When Industry meets Trustworthy AI: A Systematic Review of

AI for Industry 5.0

EDUARDO VYHMEISTER∗ and GABRIEL G. CASTAÑE∗ , University College Cork, Ireland
arXiv:2403.03061v1 [cs.CY] 5 Mar 2024

Industry is at the forefront of adopting new technologies, and the process followed by the adoption has a
significant impact on the economy and society. In this work, we focus on analysing the current paradigm in
which industry evolves, making it more sustainable and Trustworthy. In Industry 5.0, Artificial Intelligence
(AI), among other technology enablers, is used to build services from a sustainable, human-centric and resilient
perspective. It is crucial to understand those aspects that can bring AI to industry, respecting Trustworthy
principles by collecting information to define how it is incorporated in the early stages, its impact, and the
trends observed in the field. In addition, to understand the challenges and gaps in the transition from Industry
4.0 to Industry 5.0, a general perspective on the industry’s readiness for new technologies is described. This
provides practitioners with novel opportunities to be explored in pursuit of the adoption of Trustworthy AI in
the sector.
CCS Concepts: • General and reference → Surveys and overviews; • Social and professional topics →
Government technology policy; • Applied computing → Industry and manufacturing; • Computing
methodologies → Artificial intelligence.
Additional Key Words and Phrases: Trustworthy AI, Industry 5.0, Artificial Intelligence, Technology Readiness
Level
ACM Reference Format:
Eduardo Vyhmeister and Gabriel G. Castañe. 2022. When Industry meets Trustworthy AI: A Systematic Review
of AI for Industry 5.0. ACM Comput. Surv. 1, 1 (March 2022), 34 pages. https://doi.org/XXXXXXX.XXXXXXX

1 INTRODUCTION
Industrial processes and factories are evolving to transition to digitalization with novel techniques
that automate data computations more efficiently – and adequately – to improve tasks. The fourth
Industrial Revolution or Industry 4.0 incorporated cyber-physical systems into factories within
production systems, warehousing, and logistics, easing the adoption of new digital assets and
establishing them within enterprise values [43]. In this transition process, the decrease of the
hardware costs, and the vast amounts of data to be processed impulsed novel techniques that
automates, secures, and flexibly adapt factories. Foremost among these are: AI to boost data
analysis and outcome predictions [36]; Blockchain to secure and transparent record and share
data [5]; Additive Manufacturing to rapidly prototype and customise the products [6]; Internet of
the Things (IoT), Edge and Big Data analyses to enable real-time data processing and analysis for
decision making [1].
∗ Both authors contributed equally to this research.

Authors’ address: Eduardo Vyhmeister, eduardo.vyhmeister@insight-centre.org; Gabriel G. Castañe, gabriel.

gonzalezcastane@ucc.ie, University College Cork, Western Gateway Building, University College Cork, Western
Road, Cork, Cork, Ireland, T12YN60.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee
provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and
the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored.
Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires
prior specific permission and/or a fee. Request permissions from permissions@acm.org.
© 2022 Association for Computing Machinery.
0360-0300/2022/3-ART $15.00
https://doi.org/XXXXXXX.XXXXXXX

