Content uploaded by Andreas Müller
Author content
All content in this area was uploaded by Andreas Müller on Sep 25, 2024
Content may be subject to copyright.
Frontiers in Artificial Intelligence 01 frontiersin.org
Algorithmic management and
human-centered task design: a
conceptual synthesis from the
perspective of action regulation
and sociomaterial systems theory
CarstenRöttgen
1*, BrittaHerbig
2, TobiasWeinmann
2 and
AndreasMüller
1
1 Institute of Psychology, Work and Organizational Psychology, University of Duisburg-Essen, Essen,
Germany, 2 Institute and Clinic for Occupational, Social and Environmental Medicine, LMU University
Hospital, LMU Munich, Munich, Germany
This paper aims to explain potential psychological eects of algorithmic
management (AM) on human-centered task design and with that also workers’
mental well-being. For this, welink research on algorithmic management (AM)
with Sociomaterial System Theory and Action Regulation Theory (ART). Our
main assumption is that psychological eects of sociomaterial systems, such
as AM, can beexplained by their impact on human action. From the synthesis
of the theories, mixed eects on human-centered task design can bederived:
It can beexpected that AMcontributes to fewer action regulation opportunities
(i.e., job resources like job autonomy, transparency, predictability), and to lower
intellectual demands (i.e., challenge demands like task complexity, problem
solving). Moreover, it can beconcluded that AMis related with more regulation
problems (i.e., hindrance demands like overtaxing regulations) but also fewer
regulation problems (like regulation obstacles, uncertainty). Based on these
considerations and in line with the majority of current research, it can beassumed
that the use of AMis indirectly associated with higher risks to workers’ mental
well-being. However, wealso identify potential positive eects of AMas some
stressful and demotivating obstacles at work are often mitigated. Based on these
considerations, the main question of future research is not whether AMis good or
bad for workers, but rather how work under AMcan bedesigned to behumane.
Our proposed model can guide and support researchers and practitioners in
improving the understanding of the next generation of AMsystems.
KEYWORDS
digitalization, artificial intelligence, work design, Job Demands-Resources Model,
work stress, motivation, self-determination
1 Introduction
Intelligent technological systems that have the capability to learn and to make autonomous
decisions through algorithmic pattern detection permeate and fundamentally transform our
work life (Cascio and Montealegre, 2016). Such systems are also increasingly used to take over
decision-making in organizations from human actors such as managers (Benlian etal., 2022).
ese types of technological systems that overtake managerial decisions at work are referred
OPEN ACCESS
EDITED BY
Federica Caaro,
Roma Tre University, Italy
REVIEWED BY
Monica Molino,
University of Turin, Italy
Andreia de Bem Machado,
Federal University of Santa Catarina, Brazil
*CORRESPONDENCE
Carsten Röttgen
carsten.roettgen@uni-due.de
RECEIVED 31 May 2024
ACCEPTED 23 August 2024
PUBLISHED 25 September 2024
CITATION
Röttgen C, Herbig B, Weinmann T and
Müller A (2024) Algorithmic management and
human-centered task design: a conceptual
synthesis from the perspective of action
regulation and sociomaterial systems theory.
Front. Artif. Intell. 7:1441497.
doi: 10.3389/frai.2024.1441497
COPYRIGHT
© 2024 Röttgen, Herbig, Weinmann and
Müller. This is an open-access article
distributed under the terms of the Creative
Commons Attribution License (CC BY). The
use, distribution or reproduction in other
forums is permitted, provided the original
author(s) and the copyright owner(s) are
credited and that the original publication in
this journal is cited, in accordance with
accepted academic practice. No use,
distribution or reproduction is permitted
which does not comply with these terms.
TYPE Hypothesis and Theory
PUBLISHED 25 September 2024
DOI 10.3389/frai.2024.1441497
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 02 frontiersin.org
to as data-driven or algorithmic management systems (AM) (Lee
etal., 2015). So far, AMare mainly used to manage work in the
so-called platform economy (e.g., Uber, Amazon MTurk) (Rosenblat
and Stark, 2016), or in warehouse logistics (Schmierl etal., 2022) with
rather low skilled jobs. However, AMis also beginning to transform
traditional organizations and higher qualied jobs, such as engineering
(Bakewell et al., 2018) or healthcare (Mashar et al., 2023).
Consequently, one can predict that AMwill bean important factor for
the design of future workplaces which deserves a broad attention of
I-O psychology concerned with humane work design.
e increasing shi in agency from people to technology by the
introduction of AMat work is seen as a fundamental new quality of
work (Benlian etal., 2022; Gagné et al., 2022; Kellogg etal., 2020;
Parker and Grote, 2022). For example, AMis considered to radically
recongure the “contested terrain” of organizational control as one of
the most fundamental aspects of the employer – employee relation
(Kellogg etal., 2020). As the ability to inuence one’s own work is one
of the key job resources for workers, AMis also expected to impact
the mental well-being in future work places (Kinowska and
Sienkiewicz, 2022).
Although comprehensive I-O psychological research in the last
decades has expanded our understanding on the design of a
meaningful and healthy work environment (e.g., Parker etal., 2017a),
the psychological understanding of AMand its possible consequences
for workers’ well-being is only just beginning. With this conceptual
paper webuild on empirical and descriptive studies of AMpractices
to complement and extent recent psychological knowledge about the
impact of AMon work design (Parent-Rocheleau and Parker, 2022)
and workers motivation (Gagné etal., 2022) by looking at AMfrom
action regulation theory (ART) (Frese and Zapf, 1994; Hacker, 2003;
Zacher, 2017), and sociomaterial-system perspective (Cascio and
Montealegre, 2016; Landers and Marin, 2021; Orlikowski and Scott,
2008). Webelieve that by integrating both theoretical perspectives
new I-O psychological insights about the human centered design of
algorithmically managed workplaces can bederived:
(1) So far, the starting point for the consideration of AMhas
mainly been its managerial functions. e perspective of ART will
additionally contribute to identify psychologically important
functions of AMfrom the perspective of involved workers. ART will
help to further reveal the underlying mechanisms through which the
functions of AMwill aect the design of work tasks, work behavior,
and nally also workers’ well-being and vice versa. us, the
perspective of ART allows to further systematize the psychological
understanding of AMfunctions across a wide range of jobs and
industries. (2) I-O psychological research on work design has oen
been criticized to neglect that psychological and social phenomena
in organizations are related with material aspects of technologies
(Landers and Marin, 2021; Orlikowski and Scott, 2008; Parker etal.,
2017b). e sociomaterial-system perspective highlights that AMis
a fusion of material aspects (e.g., computer networks, mobile digital
devices, or the interface of a soware program) as well as
psychological/social aspects (e.g., goals and values of involved
organizations and workers) (Orlikowski and Scott, 2008). Taking the
sociomaterial-system perspective will therefore contribute to deepen
our theoretical understanding about which “objectiable”
characteristics and functions of AMaect workers’ actions and well-
being. (3) From a practical perspective, this understanding of the
interrelation between the material and psychological/social aspects
of AMis one important precondition to name and substantiate
concrete starting points for the human-centered design of modern
digitalized workplaces. us, our approach helps to better
understand the eect of technology on humane work design, an
issue that is rarely the focus in existing I-O research so far (Parker
etal., 2017b).
In the following sections, wewill rst describe what organizational
research to date understands by AM. Wethen give a brief overview of
the main assumptions of ART as well as of the conception of
technologies as sociomaterial congurations. In a next step,
wedevelop a new theoretical founded denition of AMbefore nally
deriving propositions about the eects of AMon human-centered
work design.
1.1 Algorithmic management
e term AM refers to learning algorithms that carry out
automated data-driven coordination and control of workers without
explicit involvement of human managers or other human agents at
work (Benlian etal., 2022; Gagné etal., 2022; Möhlmann etal., 2021;
Noponen etal., 2023; Parent-Rocheleau and Parker, 2022). AMtakes
over typical functions and responsibilities of lower and middle managers
like assigning tasks, scheduling work, monitoring task
accomplishment, evaluating workers performance, providing rewards
or sanctioning, and even making human resource management
decisions, like termination of work (Benlian etal., 2022; Gagné etal.,
2022; Parent-Rocheleau and Parker, 2022).
e implementation of AM is tied to interconnected digital
technologies that enable the collection, storage, processing, and
transmission of data: Basically, learning algorithms are informed by
large quantities of data that are continuously recorded by digital
devices (e.g., cell phones, tablets, or handhelds) of involved
stakeholders like workers and customers. Depending on the job, a
multiplicity of data of varying depth and magnitude is recorded, such
as behavioral data of workers (e.g., movement data), their active inputs
(e.g., about completed tasks), but also physiological reactions (e.g.,
galvanic skin response) (Cram and Wiener, 2020). e data are
transferred via internet connections and stored on digital platforms
where they are processed, and from where algorithmically generated
instructions and feedback are sent back to the digital devices
of workers.
From an economic perspective, AMparticularly oers enormous
growth opportunities and adaptability of business models, by
facilitating the exible on-time coordination of large numbers of
on-demand tasks, like food delivery or transport of passengers
(Benlian et al., 2022; Möhlmann et al., 2021). One prototypical
example is the globally operating platform company Uber that
operates a smartphone app which connects mainly independently
operating drivers with passengers to provide “ridesharing” services
(Rosenblat and Stark, 2016). Uber is an impressive example that
AM can operate global companies that have a shareholder value
comparable to large traditional corporations, such as Volkswagen,
with a comparatively small amount of material resources and regularly
employed workers. With the development of this so called “gig
economy,” AMhas opened a whole new labor market providing the
opportunity to earn money with a very low entry threshold (Wu and
Huang, 2024).
