data are missing, when a model is uncertain, and which cases require escalation. Human override should be accompanied by logging and appropriate review to reduce arbitrary interventions without discouraging justified exceptions. Evaluation should compare overall productivity, employee health, job quality, fairness, service reliability, and organizational learning. A technically impressive algorithm whose incentives erode the conditions required for competent work is a weak organizational solution.

Research Frontiers in the Artificial Era

The next stage of research concerns the integration of prediction, generative models, conversational interfaces, and increasingly adaptive allocation policies. When automated managers explain decisions through natural-language dialogue, the apparent social qualities of an interface may affect understanding and willingness to comply. Yet explanation fluency is not equivalent to causal fidelity, and a plausible generated reason may not accurately represent the decision process. Future studies should compare explanations with verifiable rule traces and measure whether employees can identify, challenge, and correct an actual error.

Adaptive systems introduce a temporal problem. An algorithm that changes scoring thresholds, routing priorities, or feedback policies during operation alters the exposure being studied. Investigators must preserve model versions, records of policy change, and the conditions under which workers encountered each version. Independent studies should evaluate not only average outcomes but path dependence: whether early low ratings reduce later opportunities, whether reduced opportunities generate further low ratings, and whether corrective interventions can reverse these trajectories. Such feedback loops can create durable inequality from comparatively small initial differences.

Another research priority is understanding collective effects. Most familiar psychological instruments describe individuals, while algorithmic management often reallocates opportunities within teams or across entire labor markets. Increased productivity at one site may be achieved by offloading waiting time to contractors; reduced response time may depend on increasing demands on workers who are nominally off duty. Systems-level measures need to account for these transfers. Network and multilevel designs can test whether outcomes change when a rule is deployed widely rather than for isolated participants.

The transition From Homo to Artificial makes the changing institutional location of decision-making a major object of psychological research. Artificial systems may become stable coordinators of work, but empirical inquiry must continue to distinguish the system's computational behavior from the employer's legal and organizational responsibilities. Foundational questions concern how human agency is supported or constrained, how confidence in algorithmic outputs is calibrated, how meaningful recourse is preserved, and how institutions establish responsibility across evolving human–artificial arrangements. These are testable propositions about work systems and their psychological consequences.

Conclusion

Algorithmic management is a consequential reorganization of workplace authority through computational systems that assign tasks, schedule labor, monitor activity, evaluate performance, and mediate access to economic opportunity. Its central psychological effects arise from the interaction of decision rights, information asymmetry, workload, perceived justice, organizational support, and the ability to contest consequential errors. Historical research on managerial control and electronic monitoring provides the foundations for understanding contemporary artificial intelligence deployments. Theoretical traditions in job demands and resources, self-determination, organizational justice, action regulation, and automation reliance offer complementary mechanisms rather than a universal explanation.

The evidence supports several robust distinctions. Qualitative studies show how workers learn, accommodate, and contest algorithmic control in specific labor environments Rosenblat & Stark (2016) Wood et al. (2019) Möhlmann et al. (2021). A broad quantitative synthesis finds that electronic performance monitoring is associated with greater worker stress and does not show an overall improvement in work performance Ravid et al. (2023). Contemporary studies indicate that distinct algorithmic management functions have different associations with job autonomy, work pace, irritation, and strain Röttgen et al. (2025) Eurofound (2026). None of these findings authorizes the claim that every deployment has the same effect or that observational associations by themselves establish causation.

A mature scientific evaluation must therefore specify the managerial function, population, work conditions, measured outcome, institutional context, and evidential design. It must test whether gains in coordination are accompanied by sustainable work conditions, valid measures of contribution, meaningful decision latitude, and effective remedies. Fairness and transparency should be evaluated through actual decisions and their consequences. Governance is successful when people can understand how rules affect them, challenge mistakes, participate in relevant design choices, and obtain appropriate correction from accountable institutions.