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
2 Castañé and Vyhmeister

The industrial ecosystem demands a transition to embrace resilient sources that promote societal
and environmental wellness while driving economic growth. Although this can seem as utopian,
the European Commission has designed a strategic path that, effectively executed, will transforms
traditional factories into resilient providers of prosperity, thereby evolving production centres into
respectful components for environmental and societal well-being [35]. This path is a transformative
journey that defines the development from Industry 4.0 into a new industrial revolution, Industry
5.0 [14].
Between all enabling technologies with potential for growth from Industry 4.0 to Industry 5.0,
Artificial Intelligence (AI) is widely recognized as a key among them. The automation or semi-
automation of large-scale production processes and extensive customizations makes it invaluable
in this transition. However, the use of AI must be underpinned by a high degree of trustworthiness
in the use of AI to bootstrap a technological transition. Hence, the economic benefits became a
secondary dimension compared to the minimization of associated risks and their impacts on society
and workers in factories [27, 93, 95]. This makes the adoption of AI technology an important case
of study encompassing socio-technical and legal challenges that must be addressed to reach human
and machine cooperation in smart-working environments [11].
However, factories are struggling to adopt operations to the standards of Industry 4.0. There are
several challenges that hamper a successful integration of necessary technologies for transitioning
to this stage [32, 96]. In addition, the emergence of Industry 5.0 can be perceived as an extra layer
of complexity for manufacturers and factories to evolve in the digitalization process [44]. Despite
that some production centres are not ready for the adoption, early studies show that transitioning
to Industry 5.0 pose benefits for companies such as: a seamless adoption of their products in the
market by making them trustworthy and explainable, advantages in product assortments [60],
improvement in maintenance processes and cost reduction, and enhancement in product quality
[13].
From a legal perspective, governments are working on several AI regulations. For example, the
AI Act [24, 85] creates awareness among companies to promote the adoption of Industry 5.0. The
major challenge for industry is to identify the correct paths and challenges within the extensive
work provided by the research community on Trustworthy AI topics. The diverse literature on
each of these challenges – when/if identified – hampers their understanding of the necessary steps
to meet the necessary requirements.
The work here presented provides a comprehensive review of the literature on state-of-the-art
Trustworthy AI technologies used in industry and the opportunities for transitioning to Industry 5.0.
The analysis explores research trends and highlights critical aspects of AI adoption by the industrial
sector. While various studies have examined the status of Industry 4.0, including bibliometric
evaluations analyzing trends and gaps [3, 58, 67, 79], different research approaches and statuses
used in Industry 4.0 [4, 52], future trends and challenges [4, 32], and the impact of Industry 4.0
on business models [45], to the best of the authors’ knowledge, no articles exist that define the
state-of-the-art dimensions of Trustworthy AI comprehensively to be applied in Industry 4.0 and,
consequently, to reach Industry 5.0. The data for this study was collected through a systematic
literature review of peer-reviewed articles to analyze the trends.
The paper is organized as follows: Section 2 provides an overview of the main aspects related
to Industry 4.0 and Industry 5.0, along with the leading AI technologies applied in these contexts.
Section 3 describes the research methodology and materials used for the literature review, including
qualitative and quantitative results for each research question. Section 4 presents a summary of the
discussion and findings. Finally, Section 5 presents the study’s conclusions and offers a general
perspective on the observed trends.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 3

2 BACKGROUND
This section provides an overview of Industry 4.0 and Industry 5.0 to give the reader a comprehensive
understanding of the existing challenges. In addition, it also describes the topics where AI is applied
to the industry are also described.

2.1 Industry 4.0 and Industry 5.0

Over the last decade, Industry 4.0 focused on incorporating cyber-physical systems and digital
platforms into factories, integrating monitoring processes [127]. By harnessing real-time data,
factories are able to gather and analyze valuable information, thus providing valuable insights
for decision-making processes. This has led companies in two directions to enhance their produc-
tion capabilities: decision making features and decentralization of production processes from the
traditional production control systems [40].
In the scope of decision-making, Industry 4.0 explores the use of technological advancements
for training and decision-making using models that replicate real-world situations in controlled
and safe environments, such as Digital Twins and Augmented Reality acting to facilitate (enable)
enhanced working environments and experiences in industrial settings, integrating virtual and
real systems, both independently and collaboratively with human counterparts. As a consequence,
this approach enables a thorough examination of business objective decisions while prioritizing
risk mitigation, leading to the minimization of adverse economic consequences on factories and
products.
The decentralization has been gathered by sensors and actuators operating on different machines
through IoT, facilitating comprehensive connectivity with computing systems and resulting in the
generation of large volumes of streaming data, commonly known as Big Data [77]. The processing
of this data is carried out using IoT devices either locally or on the Cloud/Edge, which reduces
costs and enhances scalability through the utilization of virtual resources in the Cloud or at the
network Edge [127]. The increased computational power and reduction of costs, enhanced the
ability to process data efficiently, enabling the adoption of multiple technologies that support the
goals of Industry 4.0. However, to manage such vast amounts of data, novel methods for intelligent
acquisition, collection, and process are required [122]. Effective data management and processing
techniques contribute to improved scalability, security, and efficiency but also a reduction in the
required resources. This is the case for Additive Manufacturing that streamlines the assembly
process by reducing the number of critical components required [62]. Although AI is not officially
considered a fundamental pillar of Industry 4.0, novel techniques supported by AI have a significant
role in driving its advancement. Numerous companies have developed innovative intelligent systems
enabling a certain level of process automation.
By utilizing AI, the information from various systems can be efficiently gathered and processed
at high speed, enabling the execution of tasks, such as Fault Prediction and Action Selection [7].
AI enables intelligent decision-making and predictive capabilities, allowing for more efficient
utilization of the gathered data in Industry 4.0 systems. In summary, AI plays a crucial role in
leveraging IoT services and Cloud Computing to enhance the enablers and technologies for industry.
Table 1 shows a list of the most relevant pillars that drive Industry 4.0 as extracted from [106].
The impact of the aforementioned technologies extends beyond the industrial sector, encompass-
ing home products, business models, clean energy, and broader sustainable aspects that were not
fully taken into account in previous industrial revolutions. Moreover, the industry is recognized to
be a catalyst for systemic transformation towards more sustainable economies [98]. Therefore, it is
essential to incorporate various societal and environmental aspects as new driving forces within
the industrial sector.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
4 Castañé and Vyhmeister