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 03 frontiersin.org
From the perspective of workers, these potentially tremendous
entrepreneurial advantages seem to be outweighed by an array of
drawbacks. Recent initial reviews indicate that current applications of
AM have predominantly negative eects on the quality of work
(Parent-Rocheleau and Parker, 2022), as well as on individual
outcomes such as motivation (Gagné etal., 2022) and other aspects of
mental well-being (Kinowska and Sienkiewicz, 2022). For example,
similar to traditional eciency-driven Taylorist management systems,
it is suspected that AMmight berelated to higher workload and
reduced job autonomy (Kellogg etal., 2020; Parent-Rocheleau and
Parker, 2022); a combination that is well-known to bea signicant
psychosocial health risk for workers (Karasek, 1979; see also eorell
etal., 2015). Even more signicant, AMis also suspected to possess
entirely new qualities compared to traditional management
approaches particularly by replacing important social agents at work
like human managers, thus literally contributing to a “dehumanization”
(Lamers etal., 2022) of work, and by making work more opaque and
unpredictable compared to repetitive but predictable traditional work
systems (Noponen etal., 2023).
However, AMcan bedesigned and implemented in quite dierent
ways, so that the psychosocial eects of AM are not necessarily
predetermined (Benlian et al., 2022; Cram and Wiener, 2020;
Noponen etal., 2023). One of the few available simulation studies
indicates, that single aspects of AM can also be designed in a
supportive and motivating way if psychological needs of workers are
taken into account (Sailer et al., 2017). From a psychological
perspective, it is therefore important to better understand the
opportunities for and limitations of the human-centered task design
under the conditions of AM.
1.2 Action regulation theory
Human action like taking care of a person, delivering a good,
repairing a car is (still) the core of work (Hacker, 2003). Action
Regulation Theory (ART; Frese and Zapf, 1994; Hacker, 2003)
captures the cognitive processes of human action regulation and
explains its relation to desirable individual outcomes, such as
workers’ well-being, personal growth, intrinsic motivation, and
good performance. ART assumes that these desirable individual
outcomes are closely intertwined with the extent to which the
work environment either promotes or impedes autonomous
actions of workers with scope for decision making and high levels
of personal control. With that, ART ties together task
characteristics, like job autonomy, with psychological processes
and states, like motivation and well-being. Therefore, ART can
beuseful to deepen our psychological comprehension of AMas
part of the work environment, and to better understand the
potential psychological effects of AM.
Basically, ART illuminates human action from two perspectives
(Frese and Zapf, 1994; Hacker, 2003): On the one hand, the theory
considers action as cyclical sequences of action phases with a certain
logical order: goal development, planning and orientation, executing
and monitoring, and feedback. In addition, the theory distinguishes
three hierarchical organized cognitive levels of action regulation
(Hacker, 2003): On the sensumotor level actions are automated and
regulated without conscious attention (e.g., riding a bicycle). On the
knowledge-based level well-practiced actions are executed that are
based on plans stored in memory and that must be adapted to a
specic situation. Actions on that level can but do not have to
beconsciously regulated (e.g., navigating the bicycle on the well-
known route to work). Action regulation on the highest, the
intellectual level, is characterized by a conscious development and
activation of goals and plans for the regulation of complex activities.
is takes place when no ready-to-use pattern of activity exists, that
is, when non-routine actions are regulated (e.g., nding a way in an
unknown town).
ART particularly emphasizes the importance of goals, in the sense
of the mental representation of a future outcome, for human action
(Hacker, 2003). Goals trigger actions, direct attention during action,
and are the benchmarks to evaluate the progress of action. us, goals
align the complete action sequence. Goals also integrate cognitive as
well as motivational processes of action regulation (Frese and Zapf,
1994): From a cognitive perspective, the iterative self-regulated
development of adequate goals is for example an integral demand of
intellectual level action regulation in non-routine problem-solving
tasks (Hacker, 2003). From a motivational perspective, specic and
dicult goals increase the perseverance and eort of workers during
task accomplishment (Locke and Latham, 2016). Moreover, self-set
and internalized goals contribute to self-determined and intrinsically
motivating work (Deci etal., 2017).
From these basic assumptions ART derives normative
standards for the humane design of work tasks, that are in
accordance with well-known theories about health-related work
design, like the Job Demands-Resources Model (JD-R, Bakker
and Demerouti, 2017), that distinguishes between job demands,
i.e., working conditions that require physical and mental energy
from the employee and are mainly related to strain processes, and
job resources, i.e., working conditions that directly or indirectly
satisfy basic human needs and might therefore trigger
motivational and salutogenetic processes. Basically, the JD-R
model (Bakker and Demerouti, 2017; Bakker et al., 2023;
Demerouti, 2020) distinguishes between two kinds of working
conditions that are related with two kinds of health related
psychological mechanisms: (a) Job demands require the use of
physical and mental energy. Mediated through stress-related
mechanisms, they can represent risk factors for mental health.
Health risks exist, for example, when job demands are
experienced as uncontrollable (Karasek, 1979) or when they
hinder or complicate the accomplishment of work tasks (LePine
et al., 2005). (b) Job resources can strengthen mental health
through motivational mechanisms. In this context, job resources
are those working conditions that directly correspond to basic
human needs or that are instrumental in satisfying these basic
needs [e.g., latitudes at work that correspond to the basic need
for autonomy and self-determination (Deci etal., 2017; Hackman
and Oldham, 1976)]. Job resources can also mitigate the
potentially negative effects of work demands (Karasek, 1979).
Accordingly, health risks arise when job resources are not
available to a sufficient extent (Lesener etal., 2019). Due to its
good empirical evidence (Lesener etal., 2019), its high degree of
generalization and its compatibility with other central
psychological theories of work design (Hacker, 2003; Hackman
and Oldham, 1976; Karasek, 1979), the JD-R model is suitable to
explore and classify potentially novel conditions of digitalized
work (Demerouti, 2020).
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 04 frontiersin.org
1.2.1 Job resources from the perspective of ART
From the perspective of job resources, which in ART are referred
to as regulation opportunities, ART suggests the concept of complete
tasks and activities as a gold standard for work design (Hacker, 2003):
Work tasks are sequentially complete when their design oers
latitude about the action sequence described above, particularly
having the opportunity to develop self-set goals as well as to plan the
action steps and measures to reach these goals. Moreover, tasks are
hierarchically complete when they require all levels of action
regulation, i.e., automated sensumotor regulated actions, as well as
knowledge-based and intellectual regulated actions. A typical
example of an incomplete task in a “traditional” job would be a
partialized routine task on an assembly line in the automobile
production, where a narrowly dened action step (such as attaching
a car body part) simply must beperformed over and over again,
without requiring any specic goal development or planning. A more
complete task would be the automobile production by semi-
autonomous groups that can co-determine and plan their own work
processes, as was found in some Volvo plants until the 1990s
(Sandberg, 1993).
Complete tasks contain specic job resources that are well-known
from other established work design-models like the Job Demand-
Control Model (Karasek, 1979) and the Job Characteristics Model
(Hackman and Oldham, 1976; Humphrey etal., 2007): e main
characteristic of complete tasks is a high level of job autonomy that
oers the worker leeway to develop goals and exert control about the
complete sequence of action (Frese and Zapf, 1994; Hacker, 2003).
Complete tasks should also be related with further intrinsically
motivating task characteristics such as task identity, i.e., employees’
perception that they contribute to a complete piece of work (Hackman
and Oldham, 1976; Humphrey etal., 2007). Another important job
resource included in complete tasks is feedback, that provides workers
with helpful information about their progress toward reaching goals
and thus with learning opportunities (Zacher, 2017).
According to the Job Demand-Control Model (Karasek, 1979),
complete tasks should therefore help to avoid chronic stress and
associated health risks because particularly the higher job autonomy
or decision latitude increases the internal control of workers to adjust
their work tasks according to their personal abilities and skills (Hacker,
2003). Moreover, according to the Job Characteristics Model
(Hackman and Oldham, 1976; Humphrey etal., 2007) the perception
of job autonomy, task identity and helpful feedback should increase
workers’ perceptions of psychological states like responsibility,
meaningfulness, and accomplishment in their work, which altogether
should contribute to intrinsic motivation.
ART also suggests that transparency and predictability of work
tasks are further important job resources that enable workers to take
control over work processes (Frese and Zapf, 1994). ey both
somewhat go beyond the characteristics of the specic work task and
also relate to the design of the work organization and the wider work
environment. Transparency refers to the knowledge about the
meaning of relevant task related information (e.g., the meaning of a
specic work object, like a tool or a working material) and allows the
worker to develop an adequate operative image system, i.e., an
adequate mental model, of its work task (Hacker, 2003). A lack of
transparency makes it dicult to interpret information appropriately
and limits the possibilities to develop adequate task goals and action
plans. Whereas transparency refers to the present work situation,
predictability refers to the possibility to foresee future work tasks,
changes, or problems. Predictability is an essential prerequisite for
forward-looking action planning that goes beyond a mere reaction to
action stimuli and has been shown to bea further important health
relevant job resource for employees (Väänänen etal., 2008).
1.2.2 Job demands from the perspective of ART
In ART, job demands are conceptualized as regulation
requirements that are related to properties of the hierarchic-sequential
organization of action (Frese and Zapf, 1994). In accordance with the
Challenge-Hindrance Stressor Framework (LePine et al., 2005),
regulation requirements are primarily seen as motivating and
learning-promoting intellectual demands (i.e., challenge demands).
On the one hand, coping with these intellectual demands requires
individual mental and physical resources. ese individual resources
will beexhausted at some point and must therefore beregenerated in
order to avoid impairment of well-being (Sonnentag etal., 2022). On
the other hand, they are motivating, as the accomplishment of
challenging demands meets our basic need for competence (Deci
et al., 2017). Human-centered work should therefore not aim to
reduce such challenging task demands, but to design demands in such
a way that they correspond to the skill level of the employees.