The central synthesis is that the psychological quality of algorithmically managed work depends on the institutional organization of computational authority. Algorithms can increase coordination and information while redistributing discretion, visibility, and risk. Their effects are jointly produced by technical design, labor relationships, and governance practices. Future research should connect causal intervention evidence, longitudinal work trajectories, worker-centered measurement, and independent audits to build a cumulative science of employment in the Artificial Era.

References

  • Bakker, A. B., & Demerouti, E. (2007). The job demands–resources model: State of the art. Journal of Managerial Psychology, 22(3), 309–328. https://doi.org/10.1108/02683940710733115
  • Ball, K. (2010). Workplace surveillance: An overview. Labor History, 51(1), 87–106. https://doi.org/10.1080/00236561003654776
  • 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, Article 102994. https://doi.org/10.1016/j.ijinfomgt.2025.102994
  • Colquitt, J. A. (2001). On the dimensionality of organizational justice: A construct validation of a measure. Journal of Applied Psychology, 86(3), 386–400. https://doi.org/10.1037/0021-9010.86.3.386
  • Demerouti, E., Bakker, A. B., Nachreiner, F., & Schaufeli, W. B. (2001). The job demands–resources model of burnout. Journal of Applied Psychology, 86(3), 499–512. https://doi.org/10.1037/0021-9010.86.3.499
  • Eurofound. (2026). Algorithmic management in European workplaces. Publications Office of the European Union. https://doi.org/10.2806/3220989
  • European Parliament & Council of the European Union. (2024a). Directive (EU) 2024/2831 of 23 October 2024 on improving working conditions in platform work. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024L2831
  • European Parliament & Council of the European Union. (2024b). Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
  • Karasek, R. A., Jr. (1979). Job demands, job decision latitude, and mental strain: Implications for job redesign. Administrative Science Quarterly, 24(2), 285–308. https://doi.org/10.2307/2392498
  • Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366–410. https://doi.org/10.5465/annals.2018.0174
  • Lee, M. K., Kusbit, D., Metsky, E., & Dabbish, L. (2015). Working with machines: The impact of algorithmic and data-driven management on human workers. Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, 1603–1612. https://doi.org/10.1145/2702123.2702548
  • Möhlmann, M., Zalmanson, L., Henfridsson, O., & Gregory, R. W. (2021). Algorithmic management of work on online labor platforms: When matching meets control. MIS Quarterly, 45(4), 1999–2022. https://doi.org/10.25300/MISQ/2021/15333
  • Morrison, E. W. (2014). Employee voice and silence. Annual Review of Organizational Psychology and Organizational Behavior, 1, 173–197. https://doi.org/10.1146/annurev-orgpsych-031413-091328
  • OECD. (2025). How widespread is algorithmic management in workplaces? OECD Publishing. https://doi.org/10.1787/cda7a114-en
  • Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans, 30(3), 286–297. https://doi.org/10.1109/3468.844354
  • Ravid, D. M., White, J. C., Tomczak, D. L., Miles, A. F., & Behrend, T. S. (2023). A meta-analysis of the effects of electronic performance monitoring on work outcomes. Personnel Psychology, 76(1), 5–40. https://doi.org/10.1111/peps.12514
  • Rosenblat, A., & Stark, L. (2016). Algorithmic labor and information asymmetries: A case study of Uber’s drivers. International Journal of Communication, 10, 3758–3784. https://ijoc.org/index.php/ijoc/article/view/4892
  • Röttgen, C., Dzaack, H., Herbig, B., Weinmann, T., & Müller, A. (2025). Algorithmic management: Psychological measurement and associations with work design and mental strain. BMC Psychology, 13, Article 1327. https://doi.org/10.1186/s40359-025-03680-2
  • 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
  • Skitka, L. J., Mosier, K. L., & Burdick, M. (1999). Does automation bias decision-making? International Journal of Human–Computer Studies, 51(5), 991–1006. https://doi.org/10.1006/ijhc.1999.0252
  • Tabassi, E. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.100-1
  • Wood, A. J., Graham, M., Lehdonvirta, V., & Hjorth, I. (2019). Good gig, bad gig: Autonomy and algorithmic control in the global gig economy. Work, Employment and Society, 33(1), 56–75. https://doi.org/10.1177/0950017018785616