Table 1. Industry 4.0 Technological Enablers

Category Description
Additive Also known as 3D printing, encompasses a range of techniques dedicated to the pro-
Manufacturing duction of products through layer-by-layer deposition [127]. These methods include
vat photo-polymerization, powder bed fusion, binder jetting, material jetting, sheet
lamination, material extrusion, contour crafting, cellular fabrication, d-shape, concrete
printing, and direct energy deposition [15]. The benefits include reduced costs, reduced
supply chain, worker safety, complex forms fabrication, and short turnaround times.
Augmented By overlaying digital content onto the physical environment, this technology serves as
Reality a bridge between the digital and real-world [15]. It offers diverse benefits, including
improved product development insights, enhanced maintenance, training, issue resolu-
tion, support, quality assurance, and automation [15]. AR utilizes markers, holograms,
mobile devices, tracking, and interaction methods to enable real-time information
streaming. This facilitates the monitoring and control of virtual representations known
as Digital Twins, enhancing the management of industrial processes [15]..
Simulation Mathematical representations of systems and phenomena offer approaches that enable
the evaluation of different alternatives or scenarios within a simulated environment.
These approaches find extensive application in investment assessment, production
planning, optimization and scheduling, design, capability planning, process improve-
ment, bottleneck analysis, and resource allocation. They serve as powerful tools for
decision-making and analysis, providing insights and facilitating informed choices
within complex systems and dynamic environments.
Autonomous These systems exhibit a level of autonomy as they perceive and respond to external in-
Systems formation, gathered through various sensors, thus demonstrating a form of intelligence
[86]. They are capable of executing repetitive, hazardous, and time-consuming tasks
with high accuracy and efficiency, without the need for frequent interruptions [15].
Internet of The interconnection of machinery and sensors through communication channels, such
Things as the internet, provides the foundation for IoT. In the industrial sector, this technology
brings various advantages, including cost reduction, mass customization, improved
safety, and accelerated time to market. Extensive literature coverage underscores the
significance of IoT as a crucial enabler for other Industry 4.0 advancements [59, 74]. It
plays a pivotal role in driving the transformation and progress within the Industry 4.0
landscape.
Big Data and Aggregated data’s significance in the industry became apparent as it emerged as a
Analytics valuable tool for decision-making processes [15]. Data utilisation is further enhanced
by leveraging supporting technologies, such as AI, which can analyze and harness
more extensive datasets. The accumulated information is commonly employed for
descriptive, exploratory, predictive, and prescriptive tasks [41].
Cloud The utilization of computational resources through the internet, as facilitated by Cloud
Computing Computing, is strongly intertwined with IoT and has significant implications in the
manufacturing sector, as evidenced in the literature [8, 83]. Cloud Computing’s broad
impact drives efficiency, scalability, and digital transformation across various industries.
Cybersecurity Securing data systems in cyber-physical systems is crucial. This area involves policies
and practices to prevent attacks and unauthorized access, meets manufacturer and
consumer requirements. It ensures system integrity and protects against threats and
vulnerabilities.
Horizontal and Protocols and approaches defining machine and customer integration within the pro-
Vertical duction system encompass Horizontal and Vertical Integration. These strategies involve
Integration acquiring related businesses or controlling production/distribution stages, respectively,
to consolidate market position and differentiate from competitors [88]. They expand
the traditional perspective of product-to-service integration and contribute to a com-
prehensive industrial process.
ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 5

Industry 4.0 does not fully meet Europe’s objectives as it gives priority to economic-driven busi-
ness optimization, which could lead to technological monopolies and increased inequalities [98].
For example, it lacks a comprehensive framework to promote circular economies and restorative
feedback loops within value chains; priorities for well-being as a fundamental pillar by considering
human capabilities both within and outside industrial environments (e.g., through social gover-
nance); and a strong emphasis on the environmental impact by focusing on efficiency improvements
and the adoption of clean energy sources.
Hence, in 2020, the concept of Industry 5.0 was introduced during a workshop organized by the
European Commission, where research and technology organizations and funding agencies dis-
cussed the future vision for industry. This new concept incorporated AI and the societal dimension
as catalysts for the future roadmap for European industry [34].
An early definition of Industry 5.0 is provided in the document:

“Industry 5.0 recognises the power of industry to achieve societal goals beyond jobs and growth to
become a provider of prosperity by making production respect the boundaries of our planet and placing
the well-being of the industry worker at the centre of the production process..”