Complete tasks not only contain heath promoting and motivating
job resources, but they also require a higher level of cognitive
demands, as goals and action plans may need to bedeveloped and
adapted during action execution. Wewant to distinguish between two
types of challenging job demands that go along with complete tasks:
task complexity and problem solving that are particularly important
from the perspective of ART.
Whereas job control can bedened in terms of available decision
possibilities, task complexity implies decision necessities. Task
complexity increases with the number of task goals and the degree of
interconnection between these goals. Higher task complexity leads to
a higher degree of regulation requirements, as pursuing one task goal
positively or negatively aects multiple other task goals. Previous
research suggests that high task complexity promotes satisfaction, as
coping with complex demands goes hand in hand with a high sense of
competence; however, it might also berelated with the experience of
overload, as employees might perceive those tasks as too complicated
and overtaxing (Humphrey etal., 2007).
Problem solving particularly focuses on the extent to which a task
requires the development of novel solutions or ideas (Humphrey etal.,
2007). Particularly, tasks that are regulated on the intellectual level are
oen novel tasks with initially unclear goals and therefore uncertain
action plans. While this implies high information processing demands,
it also provides the potential to learn and grow by solving unknown
problems (Frese and Zapf, 1994).
In line with the Challenge-Hindrance Stressor Framework
(LePine etal., 2005), ART also enables the derivation of a taxonomy
of stressful and demotivating hindrance demands, in terms of action
regulation problems. Roughly, regulation problems can
bedistinguished in regulation obstacles, regulation uncertainty, and
overtaxing regulation (Frese and Zapf, 1994): Regulation obstacles
refer to external barriers or hindrances (e.g., unexpected disruptions)
that individuals encounter when trying to accomplish tasks or goals.
Regulation obstacles unnecessarily increase the eort required to
perform a task and might contribute to frustration and stress
experience of workers (Baethge etal., 2015). Regulation uncertainty
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 05 frontiersin.org
refers to the ambiguity or lack of clarity in one’s task goals, or action
plans. Uncertainty can stem from factors such as unclear instructions,
conicting goals, or rapidly changing circumstances, and lead to
insecurity and doubts about the clear path forward to goal
accomplishment. Finally, overtaxing regulation occurs when a work
task places excessive demands (e.g., tight deadlines, simultaneous
tasks, information overload) on the workers. Tasks can therefore
beexperienced as overwhelming and unmanageable (Frese and Zapf,
1994; Zacher and Frese, 2018).
With the introduction of AM, parts of the action sequence are
transferred from humans to algorithms (Parent-Rocheleau and Parker,
2022) changing the job demands and resources for workers and with
that also the design of their work tasks.
In sum, from the perspective of ART, the psychological assessment
of AMshould therefore consider the extent to which AMinuences
the design of work tasks—in particular regulation opportunities (i.e.,
job resources), motivating and learning-promoting intellectual
demands (i.e., challenge demands) or stressful regulation problems
(i.e., hindrance demands).
1.3 Technologies as sociomaterial systems
As stated above, AMis tied to technologies. Landers and Marin
(2021) dene technologies as “[…] a collection of enduring physical
and/or digital materials that dynamically aord individual and/or
collective goal-directed action” (p.240). us, one basic purpose of a
technology is the reinforcement and enhancement of human
capabilities to reach goals.
is denition shows the close link between technology and
action regulation. For example, the physical and material features of a
hammer—its sturdy long handle and heavy head—enhances the
transfer of our arm power to or exert physical forces on objects.
Without such tools, with our bare hands or minds, even rather simple
tasks, like driving a nail into a wall to hang up a picture, would bevery
dicult or even impossible to perform. Similarly, the databases,
internet connections, digital devices, and soware designs of an
AM system enable a worldwide expansion of business models of
platform companies like Uber that would not bepossible without such
digital technologies. A technology is therefore seen as a merger of
individual/social aspects—e.g., individual or collective goals as well as
the human knowledge and capabilities to reach these goals—and
material aspects—e.g., the compilation of physical or digitalized
features of a tool, machine or computer—that are specically designed
to enhance human capabilities for goal-directed behavior.
Technologies thus are oen seen as sociomaterial congurations
(Leonardi, 2012), i.e., a functional amalgamation of material and
human/social aspects.
In this regards, several authors (Landers and Marin, 2021;
Orlikowski and Scott, 2008; Zammuto etal., 2007) refer to Gibson
(1977) concept of aordances that gets to the essence of the
sociomaterial congurations of technologies and interrelate them with
psychosocial phenomena and processes. “Aordances can bedened
as the perception of whether the features of a technology can beused
to achieve goal-directed actions” (Landers and Marin, 2021, p.240).
As such, aordances are possibilities and at the same time restrictions
for action that arise from the connection between people and material
objects (Gibson, 1977).
e perspective of aordances helps to understand how the objective
design of our work environment facilitates or impedes or even precludes
human actions. For example, a hammer aords actions in which physical
forces must beexerted on other objects. At the same time the nature of
a hammer impedes other actions. It is less suitable for cutting materials
precisely or it will hardly beever used to paint an object. In the same way,
the AMsystem of Uber is optimized to provide a exible demand-driven
transport service for passengers. But from the drivers’ point of view, this
might come at the expense of exerting control about their work
(Rosenblat and Stark, 2016). us, the aordances of the technical design
of our working environment may not bedetermining, but at least is
paving our behavior by enabling or hampering goal-directed actions.
Aordances can be conceptually distinguished from
operationalizations of psychosocial task design characteristics such as
job autonomy, or task complexity (Humphrey etal., 2007), that usually
lack a direct reference to the “material” working environment. e
consideration of aordances should therefore provide additional
information and concrete conclusions about specic starting points
for task design, which are oen still missing in current I-O
psychological research (Parker et al., 2017b). We assume that
aordances of AMaect task design and postulate that ART can
explain these eects. In the following two sections, wewill bring these
perspectives together.
2 Definition of algorithmic
management from the perspective of
action regulation and sociomaterial
system theory
From the synthesis of the theoretical perspectives introduced
above, it can bederived that the main aordances of AMsystems from
the perspective of the acting worker are goal-setting, action-planning,
scheduling, monitoring, and feedback. ese aordances correspond
with the functions of AMreported elsewhere (Benlian etal., 2022;
Gagné etal., 2022; Parent-Rocheleau and Parker, 2022).
From the action theoretical concept of complete tasks one can
further conclude that the more of these functions an AMsystem
incorporates, the more incomplete a task is from the perspective of
workers. It is therefore likely that AMhas an impact on job resources
like job autonomy and job control as well as on the extent of learning-
promoting and motivating challenge demands like job complexity as
well as on stressful and demotivating hindrance demands like
regulation uncertainties.
Moreover, in addition to the individual functions and aordances
of AMmentioned above, weassume that the “completeness” of AMis
an aordance in itself with an own quality. Because the single
functions of AMshould, according to ARTs concept of complete tasks,
have a close inherent logical relationship, webelieve that they also
jointly aect work design and mental well-being, and therefore must
also bestudied together and not solely separately.
Consequently, wesuggest the following working denition of AM:
An algorithmic management system is a sociomaterial system which
aords the work behavior of employees through rule-based computed
(algorithmic) goal-setting, action-planning, scheduling, monitoring, and
feedback, without explicit involvement of human managers or other
social agents at work. e amount and extent of algorithmic control of
these functions indicate the “completeness of AM.”
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 06 frontiersin.org
In the following, wewant to derive our propositions on the
relationship between the completeness of AMand the quality of work
design based on ART and examine them with the available literature.
To illustrate the current knowledge on the effects of AMon
the quality of work and action regulation, use cases of work
which is already managed by algorithms are presented. The
description of these use cases follows the sequence of action steps
according to ART. This will shed light on the perspective of
workers under AMworking conditions. The effects of AMon the
quality of task design and thus on workers mental well-being are
discussed after the use cases to (1) highlight the key differences
between AMwork quality and traditionally managed work and
(2) to show the similarities of AMwork quality in different
work contexts.
3 Use cases of algorithmic
management
3.1 Algorithmic management in the
ridesharing industry
e advent of algorithmic management has ushered in a new era
in the so-called gig economy. e gig economy is a part of the labor
market in which workers engage on a job-by-job basis (Manyika etal.,
2016). For example, in contrast to traditional taxi services, companies
like Uber and Ly do not employ drivers but provide a platform which
enables the matching of self-employed drivers and customers. is
puts the drivers into a freelance status, reducing the employer
obligations (e.g., occupational health and safety) and shiing a great
part of the business risk from companies to workers (Rosenblat and
Stark, 2016). To organize this complex network of drivers, algorithmic
systems are used (Lee etal., 2015).
Once workers are registered in the Uber app, they are oering
transportation of other persons like traditional taxis would do. e
interaction between drivers and the algorithmic management system
begins with matching customer and driver, where algorithms dene
targets based on a myriad of variables (Rosenblat and Stark, 2016).
Previous performance, customer ratings, and geographical
considerations all contribute to a dynamic goal-setting process. Once
a customer requests a transportation, the algorithm chooses the
closest driver in that area. e driver cannot inuence whom heor she
should transport or where the journey should go to. In fact, almost no
information about the customer is shared with the driver before the
ride is accepted. e only way to “decline” a ride is to wait out the 15 s
time window in which the ride needs to beaccepted (Cropanzano
etal., 2023; Lee etal., 2015).
As drivers embark on their journeys, algorithms function as
co-pilots, planning actions through a continuous exchange of
information. Real-time variables such as passenger demand, trac
conditions, and driver proximity are considered. e algorithmic
system plans routes, selects rides, and adapts to unforeseen
circumstances. Once a driver picks up a customer, the app provides
the routing to the target destination. In addition, the algorithm already
starts nding potential new clients for the driver once the current ride
is over. In contrast to traditional taxi drivers, ridesharing drivers do
not need to know the area they are working in. Even if they knew any
shortcuts, the app would provide the directions to take (Wood, 2021).