Several initiatives have been designed since then to support Industry 5.0, including efforts on
up-skilling and re-skilling European workers, especially in digital skills - Skills Agenda and Digital
Education Action Plan [28]; a more competitive industry fostered by speeding up investment in
research and innovation - Industrial Strategy; advancement on sustainable development, which is
translated into resource efficient and sustainable industries environment-friendly that lead to a
transition towards a circular economy - Green Deal [21, 94]; and the adoption of a human-centric
approach for digital technologies, defined through different proposals for AI regulation - including
the AI Act, the white paper, and the Trustworthy AI requirements [23, 37, 85].
A core pillar of the Industry 5.0 is the adoption of AI. The process targets several categories:
processing data at speed, skilled employees to operate heterogeneous technologies – AI, computing
resources, data – and the incorporation of ethical dimensions in the AI life-cycle to generate trust
and ensure safe working environments [24, 33].
However, despite the efforts of researchers to incorporate ethical considerations in AI applications
at large, the AI technology used in operation environments, and the application area where these
technologies are deployed, introduce unique concerns. In the case of Industry 5.0, the connection
between the factory’s challenges along with the AI application domains and technological pillars,
remains unclear [101].

2.2 Towards Industry 5.0

Industry 5.0 represents a paradigm shift that goes beyond solely technological and economic consid-
erations in the industrial domain. It emphasizes the integration of human progress and well-being,
encompassing sustainable, circular, and regenerative economies. This evolution acknowledges the
symbiotic interaction between machines and humans, leveraging the strengths of each counterpart
instead of pursuing replacement [14]. It envisions a collaborative environment where robots and
technologies are designed with a human-centric approach, aiming to optimize the processes in the
factories and the performance of workers.
There are different technologies that are defined as technology enablers listed by authors at
[114, 123]. These are summarised in Table 2, describing each of these as core components associated
with Industry 5.0.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
6 Castañé and Vyhmeister

Table 2. Industry 5.0 Technology Enablers

Category Description
Human - This area involves concepts that involve collaborative work – Cobots –, multi-lingual
Machine speech, gesture recognition and human intention prediction, AR in interactive pur-
interaction poses (e.g. training and inclusiveness), enhancing human physical and cognitive
capabilities.
Bio-inspired The technologies focus on materials that include properties such as self-repairing,
and smart recyclables, and embedded sensor technologies with intrinsic traceability. They can
materials be integrated within living organisms/materials, among others.
Digital Twins Higher-level models corresponding to digital representations of physical objects
and or processes. The models are linked to the real environment through the sensors’
Simulation information. These can support complex processes, such as autonomic resource
reconfiguration, replacement, or movement, in factories, increasing the flexibility of
the resources.
Data The novelty in this area is related to energy-efficient and secure data management.
sharing and In addition, given the impact of AI on mining data, Big Data analytic techniques are
processing considerably improved, becoming a new hot spot [118].
Artificial Automation and classification tasks that include, among others, causality-based AI,
Intelligence AutoML for adaptability without human intervention, human-centric AI, improved
AI components through the use of oracles/expert knowledge, improvements on the
energy consumption of AI, and scaling capabilities in dynamic systems.
Technologies These technologies develop and operationalise approaches to achieve carbon neu-
for energy trality. Foremost among these are included renewable energy sources, hydrogen and
efficiency, hydrogen carriers, low data transmission and data analyses. The European Commis-
renewable, sion has also involved initiatives to foster the sound development of these technologies
storage, – Green Deal [21]. This initiative does not only include industrial concepts (e.g. healthy
and autonomy and affordable food). Nevertheless, it clearly states the need for the industry to foster
cleaner energy and cutting-edge clean technological innovation.
Reconfigurability Reconfigurability is crucial in manufacturing systems, allowing swift adaptation to
dynamic market demands. It involves dynamic management of physical, informational,
and organizational elements to achieve flexibility and agility in the production process.
This capability addresses challenges posed by changing consumer preferences, market
trends, and the need for rapid product development. It integrates technologies like IoT,
AI, and Additive Manufacturing enabling dynamic customization, rapid prototyping,
and mass customization strategies while promoting sustainability and environmental
responsibility by reducing waste and energy consumption.

These technologies show the inclusion in Industry of new forms of sustainable, circular and
regenerative economic value creation and equitable prosperity. It is expected that future industries
will have a broader impact on societal goals and contribute to a greener ecosystem [25].

2.3 AI, Trustworthy AI and its link to Industry 5.0

AI technologies are widely employed in various digital services, within prominent technological
leading companies such as Google, Amazon, Apple, Facebook, and Microsoft. These companies
leverage AI to power a range of services, including e-commerce, online advertising, media streaming,
smart home systems, self-driving cars, and social networking. The success of these services is
attributed to the convergence of media and information technology, which fosters their growth
[108]. Within these companies, AI is utilized for diverse purposes. It enables the creation of virtual
representations of the real world, converts photos into 3D images, generates interactive maps and

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 7

immersive views, enhances personal assistants, improves language recognition, and offers a wide
array of Edge Computing-based services like predictive maintenance for device management and
Augmented Reality experiences [111].
Table 3 presents an overview of AI applications in industrial fields classified based on the research
conducted by [15, 38].
Table 3. AI Application Methods in Industrial Environment