To ensure eective scheduling, algorithms anticipate and
orchestrate driver movements. Leveraging predictive analytics, these
systems forecast demand, dynamically allocating drivers to specic
locations. Drivers, guided by the anticipatory algorithms, become
integral components of a “synchronized dance,” strategically
positioned in high-demand areas to meet the ebb and ow of the
ridesharing ecosystem (Rosenblat and Stark, 2016).
e driver’s performance is constantly monitored. Metrics such
as completion rates and customer ratings are observed in real-
time. Deviations from performance standards trigger immediate
adjustments, shaping the algorithm’s decision-making process.
is interactive monitoring fosters a dependent relationship, as
drivers strive to align their actions with the algorithmic
expectations to maximize eciency and earnings (Rosenblat and
Stark, 2016).
Feedback to drivers is provided by algorithms through real-time
evaluations and performance dashboards. Passengers provide
immediate feedback, shaping driver behavior. Simultaneously,
performance metrics and earnings insights empower drivers to adapt
their strategies (Rosenblat and Stark, 2016).
3.2 Algorithmic management in the food
delivery industry
Algorithmic management has also revolutionized the food
delivery industry, introducing intelligent systems that dynamically
guide the actions of delivery drivers (Ivanova etal., 2018). In most
food delivery companies the work starts with logging into the
respective app. Once online, workers receive requests for delivery and
the corresponding restaurant address (Veen etal., 2020). In some
apps workers have approximately 10 s to accept a request, in some
apps the default option is acceptance with 10 s to decline (Veen
etal., 2020).
Once the request is accepted, the work follows a strict step by step
process managed by the app. Aer picking up the order at the
restaurant the worker needs to conrm the completeness of the order.
Only aer this step the delivery address is provided (Ivanova etal.,
2018). e app proposes the route to the client. Although the rider is
able to choose another route, the app will oen nudge the worker with
notications in case the chosen route seems to beslower (Ivanova
etal., 2018).
Real-time demand patterns, geographic variations, and driver
availability inform the dynamic scheduling process. Drivers are in
theory able to choose their shis according to their preferences and
availability through the platform interface. In practice however,
workers are oen organized in badges depending on their rating by
the app. e highest rating badge can choose their work schedule for
the next week rst, then the second badge and so forth (Ivanova etal.,
2018). at way only workers in full compliance with the apps rating
criteria have some level of control about their working schedule
(Griesbach etal., 2019).
Continuous monitoring is a hallmark of algorithmic management,
with real-time data informing performance assessments. Workers are
being monitored via GPS and customer rating are gathered (Ivanova
etal., 2018).
Automated feedback systems provide drivers with real-time
feedback. Although this information is helpful for the worker, it is not
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 07 frontiersin.org
directly obtained by the worker through interaction with the customer
or restaurant but only received through the app with no means to ask
for clarication if needed.
3.3 Algorithmic management in
warehouses
As an example of more traditional work arrangements that has
been fundamentally changed by algorithmic management, warehouse
logistics show a substantial integration of AMinto the work process.
Some in fact argue that human workers are as much integrated into a
system as a robot would be(Delfanti, 2019). Although there are still
human managers present in warehouses, their interaction with
workers is reduced to a minimum; and remaining managers oen rely
on the algorithmic management system to receive information about
the workers productivity (Delfanti, 2019).
All workers receive a barcode scanner, oen referred to as “gun”
at the beginning of their shi as their main instrument of work. is
barcode scanner is the equivalent of the app in the platform economy
as it mediates between workers and management, setting goals,
assigning, and planning tasks, monitoring the work completion and
providing feedback (Delfanti, 2019; Gent, 2018).
e work itself is organized into four core processes: receive
incoming wares, stow incoming wares into the storage system, pickup
outgoing wares out of the storage system and pack outgoing wares for
shipment. At the “receive” workstation, workers unpack pallets and
scan the barcodes of the wares and pack them on a tote. e tote
travels to the “stow” area, where workers group the wares into bins and
carry those to the automated storage system. “Pick”-workers retrieve
items from the storage system and bring them to sorting workstations.
In the “pack” area, workers receive, pack and label wares for shipment.
Work starts once workers pickup their barcode scanner and scan their
work batch at the beginning of each shi (Delfanti, 2019).
Goals are set and communicated by the barcode scanner to the
worker. Once an order has been placed, the system calculates which
worker should retrieve the item from the storage system depending
on the locations of the item and the worker (Delfanti, 2019). e goals
are based to the data previously collected from workers (Gent, 2018).
In case of batches of orders, the barcode scanner provides not the
complete overview, but only the next item to pick up to the worker.
at way the sequence of item retrieval is “known” only to the
AMsystem and workers cannot look ahead past their current activity
(Delfanti, 2019). At the packing workplace, the scanner tells the packer
which box size to use (Gent, 2018).
e time when a worker needs to receive, stow, pick up or pack an
item is determined by customer orders and communicated by the
barcode scanner to the worker. ere is no option for the worker to
change the time or sequence of the actions required to fulll the task
(Delfanti, 2019; Gent, 2018).
All information about the work is gathered and communicated by
the barcode scanner. It constantly monitors the location of the worker
and calculates the speed workers need to fulll their tasks, such as
items packed per hour (Delfanti, 2019; Gent, 2018).
e information about the workers performance is provided to
the worker and to the warehouse management. at way performance
feedback is available to workers to know if they meet the set targets.
e barcode scanner does not provide any information to improve in
case these targets are not met, however. It is not even communicated
what data the performance evaluation, communicated in percent
target achievement, is based on. Even warehouse managers have no
oversight therefore fully dependent on that system (Delfanti, 2019;
Gent, 2018). e inherent necessity of chaotic warehousing not only
aects workers and managers but also leads to a complete dependence
of the company/organization on the system as all overview and control
of locations and stock is transferred to the technical system.
4 Propositions on the eects of
algorithmic management on
human-centered task design
From an ART perspective, one can conclude out of these three use
cases that AM is associated with sequentially and hierarchically
incomplete work tasks, which in turn might have an impact on task
design and indirectly also on the stress experience and motivation
of workers:
1 Under AMworkers usually cannot autonomously decide about
their task goals. e setting of goals is done by the algorithm
and goals are merely communicated to the worker via technical
devices. Within exible gig work, workers are additionally
oen motivated by nudges or additional extrinsic incentives to
pursue these goals. For example, surge pricing is used to
increase ridesharing driver availability in certain areas or to a
certain time of the day with many customers (Rosenblat and
Stark, 2016). Moreover, task goals are strongly connected to
real time performance data and customer feedback which can
result in dynamic changes of goals and limited predictability
(Griesbach etal., 2019; Lee etal., 2015; Rosenblat and Stark,
2016). Particularly, the latter distinguishes AMfrom traditional
eciency-oriented forms of work organization, such as
externally determined but predictable assembly line work, and
it can be assumed that this severely restricts the workers’
control over the entire task process.
2 Also the planning of required actions is mostly done by the
algorithm (e.g., negotiating prices, identifying the fastest route
to the destination, selecting the next product to pick from a
shelf) (Gent, 2018; Wood, 2021).
3 e same applies to scheduling. For example, the Uber
algorithm is said to beoptimized for matching drivers and
customers to provide the drivers with as many customers as
possible and the customers with as little waiting time as
possible. is is accomplished by using real time data provided
by drivers and customers. Hereby, the amount of processed
data considered by the algorithm is larger than humans can
process. At the moment, in most cases, this data is not made
available to the drivers, nor is it processed in a way that would
becompletely understandable for the drivers. us the drivers
can oen only choose when to start working and when to stop
(Lee etal., 2015). e same is true for the food delivery services
(Ivanova et al., 2018). e chaotic storage systems in
warehouses have a similar eect. Workers can hardly know
where a certain product is stored, so only the algorithm can
determine where and when to pick up which product to ensure
an ecient workow (Delfanti, 2019). us, action planning
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 08 frontiersin.org
and scheduling by AMmay not only reduce the challenging
intellectual regulation requirements for workers. It might also
limit the transparency of their work task and conditions for
task execution, which are necessary to get an adequate mental
model or, in action regulation terms, an operative image system
of the work task. Both are a prerequisite for goal setting and
action planning and thus individual control over the work
process (Hacker, 2003).
4 In a similar vein is the monitoring of actions performed mainly
by technical devices (Griesbach etal., 2019; Rosenblat and
Stark, 2016). Monitoring in AMtends to bemainly used to
evaluate the performance of workers, which seems to have an
additional eect on workers focus on their actions. As for
example the rating by customers is an important factor for
future earnings, Uber drivers tend to optimize their relationship
with the customers at their own expense to receive better
ratings (e.g., providing free drinks) (Rosenblat and Stark,
2016). us, workers under AMtend to “work for data,” which
means that they focus their action regulation more on the
aspects of a task that are recorded by the monitoring system
and less on the things that seem relevant, meaningful and
motivating to them (Moore and Robinson, 2016). Moreover,
monitoring might increase the quantitative demands of
workers. For example, warehouses worker, where the
performance monitoring is more focused on the number of
simple tasks completed in a given timeframe, workers are
forced to increase their work speed (Gent, 2018).
5 Lastly, also the feedback is provided by a technical device.
AMusually provides workers with timely and understandable
feedback, e.g., whether a task has been nished or how many tasks
have been completed. is information may strengthen the feeling
of workers’ mastery and their experience of competence (Gagné
et al., 2022). However, certain aspects of feedback might not
always beacceptable to the worker (Delfanti, 2019). is again
might bedue to the perceived lack of legitimacy of the feedback
and the missing transparency of the feedback criteria, because
studies report that some sources of information are not completely
reliable (e.g., unjustied angry customer) (Ivanova etal., 2018;
Rosenblat and Stark, 2016), or because it is oen not transparent
which data feeds the performance evaluation of performance
(Delfanti, 2019). From an action theoretical perspective, on the
one hand this lack of legitimacy and transparency of feedback can
impair an important source of learning at work, as unaccepted
feedback might beless likely integrated to improve one’s action
regulation and on the other hand the action is directed toward less
accepted and therefore less intrinsically motivating goals
(Christensen-Salem etal., 2018).
e three use cases indicate that AMas it is predominantly used
at present might have similar eects on the completeness of work
tasks, independent on the individual work settings. From this, the
following propositions can bederived regarding the eects of AMon
job resources and job demands, which are important for the regulation
of employees’ actions and thus also for their stress experience and
motivation. Our propositions are summarized in Figure1.
Overall we assume that more complete AM systems are related to
reduced job resources (Parent-Rocheleau and Parker, 2022): As wesaw in
the use cases, across work settings AM predenes task goals and
considerably limits the possibility for workers to inuence the goal.
Whereas in traditional managed work no constant contact between
manager and worker necessitates some form of individual task planning,
the close coupling of task assignment and monitoring employs strict rules
which hinders autonomy (De Cremer, 2020; Moore and Hayes, 2017).
Algorithms are designed in a such a way that make it dicult to intervene
in the work process, as human intervention can interfere with the
algorithm. Neither the task assignment nor its scheduling is to beeasily
negotiated or declined. Moreover, especially in the platform economy,
workers are oen sanctioned when they decline tasks assignment too oen
(Goods etal., 2019; Gregory, 2021; Griesbach etal., 2019; Lee etal., 2015;
Rosenblat and Stark, 2016). is means that AMpresumably not only
makes it more dicult for workers to make own decisions, but
autonomous decisions are also sanctioned in some cases. Finally, as
AMsystems rely on the quantication of work, workers might focus only
on the aspects of work that are reected in the AM’s quantication of
performance. is so called “working for data” has been reported to
berelated to reduced autonomy (Gal etal., 2020; Schaeitle etal., 2020).
us, more complete AMshould berelated to lower job autonomy as the
central resource for action regulation according to ART (Frese and
Zapf, 1994).
Proposition 1: More complete AM is associated with lower
job autonomy.
We can derive from the above, that AM systems seem to
defragment complete tasks. is reduction in task completeness seems
to be implied by the need to quantify job demands to enable
AMsystems to function. Once job demands are fully quantied by an
AMsystem, the optimization and expansion of the system requires
that tasks to berepeated as oen as possible in the same way. us, in
logistic warehouses the worker might beonly involved in in a small
section of the complete task, like taking a product from a shelf or
packing it into a box which is then transported to the next step in the
delivery process by a conveyer belt. us, more complete AMshould
berelated to lower task identity.
Proposition 2: More complete AM is associated with lower
task identity.
Due to the large amount of data available, AMsystems have the
potential to provide valuable feedback to workers to learn and develop.
Unfortunately, research on AMsystems implies that this information
is rarely shared. Whereas Uber drivers get tips how to improve
customer ratings (Rosenblat, 2018; Rosenblat and Stark, 2016),
Amazon warehouse workers oen not even know on what data their
target achievement is based on Delfanti (2019). Moreover, studies
report a lack of legitimacy and transparency of feedback criteria, and
unreliability of the information source (Ivanova etal., 2018; Rosenblat
and Stark, 2016). us, although more complete AMshould usually
provide a large quantity of feedback it should bealso related to lower
feedback quality.
Proposition 3: More complete AM is associated with lower
feedback quality.
e constant integration of huge amounts of data by AMthat is
used to adapt action plans and schedules in real time, makes it
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 09 frontiersin.org
impossible for the worker to oversee the complete details of a given
task before starting to work on it. For example, in the ride sharing
business, drivers are only presented with the very next potential
passenger pickup address without any details where the customer
wants to betaken (Rosenblat and Stark, 2016). e same is true for
food delivery services, where a rider needs to accept a task without
knowing the delivery address (Griesbach etal., 2019). It is in fact
impossible for workers (and managers) to grasp the complete
workow as required information is locked in the AMsystem (e.g.,
chaotic storage) (Gent, 2018). us, more complete AM should
be related to lower transparency and predictability of work tasks
(Delfanti, 2019; Rosenblat, 2018).
Proposition 4: More complete AM is associated with
lower transparency.
Proposition 5: More complete AM is associated with
lower predictability.
We further assume that more complete AMsystems are related to
reduced challenge demands: According to ART learning is facilitated by
action. With sequentially and hierarchically complete actions workers
have the most potential to learn and develop during work (Zacher,
2017). is learning is enabled by both, errors and accomplishments,
which the worker integrates via the monitoring and feedback of the
action (Frese and Zapf, 1994). In algorithmically managed jobs,
however, the work is highly standardized. As already stated above, tasks
are broken down into quantiable steps and oen the steps related to
one single task are distributed between dierent workers. Tasks are
simplied to a point where workers do not have the opportunity to
solve problems by their own (Parent-Rocheleau and Parker, 2022), like
being told which box size to use to pack an item or how much tape is
required to seal that box (Delfanti, 2019). In addition, the tasks do not
vary in complexity. ey are merely repetitions of the previous tasks
with a dierent passenger, dierent food order or dierent item to pack.
Proposition 6: More complete AM is associated with lower
task complexity.
Proposition 7: More complete AM is associated with lower
problem solving.
We nally conclude that more complete AMhas mixed eects on
hindrance demands or regulation problems: In traditionally managed
jobs, retrieving required information or getting physically to a desired
location or object might pose a challenge. With work being very
structured by AM systems with step-by-step guidance, the risk
decreases of not having the required information or access rights for
the current task. And as there is no choice in which tasks to do next,
there is no need to obtain and process such information to structure
one’s work (Reyes, 2018). e same seems to betrue for the risk of
disruptions. In traditionally managed work settings, the unplanned
interaction with managers or fellow workers might interrupt the ow
of work. With AM workers usually working without any such
interactions, a circumstance that represents, however, a work design
problem of its own, the main obstacle in an AMworkplace would
bethe malfunction of the AMsystem.
Proposition 8: More complete AM is associated with lower
regulation obstacles (i.e., interruptions).
FIGURE1
Propositions of associations between completeness of AMand task design. P = Proposition. Black circles indicate detrimental eects on task design
according to action regulation theory, as well as on motivation and stress. White circles indicate desirable eects. Dashed lines indicate eects that are
not the direct focus of this article but can bederived from the existing literature.
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 10 frontiersin.org
Due to the simplication of work tasks, tasks goals and roles of
workers are usually very clearly dened, and role ambiguity should
therefore bequite low. Moreover, also regulation uncertainty resulting
from the dierent role expectations of colleagues should below as
workers only very rarely interact at all (Parent-Rocheleau and Parker,
2022). And although the workload is reported to beoen high in
AMwork settings, there is usually a quite low uncertainty on how to
achieve a certain goal (Benlian etal., 2022; Cram and Wiener, 2020).
Proposition 9: More complete AM is associated with lower
regulation uncertainty (i.e., ambiguity or lack of clarity in one’s
task goals, or action plans).
Overtaxing regulations on the other hand seem to beincreased in
an AMwork environment. AMis reported to increase work demands
in terms of workload due to its constant optimization of work ecacy
(Gregory, 2021; Reyes, 2018). As performance targets are based on
data of past workers AMsets an environment of ever-increasing
standards of productivity (Guendelsberger, 2019). For example, in
warehouses the daily performance targets are displayed by the scanner
and enforced by supervisors (Delfanti, 2019). Due to its reliance on
data, AMmonitors as many aspects of the task accomplishment as
possible and uses this information to control workers behavior with
nudging. On several platforms, the algorithmic system modies pay
rates in real time according to demand, thus spurring workers to
“chase” lucrative hours, and, consequently, to work long and irregular
hours, at times most other people enjoy their free time (Cram and
Wiener, 2020; Oppegaard, 2021; Wood et al., 2019). Feedback
provided by AMalso changes rapidly with the constant updating of
algorithms (Stark and Pais, 2020).
Proposition 10: More complete AMis associated with higher
overtaxing regulation (i.e., time demands, more
simultaneous tasks).
5 Discussion
e aim of this paper was to complement recent rst attempts to
establish our conceptual understanding of the psychological impact of
algorithmic management (AM) (Gagné etal., 2022; Parent-Rocheleau
and Parker, 2022). In line with sociomaterial understandings of
technology (Leonardi, 2012) and action regulation theory (ART) (Hacker,
2003; Zacher and Frese, 2018), wepropose that the perspective of human
action regulation is particularly suited to identify psychologically relevant
mechanisms through which AMexerts an eect on workers. From this
perspective, we suggest the concept of “completeness” of AM that
describes the extent to which AMdetermines the action steps of a worker,
from goals setting to giving feedback. We assume that the degree of
“completeness” of AMhas a profound eect on the humane design of
work tasks, and through task design indirectly also on the motivation and
stress experience of workers.
Against this background, wepropose that the application of
AMcan have both negative and positive eects on the human design
of work tasks.
On the negative side weexpect algorithmically managed tasks to
show lower job resources in terms of job autonomy, task identity and
feedback quality. e potential to learn and grow is likely to below as
such tasks seem to provide little complexity and opportunities for
problem solving. In terms of overtaxing regulation, weexpect higher
eects due to lower transparency and predictability as well as
increased time demands and work intensication. In line with the
Challenge-Hindrance Stressor Framework (Podsako etal., 2023),
weexpect some challenge stressors (e.g., complexity) to belimited and
some hindrance stressors (e.g., resource inadequacies) to be high
under conditions of AMand thus potentially decreasing motivation
and increasing stress.