Category Description
Process planning Research in this field is linked to the scheduling problem in the manufacturing
sector. Different approaches such as Q-learning, RF, and decision trees have
been applied for this task [72] for cost prediction, energy and resource efficiency,
workers localisation, and load forecasting, among other tasks.
Quality control This research can be seen as the root of product and process quality con-
trol (which adds process control and predictive maintenance considerations).
Nevertheless, reference settle a distinction by linking quality control to im-
plementations related to the problem of reducing quality assurance costs (e.g.
quality detection - CNN, SVM [73]; Root cause analyses [71] and classification
[109] - bayesian network,decision tree).
Predictive maintenance The research conducted in this field is dedicated to estimating the valuable
lifetime of parts and components. Literature is broad with using different AI
techniques for regression and classification [128].
Logistics This area of study shares some commonalities with the scheduling problem.
Both consider the efficient use, flow, and storage of material (i.e. optimisation).
Nevertheless, logistics considers the products from their origin, while schedul-
ing refers to the internal transformation of raw materials into products. Based
on these, some strategies are employed for system representation in scheduling
trends in the field of logistics [19].
Assistance and It focuses on supporting and enhancing employees’ capabilities. Two key as-
learning systems pects are referenced to consider the assistance; guidance of individual learning
processes and the control of competences saturation [76, 113]. Furthermore,
this cluster defines research dedicated to training concepts in manufacturing
environments.
Robotics The research is dedicated to incorporating AI within robots. The application of
Machine Learning (ML) components for automation and human collaboration
is wide and includes motion, object, and human recognition, path planning,
and improvement (i.e. optimisation) of automation tasks.
Process control and It is dedicated to implementing AI in a short-time response to systems monitor-
optimisation ing and modification by plant-wide and individual unit real-time optimisation,
parameter estimation, supervisory control, data reconciliation, alarm manage-
ment, emergency shutdown, and sensor and actuator validation, among other
tasks. [30, 78]

A key aspect is trust in technology. Building trust is needed for consumer confidence across
all market segments. To achieve this goal, the European Commission has prioritized Trustworthy
AI, aiming to establish a framework that fosters trust in the development and adoption of AI
technology. A first step on this are the Trustworthy AI requirements, based on 84 ethical guidelines,
highlighting principles, such as transparency, justice and fairness, non-maleficence, responsibility,
and privacy [51]. These requirements are defined in seven pillars, summarized in Table 4.
Furthermore, Human-centric AI emphasizes the collaboration and interaction between AI assets
and human agents. Algorithms can be continuously updated to consider the state, needs, experiences,

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
8 Castañé and Vyhmeister

Table 4. Trustworthy AI Requirements

Category Description
Human Agency Based on the principles of human autonomy and decision-making, this requirement
and Oversight includes fundamental rights and agency. It links to human-centric considerations
and Accountability. Approaches like Human-in-the-loop (HITL), Human-on-the-
loop (HOTL), and Human-in-command (HIC) extend AI autonomy and collaboration
with humans, reinforcing Human Agency and Oversight. However, HITL feasibility
depends on system granularity and functionality, limiting real-time human partici-
pation in dynamic processes. Human oversight alone doesn’t address dependency
issues and potential overconfidence in algorithms [63].
Technical Technical robustness and safety are vital in the manufacturing sector, addressing
Robustness and resilience, security, fallback plans, safety, accuracy, reliability, and reproducibility
Safety [85]. Manufacturing processes aim to withstand variability in various factors. How-
ever, the extent of robustness considerations may vary depending on the level of
risk involved.
Privacy and This requirement includes privacy, quality and integrity of data, and access to data.
Data Governance Regulatory needs regarding data managing and handling have already been set
(e.g. the GDPR regulations and [85]). The data protection regulations set rules for
businesses and organisations and, simultaneously, set rights for citizens regarding
their data rights and redress.
Transparency This requirement encompasses AI considerations, including explainability, as well
as management considerations, such as retraceability and communication. Explain-
ability involves the ability to inspect and replicate the decision-making mechanisms
employed by AI systems, while establishing clear responsibilities and ensuring
Accountability for the outcomes is essential.
Diversity, Mitigating unfair bias, ensuring accessibility, universal design, and promoting
Non-discrimination stakeholder participation are key aspects of this requirement [47]. Bias, which
and Fairness involves systematic errors or deviations from truth, is inherent to human nature and
can be observed in social information and analyses. It is essential to recognize that
AI systems may perpetuate biases present in complex historical data. Addressing
bias is crucial to prevent discrimination and unfairness, emphasizing the importance
of trustworthiness considerations in AI that prioritize the human factor, agents,
and interactions.
Environmental and This requirement encompasses sustainability, environmental friendliness, social im-
Societal well-being pact, and the principles of a democratic society. According to [20], AI offers at least
18 capabilities that can benefit the community. However, the scale of implementa-
tion determines the impact and associated risks of AI. While these considerations
have long-term implications for society, there is currently no established process
for managing and ensuring sustainable protocols in this context. It is therefore
essential for the manufacturing sector to incorporate societal well-being into its
policies, even if not explicitly stated in corporate social responsibilities.
Accountability This requirement minimizes negative impact, considers trade-offs, and ensures
redress. Accountability is crucial among AI developers, consumers, and users. Clear
definitions of uncertainties and responsibilities promote accountable behaviour
by AI developers. In the manufacturing sector, Accountability is weakly linked to
agents and their interactions. It is especially important for high-risk AI elements
and establishing legal obligations. Incorporating AI assets into business products
involves multiple stakeholders.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 9

and physical interactions with humans. To achieve this, a combination of sensed and historical
data is used to extract patterns, choices, and other trends in their behaviour. For an AI component
to be considered human-centered, it should be explainable, verifiable, physical, collaborative,
and integrative. Explainability and verifiability are linked to properties such as reliability, safety,
availability, confidentiality, integrity, and maintainability. The human-centric approach to AI aligns
in many directions with the Trustworthy AI requirements as defined by the European Commission’s
and upcoming regulations on AI.
In the context of Industry 5.0, a digitalisation strategy should consider both the enterprise’s
interests along the Trustworthy AI pillars. For sustainable development, economic, societal, and
environmental aspects are crucial for its adoption. Additionally, the resilience of the European
industry, especially in the face of disruptions due to unexpected events like a pandemic, requires
strategies that focus on adaptable production capacity, flexible business processes, and robust value
chains. In this context, AI technologies can play a significant role in enabling industrial adaptability,
supporting these objectives.
In summary, the Trustworthy AI requirements serve as essential definitions for AI systems in
transitioning from Industry 4.0 to Industry 5.0, ensuring the development of an ecosystem of trust
and sustainability required for a new industrial revolution.

3 RESEARCH METHODOLOGY
Figure 1 shows a schematic in which the research methodology is described. The processes, on
the top row of the Figure, are linked to their primary outcomes, on the bottom row in the Figure.
These are review scope, all papers, bibliometric analysis, relevant papers, classification scheme, and
systematic map results. Although the Bibliometric Analysis is not the core of the methodology, it is
added to ground a better understanding of the topic trends. These are developed in the subsequent
Subsections according to the Figure.

Definition of Keywording Data extraction

Conduct Bibliometric Screening
Research Using and Mapping
Research Analysis Papers
Question Abstracts process

Processes

Outcomes
Review Scope Classification Systematic
All Papers Analysis All Papers
RQ1, Scheme Map
2373 Results 369
RQ2, RQ3 Industrial Domain 2 Results
AI Methodological
Trustworthy AI
New Topic

Fig. 1. Systematic Mapping Approach

3.1 Definition of Research Question

The goal is defined as follows:

“To provide an understanding of trends and status on the implementation of human-

centric and Trustworthy AI technologies in the industrial sector”.

From the goal, the research questions are the following:

• RQ1 - What are the key Trustworthy AI requirements prioritized in Industry 4.0 technological
enablers and their AI applications, and how can they be further promoted?

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
10 Castañé and Vyhmeister

• RQ2 - What are the current trends in the relationship between Industry 4.0 enablers and AI
applications?
• RQ3 - What innovative approaches can facilitate the integration of Trustworthy AI into
Industry 4.0 to transition to Industry 5.0?