On the positive side, weexpect algorithmically managed tasks to
show few regulation obstacles and low regulation uncertainty as
algorithmically managed tasks follow strict rules and have clearly
dened roles and responsibilities. Wealso expect some challenge
stressors (e.g., time pressure) to behigh and some hindrance stressors
(e.g., role and interpersonal conict) to below, having a positive eect
on motivation but at the same time also contribute to higher stress.
Further positive aspects are assumed to bean increase in exibility
and performance opportunities (Benlian etal., 2022), as well as a
positive perception of procedural justice (Bujold and Parent-
Rocheleau, 2024).
us, in accordance with the Job Demands-Resources Model
(Bakker and Demerouti, 2017) weexpect a mixed picture of AMon
job demands and job resources. As complexity decreases, weexpect
the tasks to become easier to fulll while providing less potential for
learning. In the same vein, regulation obstacles are expected to below
suggesting a reduction of demotivating and stressful eects. On the
other hand, with lower autonomy and less information resources,
AMtasks provide less job resources suggesting lower motivational and
stress reducing eects as well.
Our paper complements recent psychological conceptualizations
of AMwithin frameworks of work design theory (Parent-Rocheleau
and Parker, 2022) and self-determination theory (Gagné etal., 2022).
By linking sociomaterial system theory that considers technologies as
amalgamation of material and human goals (e.g., Landers and Marin,
2021) with ART (Hacker, 2003; Zacher and Frese, 2018) weshow
potential underlying cognitive mechanisms in terms of action
regulation through which the functions of AM might aect
psychosocial important aspects of task design, and with that also
workers’ well-being.
Our approach also complements the level of automation (LOA)
research (Endsley, 2018) in a meaningful way: From a conceptual
perspective, like ours, the LOA approach deals with the psychological
eects of dierent degrees of automation of action steps through the
application of technologies. However, LOA focuses specically on
immediate performance-relevant cognitive eects, such as possible
impairments of situational awareness, which, among other things,
make it more dicult for workers to intervene in critical situations
(i.e., out-of-the-loop performance problems). In addition to this,
wetake a broader perspective by discussing the eects of automation
on the design aspects of the work tasks and entire work activity. is
perspective allows drawing additional conclusions about long-term
motivational and health-related eects of automation which are not
in the focus of LOA.
From a more practical perspective, AMcan beconsidered as a
specic application of automation that so far is not as much a focus of
LOA research. LOA mainly refers to applications were the worker
controls the execution of actions by the machine instead of doing the
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 11 frontiersin.org
work him- or herself (Endsley, 2018; Vagia etal., 2016). In contrast, in
AMmachines do not take over action execution from workers to ease
or speed up manual work, elevating the workers cognitive eorts
(Vagia etal., 2016). Instead, in AMit almost appears to bethe opposite
split of responsibilities within the human-machine-interaction:
workers need to do the manual work machines cannot do yet, losing
the cognitive challenging aspects of the task to the AMsystem.
5.1 Limitations of our perspective and
recommendations for future research
Our proposed action theoretical approach ignores some
perspectives that are also important for the psychological
understanding of AM:
Action regulation theory mainly focusses on the individual
cognitive processes of actions. It focusses less on the social interactions
during these actions as regulation requirements or resources (Zacher
and Frese, 2018). According to self-determination theory, social
relationships can act as important personal resources leading to the
experience of joy and intrinsic motivation as central characteristics of
good mental well-being (Deci etal., 2017). Good social relations at
work, like social support by managers and colleagues, are one of the
main job resources for mental well-being at the workplace (Montano
etal., 2017). As the described use cases of AMwork environments
show mostly solitary tasks at work, the lack of social relationships at
work might pose an additional eect on worker well-being
(Gagné et al., 2022) that we did not integrate directly into
our conceptualization.
Moreover, role-based identity theory (Ramarajan, 2014) suggests that
there are work tasks that are perceived as more central to one’s work role
than others. Such work tasks are called direct tasks (Gabriel etal., 2011)
or core tasks (Semmer etal., 2015). Research indicates that satisfaction of
workers with the accomplishment of such core tasks is more strongly
related with well-being than satisfaction with the accomplishment of
non-core tasks (Gabriel etal., 2011). It seems plausible that the assumed
shi of organizational control from workers to technology by the
introduction of AM(Kellogg etal., 2020; Möhlmann etal., 2021) might
fundamentally change core aspects of work tasks and with that also the
workers’ roles. us, negative eects of AMcan beexpected particularly
when accepted and internalized work roles are transformed and the work-
related identity of employees changes fundamentally (Semmer etal.,
2015). is can be particularly expected when AM is introduced in
“traditional” jobs, where understanding of one’s own work task has
developed across organizations over longer periods of time, in contrast to
the new “gig economy,” where such traditional work roles do not yet exist
and AMmay therefore bemore accepted (Baiocco etal., 2022). Although
this example is not directly an implementation of AM, in medicine
algorithms using articial intelligence technologies like machine learning
have already been proven to beas eective as humans when it comes to
diagnosing certain medical conditions. is benet for patients comes
with severe changes for the role of doctors (Loh, 2018). Similar changes
occur in human resources management as the activities of sourcing and
assessment are more and more handled by algorithms instead of
humans (Li et al., 2021; Loh, 2018). One can conclude, that an
investigation of AM eects across dierent professional contexts is
needed. In particular, the distinction between jobs in the new “gig
economy” and traditional jobs seems appropriate here.
e current body of literature about AM is mainly based on
qualitative research (Benlian etal., 2022; Gagné etal., 2022; Parent-
Rocheleau and Parker, 2022). Although these studies have provided
us with the valuable information this review is built on, quantitative
research is required to enable more robust testing of the current
hypotheses about the impact of AMon work and task design as well
as workers’ well-being. e recently developed algorithmic
management questionnaire (Parent-Rocheleau etal., 2023) is one
existing instrument starting to beused to quantify the impact of
AM(Bujold and Parent-Rocheleau, 2024).
Furthermore, the currently available studies are mostly focusing
on very pure (or extreme) applications of AM. Most data were
obtained from the platform and gig economy, namely ride-sharing,
food delivery or clickworkers. In these work environments, AMoen
fully replaces human managers and workers are mostly performing
tasks requiring low formal education. To gather a more general
understanding about AM, weneed to see its impact in more traditional
work settings as already outlined above.
Moreover, research is needed in settings were AMand human
managers coexist: Is there a possible combination of AMand human
management which maintains the performance benets from
AMwithout the supposed negative eects on well-being? One might
think of augmented leadership, where machine learning algorithms
process employee data to provide leadership suggestions to human
managers. Today wealready see that more transactional management
tasks are carried out by algorithms. It is therefore very probable that
at some point empathic and motivational leadership tasks will
be supported or even taken over by AM (Quaquebeke and
Gerpott, 2023).
5.2 Conclusion
ere is wide agreement across dierent academic disciplines
that AMwill continue to change the way wework (Benlian etal.,
2022). e main question is therefore not whether AMis good or
bad for workers, but rather how work under AMcan bedesigned to
behumane. Wethus want to stress the importance of work and
organizational psychological knowledge in the development of the
next generation of AMsystems. Today, AMsystems are designed
and implemented by companies pursuing economic goals. is
approach needs to becomplemented by scientically validated
design choices reecting the needs of humans (Parker and Grote,
2022). While some of the negative eects described above should
be attributed to inherent and hardly changeable functions of
AM systems (such as the restrictions on job autonomy, or the
reduction of task identity), other negative eects of AMprobably
derive not from the technology itself, but from the design choices
made by human developers. Managers and developers of
AMsystems should beinformed about potential consequences of
design choices for worker well-being. Our considerations suggest
that more complete AMsystems may beassociated with detrimental
task design and therefore with risks to the well-being of workers.
e technical possibilities of AMsystems should therefore not
befully exploited. Instead, human capabilities und needs should
bethe starting point for the design of AMsystems. From an action
regulation perspective, AMsystems should beused to specically
increase regulation opportunities for workers and eliminate
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 12 frontiersin.org
regulation problems. AMsystems could provide workers with
information they need to make informed and autonomous decisions
for themselves and not dictate the decisions to the workers. For
example, it is possible to give an Uber driver more information
about the location of a potential passenger and not to sanction
refusals of ride requests. In this way, AMsystems can also minimize
obstacles and uncertainties at work. Both could increase the
workers’ sense of autonomy and control over their work, and with
that also their motivation and mental well-being. Such a human-
centered design of AMsystems makes it necessary that workers are
continuously involved in the development of AMsystems, including
decisions about the data to beprocessed and the corresponding
algorithms. From the perspective taken here, both example
measures should contribute to higher motivation and lower stress
levels among workers. e common goal of industry and academia
needs to bethe improvement of AMas socio-technical systems
consisting of both, algorithms and humans, making use of and
reecting their capabilities and needs. In sum, against the
background of the above-described state of psychological research
on AM, future research should hence advance and systematize the
psychological understanding of AM and its functions across
dierent jobs and industries. e here proposed action-theoretical
perspective might be a starting point for developing further
hypotheses and explaining phenomena.
Data availability statement
e original contributions presented in the study are included in
the article/supplementary material, further inquiries can bedirected
to the corresponding author.
Author contributions
CR: Conceptualization, Investigation, Writing – original dra,
Writing – review & editing. BH: Conceptualization, Writing – review &
editing. TW: Conceptualization, Writing – review & editing. AM:
Conceptualization, Supervision, Writing – original dra, Writing –
review & editing.
Funding
e author(s) declare that nancial support was received for the
research, authorship, and/or publication of this article.
Weacknowledge support by the Open Access Publication Fund of the
University of Duisburg-Essen.
Conflict of interest
e authors declare that the research was conducted in the
absence of any commercial or nancial relationships that could
beconstrued as a potential conict of interest.
Publisher’s note
All claims expressed in this article are solely those of the authors
and do not necessarily represent those of their aliated organizations,
or those of the publisher, the editors and the reviewers. Any product
that may beevaluated in this article, or claim that may bemade by its
manufacturer, is not guaranteed or endorsed by the publisher.
References
Baethge, A., Rigotti, T., and Roe, R. A. (2015). Just more of the same, or dierent? An
integrative theoretical framework for the study of cumulative interruptions at work. Eu r.
J. Work Organ. Psy. 24, 308–323. doi: 10.1080/1359432X.2014.897943
Baiocco, S., Fernández-Macías, E., Rani, U., and Pesole, A. (2022). e algorithmic
management of work and its implications in dierent contexts, JRC Working Papers Series on
Labour, Education and Technology, No. 2022/02, European Commission, Joint Research
Centre (JRC), Seville.
Bakewell, L., Vines, J., Vasileiou, K., Long, K. S., Atkinson, M., Rice, H., et al.
(2018). Everything We Do, Everything We Press: Data-Driven Remote Performance
Management in a Mobile Workplace. In Proceedings of the 2018 CHI
Conference on Human Factors in Computing Systems (CHI ‘18). Association for
Computing Machinery, New York, NY, USA, Paper. 371, 1–14. doi:
10.1145/3173574.3173945
Bakker, A. B., and Demerouti, E. (2017). Job demands-resources theory: Taking stock
and looking forward. J. Occup. Health Psychol. 22, 273–285. doi: 10.1037/ocp0000056
Bakker, A. B., Demerouti, E., and Sanz-Vergel, A. (2023). Job demands–resources
theory: ten years later. Annu. Rev. Organ. Psych. Organ. Behav. 10, 25–53. doi: 10.1146/
annurev-orgpsych-120920-053933
Benlian, A., Wiener, M., Cram, W. A., Krasnova, H., Maedche, A., Möhlmann, M., et al.
(2022). Algorithmic management. Bus. Inf. Syst. Eng. 64, 825–839. doi: 10.1007/
s12599-022-00764-w
Bujold, A., and Parent-Rocheleau, X. (2024). “e two faces of algorithmic management
in the gig economy” in 57th Hawaii International Conference on System Sciences.
Cascio, W. F., and Montealegre, R. (2016). How technology is changing work and
organizations. Annu. Rev. Organ. Psych. Organ. Behav. 3, 349–375. doi: 10.1146/annurev-
orgpsych-041015-062352
Christensen-Salem, A., Kinicki, A., Zhang, Z., and Walumbwa, F. O. (2018). Responses
to feedback: e role of acceptance, aect, and creative behavior. J. Leadersh. Organ.
Stud. 25, 416–429. doi: 10.1177/1548051818757691
Cram, W. A., and Wiener, M. (2020). Technology-mediated control: case examples
and research directions for the future of organizational control. Commun. Assoc. Inf.
Syst. 46, 70–91. doi: 10.17705/1CAIS.04604
Cropanzano, R., Keplinger, K., Lambert, B. K., Caza, B., and Ashford, S. J. (2023). e
organizational psychology of gig work: An integrative conceptual review. J. Appl. Psychol.
108, 492–519. doi: 10.1037/apl0001029
De Cremer, D. (2020). Leadership by Algorithm: Who Leads and Who Follows in the
AI Era?. USA: Harriman House Limited.
Deci, E. L., Olafsen, A. H., and Ryan, R. M. (2017). Self-determination theory in work
organizations: the state of a science. Annu. Rev. Organ. Psych. Organ. Behav. 4, 19–43.
doi: 10.1146/annurev-orgpsych-032516-113108
Delfanti, A. (2019). Machinic dispossession and augmented despotism: Digital work
in an Amazon warehouse. New Media Soc. 23, 39–55. doi: 10.1177/1461444819891613
Demerouti, E. (2020). Turn Digitalization and Automation to a Job Resource. Appl.
Psychol. Int. Rev., 1–6. doi: 10.1111/apps.12270
Endsley, M. R. (2018). Level of automation forms a key aspect of autonomy design. J.
Cogn. Eng. Decis. Mak. 12, 29–34. doi: 10.1177/1555343417723432
Frese, M., and Zapf, D. (1994). “Action as the core of work psychology: A German approach”
In H. C. Triandis, M. D. Dunnette & L. M. Hough (Eds.), Handbook of industrial and
organizational psychology, Volume 4 (pp. 271–340). Palo Alto: Consulting Psychologists Press.
Gabriel, A. S., Diefendor, J. M., and Erickson, R. J. (2011). e relations of daily task
accomplishment satisfaction with changes in aect: a multilevel study in nurses. J. Appl.
Psychol. 96, 1095–1104. doi: 10.1037/a0023937
Gagné, M., Parent-Rocheleau, X., Bujold, A., Gaudet, M.-C., and Lirio, P. (2022). How
algorithmic management inuences worker motivation: A self-determination theory
perspective. Can. Psychol. 63, 247–260. doi: 10.1037/cap0000324
Gal, U., Jensen, T. B., and Stein, M.-K. (2020). Breaking the vicious cycle of algorithmic
management: A virtue ethics approach to people analytics. Inf. Organ. 30:100301. doi:
10.1016/j.infoandorg.2020.100301
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 13 frontiersin.org
Gent, C. (2018). e politics of algorithmic management class: composition and
everyday struggle in distribution work. [PhD, University of Warwick]. Available at:
https://wrap.warwick.ac.uk/132957/.
Gibson, J. J. (1977). e theory of aordances. Perceiving, acting and knowing:
Towards an ecological psychology/Erlbaum.
Goods, C., Veen, A., and Barratt, T. (2019). “Is your gig any good?” Analysing job
quality in the Australian platform-based food-delivery sector. J. Ind. Relat. 61, 502–527.
doi: 10.1177/0022185618817069
Gregory, K. (2021). ‘My life is more valuable than this’: understanding risk among
on-demand food couriers in Edinburgh. Work Employ. Soc. 35, 316–331. doi:
10.1177/0950017020969593
Griesbach, K., Reich, A., Elliott-Negri, L., and Milkman, R. (2019). Algorithmic
control in platform food delivery work. Socius 5:2378023119870041. doi:
10.1177/2378023119870041
Guendelsberger, E. (2019). On the clock: What low-wage work did to me and how it
drives America insane. Hachette UK.
Hacker, W. (2003). Action Regulation eory: A practical tool for the design of
modern work processes? Eur. J. Work Organ. Psy. 12, 105–130. doi:
10.1080/13594320344000075
Hackman, J., and Oldham, G. R. (1976). Motivation through the design of work: Test
of a theory. Organ. Behav. Hum. Perform. 16, 250–279. doi: 10.1016/0030-5073(76)90016-7
Humphrey, S. E., Nahrgang, J. D., and Morgeson, F. P. (2007). Integrating motivational,
social, and contextual work design features: A meta-analytic summary and theoretical
extension of the work design literature [Meta Analysis]. J. Appl. Psychol. 92, 1332–1356.
doi: 10.1037/0021-9010.92.5.1332
Ivanova, M., Bronowicka, J., Kocher, E., and Degner, A. (2018). e App as a Boss?
Control and Autonomy in Application-Based Management. Arbeit | Grenze | Fluss -
Work in Progress interdisziplinärer Arbeitsforschung Nr. 2, Frankfurt (Oder): Viadrina.
doi: 10.11584/Arbeit-Grenze-Fluss.2
Karasek, R. A. (1979). Job demands, job decision latitude, and mental strain:
Implications for job redesign. Adm. Sci. Q. 24, 285–308. doi: 10.2307/2392498
Kellogg, K. C., Valentine, M. A., and Christin, A. (2020). Algorithms at work: the new
contested terrain of control. Acad. Manag. Ann. 14, 366–410. doi: 10.5465/
annals.2018.0174
Kinowska, H., and Sienkiewicz, Ł. J. (2022). Inuence of algorithmic management
practices on workplace well-being – evidence from European organisations. Inf. Technol.
People 36, 21–42. doi: 10.1108/itp-02-2022-0079
Lamers, L., Meijerink, J., Jansen, G., and Boon, M. (2022). A capability approach to
worker dignity under algorithmic management. Ethics Inf. Technol. 24:10. doi: 10.1007/
s10676-022-09637-y
Landers, R. N., and Marin, S. (2021). eory and technology in organizational psychology:
a review of technology integration paradigms and their eects on the validity of theory. Annu.
Rev. Organ. Psych. Organ. Behav. 8, 235–258. doi: 10.1146/annurev-orgpsych-012420-060843
Lee, M. K., Kusbit, D., Metsky, E., and Dabbish, L. (2015). Working with machines:
the impact of algorithmic and data-driven management on human workers. In:
Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing
Systems (CHI ‘15). (New York, NY, USA: ACM) 1603–1612. doi:
10.1145/2702123.2702548
Leonardi, P. M. (2012). Materiality, sociomateriality, and socio-technical systems: what
do these terms mean? how are they related? do weneed them? in Materiality and
organizing: Social interaction in a technological world. (Eds.) P. M. Leonardi, B. A. Nardi
and J. Kallinikos (Oxford University Press), 25–48.