3.2 Search Methodology

To assess the status of AI in Industry 4.0 and establish the foundations for Industry 5.0, a mapping
process and a systematic approach were conducted. The process starts with the definition of the
key pillars that drive the topic and the use of targeted keywords to conduct literature searches [89].
These keywords were used as markers, enabling efficient search across various platforms. Keywords
and pillars were combined to address the research questions. Specifically, keyword selection was
approached by establishing specific groups to help with the definition of the components of
the keywords, taking a bottom-up approach. The methodology applied to construct queries and
definition of keywords follow the PICOC methodology – Population, Intervention, Comparison,
Outcomes, and Context [53, 90, 91]. Each of these topics is defined next:
• Population: It refers to the specific group of individuals or subjects under the interest of the
study. In the context of this work, the population is the group of manuscripts from which
research questions are formulated.
• Intervention: The intervention refers to the approach or technique applied in the empirical
study. This study involves software methodologies, tools, technologies, or procedures. Dif-
ferent AI techniques that can be applied to specific procedures in the industrial sector are
considered. It is represented by the AI Application Methods set, which includes AI applied
in Industry 4.0 (Table 3), and the Trustworthy AI set, and that defines the requirements
from the Trustworthy AI guidelines (Table 4).
• Comparison: The comparison component involves differentiating methods, processes, or
strategies. There are different AI techniques that can be applied to specific procedures in the
industrial sector, leading by a specific implementation of the EC initiatives to Industry 5.0 –
i.e. one set named AI Application Methods that is linked to the AI applied in Industry 4.0
(Table 3), and another set named Trustworthy AI that defines the requirements from the
Trustworthy AI guidelines (Table 4).
• Outcomes: Since empirical approaches or comparisons are not considered, no specific out-
comes are defined in this component.
• Context: The context provides a comprehensive view of whether the study is conducted
in academia or industry, the industrial segment, and the subject’s incentives. Thus, the
technological context imposes keywords to constrain the search, for example, the use of the
word AI or Artificial Intelligence.
The keywords are grouped into three categories: Industry, AI Application methods, and Trust-
worthy AI.
For the first category, Industry 4.0, these are: Additive Manufacturing, Augmented Reality,
Digital Twin, Autonomous Robots, Robotics, IoT, Big Data, Cloud Computing, Cybersecurity,
Horizontal and Vertical Integration. For the second category, AI Application methods: Process
planning, Quality Control, Predictive maintenance, Logistics, Robotics, Learning systems, Process
Control, or Process Optimization. Finally, for the third category, Trustworthy AI requirements:
Responsible AI, Trustworthy AI, Transparency, Explainability, Traceability, Human Agency, Human
Oversight, Privacy, Data Governance, Accountability, Fairness, Technical Robustness, and Safety.
Additionally, the list of keywords is completed with the terms AI and Artificial Intelligence, included
as population components.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 11

The search engines used for the queries are IEEE, Scopus, ACM, and Google Scholar (GS). The
first three have a common search interface for their database while in this case Google Scholar is
limited to tuples to perform logic queries.
The queries on IEEE, Scopus, and ACM are designed to focus on gathering the Industry 5.0
coverage or Industry 4.0 with enablers that allow its transition to Industry 5.0.
The Google Scholar queries are structured as follows: 𝐺𝑆1 is used to examine the correlation
between industrial topics and on Responsible AI or Trustworthy AI; 𝐺𝑆2 is designed to understand
the global trend of Industry 4.0, and it is a superset of 𝐺𝑆3 to 𝐺𝑆11; and finally, 𝐺𝑆 3 to 𝐺𝑆 1 1 focus
are on specific aspects of Industry 4.0 with AI and Manufacturing.
In the case of 𝐺𝑆4, the query has been split into two queries. One is for Simulation or modelling
in combination with AI, while the second targets AI applications in the context of Digital Twins.
The time-frame for the queries is from 2012 to June 2022 to capture temporal trends. Only English
publications were considered, and the analysis focused on journals, standards, and peer-reviewed
conference publications.
The queries using the keywords above mentioned are shown in Table 5.
Table 5. Queries to the databases IEEE, Scopus, ACM and Google Scholar

Database Search
IEEE (“Industry 4.0” OR “Additive Manufacturing” OR “Augmented Reality” OR “Digital Twin”
OR “Autonomous Robots” OR “Robotics” OR IoT OR “Internet of Things” OR “Big Data”
OR “Cloud Computing” OR “Cybersecurity” OR “Horizontal Integration” OR “Vertical
Integration”) AND (“Process Planning” OR “Quality Control” OR “Predictive Maintenance”
OR “Logistics” OR “Robotics” OR “Learning Systems” OR “Process Control” OR “Process
Optimization”) AND ( “Responsible AI” OR “Trustworthy AI” OR “Transparency” OR
“Explainability” OR “Traceability” OR “Human Agency” OR “Human Oversight” OR “Privacy”
OR “Data Governance” OR “Accountability” OR “Fairness” OR “well-being” OR “Technical
Robustness” OR “Safety”) AND ( “AI” OR “Artificial Intelligence”)
Scopus (Industry 4.0” OR “Additive Manufacturing” OR “Augmented Reality” OR “Digital Twin”
OR “Autonomous Robots” OR “Robotics” OR “IoT OR Internet of Things” OR “Big Data”
OR “Bigdata” OR “Cloud Computing” OR “Cybersecurity” OR “Horizontal Integration” OR
“Vertical Integration”) AND (“Process Planning” OR “Quality Control” OR “Predictive Main-
tenance” OR “Logistics” OR “Robotics” OR “Learning Systems” OR “Process Control” OR
“Process Optimization”) AND (“Responsible AI” OR “Trustworthy AI” OR “Transparency”
OR “Human Agency and Oversight” OR “Privacy and Data Governance” OR “Accountabil-
ity” OR “Fairness” OR “Well-being” OR “Technical Robustness and Safety”) AND (“AI” OR
“Artificial Intelligence”)
ACM (“Industry 5.0” AND “AI”) OR ((“Industry 4.0” AND (“Industry” OR “Manufacturing” OR
“Enterprise”)) AND ((“Ethics” AND “AI”) OR (“Sustainability” AND “AI”) OR “Responsible
AI” OR “Trustworthy AI”))
GS1 “Trustworthy AI” OR “Responsible AI”
GS2 (“Manufacturing” OR “Industry”)(“AI” OR “Artificial Intelligence”)(“responsible AI” OR
“Trustworthy AI”)(“Industry 4.0” OR “Additive Manufacturing” OR “Augmented Reality”
OR “Autonomous Robots” OR “IoT” OR “Big Data” OR “Cybersecurity”)