LePine, J. A., Podsako, N. P., and Lepine, M. A. (2005). A meta-analytic test of the
challenge stressor-hindrance stressor framework: An explanation for inconsistent
relationships among stressors and performance. Acad. Manag. J. 48, 764–775. doi:
10.5465/amj.2005.18803921
Lesener, T., Gusy, B., and Wolter, C. (2019). e job demands-resources model: A
meta-analytic review of longitudinal studies. Work Stress 33, 76–103. doi:
10.1080/02678373.2018.1529065
Li, L., Lassiter, T., Oh, J., and Lee, M. K. (2021). Algorithmic Hiring in Practice:
Recruiter and HR Professional’s Perspectives on AI Use in Hiring. in Proceedings of the
2021 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’21), May 19–21, 2021,
Virtual Event, (New York, NY, USA: ACM). 11. doi: 10.1145/3461702.3462531
Locke, E. A., and Latham, G. P. (2016). New directions in goal-setting theory. Curr.
Dir. Psychol. Sci. 15, 265–268. doi: 10.1111/j.1467-8721.2006.00449.x
Loh, E. (2018). Medicine and the rise of the robots: a qualitative review of recent
advances of articial intelligence in health. BMJ Leader 2, 59–63. doi: 10.1136/
leader-2018-000071
Manyika, J., Lund, S., Bughin, J., Robinson, K., Mischke, J., and Mahajan, D. (2016).
Independent-Work-Choice-necessity-and-the-gig-economy. McKinsey Global Institute.
Mashar, M., Chawla, S., Chen, F., Lubwama, B., Patel, K., Kelshiker, M. A., et al. (2023).
Articial intelligence algorithms in health care: is the current food and drug
administration regulation sucient? JMIR AI 2:e42940. doi: 10.2196/42940
Möhlmann, M., Zalmanson, L., Henfridsson, O., and Gregory, R. W. (2021).
Algorithmic management of work on online labor platforms: when matching meets
control [Article]. MIS Q. 45, 1999–2022. doi: 10.25300/MISQ/2021/15333
Montano, D., Reeske, A., Franke, F., and Hümeier, J. (2017). Leadership, followers' mental
health and job performance in organizations: A comprehensive meta-analysis from an
occupational health perspective. J. Organ. Behav. 38, 327–350. doi: 10.1002/job.2124
Moore, S., and Hayes, L. J. B. (2017). Taking worker productivity to a new level?
Electronic Monitoring in homecare-the (re)production of unpaid labour. N. Technol.
Work. Employ. 32, 101–114. doi: 10.1111/ntwe.12087
Moore, P., and Robinson, A. (2016). e quantied self: What counts in the neoliberal
workplace. New Media Soc. 18, 2774–2792. doi: 10.1177/1461444815604328
Noponen, N., Feshchenko, P., Auvinen, T., Luoma-Aho, V., and Abrahamsson, P.
(2023). Taylorism on steroids or enabling autonomy? A systematic review of algorithmic
management. Manag. Rev. Q. 74, 1695–1721. doi: 10.1007/s11301-023-00345-5
Oppegaard, S. M. N. (2021). Regulating exibility: Uber’s platform as a technological
work arrangement. Nordic Journal of Working Life Studies, 11.
Orlikowski, W. J., and Scott, S. V. (2008). 10 Sociomateriality: challenging the
separation of technology, work and organization. Acad. Manag. Ann. 2, 433–474. doi:
10.5465/19416520802211644
Parent-Rocheleau, X., and Parker, S. K. (2022). Algorithms as work designers: How
algorithmic management inuences the design of jobs. Hum. Resour. Manag. Rev.
32:100838. doi: 10.1016/j.hrmr.2021.100838
Parent-Rocheleau, X., Parker, S. K., Bujold, A., and Gaudet, M.-C. (2023). Creation of
the algorithmic management questionnaire: A six‐phase scale development process.
Hum. Resour. Manag. 63, 25–44. doi: 10.1002/hrm.22185
Parker, S. K., and Grote, G. (2022). Automation, Algorithms, and Beyond: Why Work
Design Matters More an Ever in a Digital World. J. Appl. Psychol. 71, 1171–1204. doi:
10.1111/apps.12241
Parker, S. K., Morgeson, F. P., and Johns, G. (2017a). One hundred years of work
design research: Looking back and looking forward. J. Appl. Psychol. 102, 403–420. doi:
10.1037/apl0000106
Parker, S. K., Van den Broeck, A., and Holman, D. (2017b). Work design inuences:
a synthesis of multilevel factors that aect the design of jobs. Acad. Manag. Ann. 11,
267–308. doi: 10.5465/annals.2014.0054
Podsako, N. P., Freiburger, K. J., Podsako, P. M., and Rosen, C. C. (2023). Laying
the Foundation for the Challenge–Hindrance Stressor Framework 2.0. Annu. Rev.
Organ. Psych. Organ. Behav. 10, 165–199. doi: 10.1146/annurev-orgpsych-080422-052147
Quaquebeke, N. V., and Gerp ott, F. H. (2023). e now, new, and next of digital leadership:
how articial intelligence (AI) will take over and change leadership as weknow it. J. Leadersh.
Organ. Stud. 30:15480518231181731. doi: 10.1177/15480518231181731
Ramarajan, L. (2014). Past, present and future research on multiple identities: toward
an intrapersonal network approach. Acad. Manag. Ann. 8, 589–659. doi:
10.5465/19416520.2014.912379
Reyes, J. F. (2018). Hotel housekeeping on demand: Marriott cleaners say this app
makes their job harder. e Philadelphia Enquirer. https://www.inquirer.com/philly/
news/hotel-housekeepers-schedules-app-ma rriott-union-hotsos-20180702.html
Rosenblat, A. (2018). Uberland. How Algorithms Are Rewriting the Rules of Work. 1st Edn:
University of California Press Available at: http://www.jstor.org/stable/10.1525/j.ctv5cgbm3
Rosenblat, A., and Stark, L. (2016). Algorithmic labor and information asymmetries:
A case study of Uber’s drivers. Int. J. Commun. 10:27.
Sailer, M., Hense, J. U., Mayr, S. K., and Mandl, H. (2017). How gamication motivates:
An experimental study of the eects of specic game design elements on psychological
need satisfaction. Comput. Hum. Behav. 69, 371–380. doi: 10.1016/j.chb.2016.12.033
Sandberg, Å. (1993). Volvo human-centred work organization—the end of the road?
N. Technol. Work. Employ. 8, 83–87. doi: 10.1111/j.1468-005X.1993.tb00038.x
Schaeitle, S. D., Weibel, A., Ebert, I. L., Kasper, G., Schank, C., and Leicht-Deobald, U.
(2020). No stone le unturned? Towards a framework for the impact of datacation
technologies on organizational control. Acad. Manag. Discov. 6, 455–487. doi: 10.5465/
amd.2019.0002
Schmierl, K., Schneider, P., Struck, O., and Ganesch, F. (2022). Digitale Logistik:
Digitalisierungstechnik, Arbeitsbedingungen, Leistungspolitik und Mitbestimmung in
Transportlogistik und Kurier-, Express-und Paketdiensten. Düsseldorf: Study der Hans-
Böckler-Stiung.
Semmer, N. K., Jacobshagen, N., Meier, L. L., Elfering, A., Beehr, T. A., Kalin, W., et al.
(2015). Illegitimate tasks as a source of work stress. Work Stress. 29, 32–56. doi:
10.1080/02678373.2014.1003996
Sonnentag, S., Cheng, B. H., and Parker, S. L. (2022). Recover y from work: advancing
the eld toward the future. Annu. Rev. Organ. Psych. Organ. Behav. 9, 33–60. doi:
10.1146/annurev-orgpsych-012420-091355
Stark, D., and Pais, I. (2020). Algorithmic management in the platform economy.
Sociologica 14, 47–72. doi: 10.6092/issn.1971-8853/12221
eorell, T., Hammarström, A., Aronsson, G., Träskman Bendz, L., Grape, T., Hogstedt, C.,
et al. (2015). A systematic review including meta-analysis of work environment and depressive
symptoms. BMC Public Health 15:738. doi: 10.1186/s12889-015-1954-4
Väänänen, A., Koskinen, A., Joensuu, M., Kivimäki, M., Vahtera, J., Kouvonen, A., et al.
(2008). Lack of predictability at work and risk of acute myocardial infarction: An 18-year
prospective study of industrial employees. Am. J. Public Health 98, 2264–2271. doi: 10.2105/
AJPH.2007.122382
Röttgen et al. 10.3389/frai.2024.1441497
Frontiers in Artificial Intelligence 14 frontiersin.org
Vagia, M., Transeth, A. A., and Fjerdingen, S. A. (2016). A literature review on the levels of
automation during the years. What are the dierent taxonomies that have been proposed?
Appl. Ergon. 53, 190–202. doi: 10.1016/j.apergo.2015.09.013
Veen, A., Barratt, T., and Goods, C. (2020). Platform-capital’s ‘app-etite’for control: A
labour process analysis of food-delivery work in Australia. Work Employ. Soc. 34,
388–406. doi: 10.1177/0950017019836911
Wood, A. J. (2021). Algorithmic management consequences for work organisation and
working conditions. JRC Working Papers Series on Labour, Education and Technology.
Wood, A. J., Graham, M., Lehdonvirta, V., and Hjorth, I. (2019). Good gig, bad gig:
autonomy and algorithmic control in the global gig economy. Work Employ. Soc. 33,
56–75. doi: 10.1177/0950017018785616
Wu, D., and Huang, J. L. (2024). Gig work and gig workers: An integrative review and
agenda for future research. J Organ Behav, 45, 183–208.
Zacher, H. (2017). Action regulation theory. In Oxford research encyclopedia of
psychology. (Oxford: Oxford University Press).
Zacher, H., and Frese, M. (2018). Action regulation theory: Foundations, current
knowledge and future directions. The SAGE handbook of industrial, work and
organizational psychology, (London: Oxford), 2, 80–102.
Zammuto, R. F., Grith, T. L., Majchrzak, A., Dougherty, D. J., and Faraj, S. (2007).
Information technology and the changing fabric of organization. Organ. Sci. 18,
749–762. doi: 10.1287/orsc.1070.0307