GS3 (“Industry” OR “Manufacturing”)(“Additive Manufacturing”)(“AI” OR “Artificial Intelli-

gence”)
GS4 (“Industry” OR “Manufacturing”)(“Simulation” OR “Modelling”)(“AI” OR “Artificial Intelli-
gence”)
GS4_2 (“Industry” OR “Manufacturing”)(“Digital Twin”)(“AI” OR “Artificial Intelligence”)

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
12 Castañé and Vyhmeister

GS5 (“Industry” OR “Manufacturing”)(“Augmented Reality”)(“AI” OR “Artificial Intelligence”)

GS6 (“Industry” OR “Manufacturing”)(“Robot” OR “Robotics”)(“AI” OR “Artificial Intelligence”)
GS7 (“Industry” OR “Manufacturing”)(“IoT” OR “Internet of Things”)(“AI” OR “Artificial Intelli-
gence”)
GS8 (“Industry” OR “Manufacturing”)(“Big Data” OR “Bigdata”)(“AI” OR “Artificial Intelligence”)
GS9 (“Industry” OR “Manufacturing”)(“Cloud Computing”)(“AI” OR “Artificial Intelligence”)
GS10 (“Industry” OR “Manufacturing”)(Cybersecurity”)(“AI” OR “Artificial Intelligence”)
GS11 (“Industry” OR “Manufacturing”)(“Vertical Integration” OR “Horizontal Integration”)(“AI”
OR “Artificial Intelligence”)

3.3 Bibliometric Analysis

Table 6 presents the findings before the screening phase, structured on publications per year.
The last three columns of the table show: the total number of publications for the analyzed
period, labelled as Total; the correlation of 𝐺𝑆1 – as per Table 5 – with each query individually,
labelled as 𝐶𝑜𝑟𝑟𝐺𝑆1 ; and finally, the last column labelled as %, shows to the reader the relative
percentage of queries that contain terms related to Industry 4.0 enablers – i.e. 𝐺𝑆3, 𝐺𝑆4_2, and
𝐺𝑆5-𝐺𝑆11.
Table 6. Searches in IEEE, Scopus, ACM and Google scholar databases between the years 2012 and 2022

Database 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Total 𝐶𝑜𝑟𝑟𝐺𝑆1 %
IEEE 29 26 19 30 31 51 137 202 266 439 305 1535 0.962 -
Scopus 5 5 12 15 51 63 94 97 127 155 160 788 0.862 -
ACM 0 0 0 0 3 5 13 13 21 24 18 97 0.873 -
GS1 8 5 10 10 17 21 106 875 2320 4500 3430 11302 1.000 -
GS2 1 1 1 2 9 6 46 347 859 1640 1750 4662 0.981 -
GS3 177 326 553 887 1410 2150 3610 5140 4360 10900 9900 42413 0.959 0.048
GS4 73200 82000 87300 97600 101000 103000 94900 96700 99700 74900 122000 1032300 0.121 -
GS4_2 28 23 38 26 47 239 807 1870 3710 6970 7230 20988 0.965 0.024
GS5 898 1110 1240 1580 1990 3470 5450 8230 11200 15000 12200 62368 0.943 0.070
GS6 11800 15500 16900 17700 19400 23600 32800 46100 48600 44400 42800 319600 0.766 0.361
GS7 940 1330 1950 3380 5590 11200 18100 23900 31900 42200 37900 178390 0.923 0.201
GS8 329 817 1640 2750 4140 7190 12500 16200 17000 17500 16300 96366 0.777 0.109
GS9 1820 2310 2830 3360 4150 6180 9860 14200 17100 17200 14000 93010 0.829 0.105
GS10 754 774 925 1240 2240 3320 4760 7060 9470 13700 8850 53093 0.929 0.060
GS11 1040 1010 1040 1090 1190 1410 1880 2430 2780 3310 2550 19730 0.896 0.022

The initial five rows after the header of Table 6 display the queries conducted across IEEE,
ACM, Scopus, and Google Scholar databases (𝐺𝑆1, and 𝐺𝑆2). These queries adhere to the search
approach outlined in the research methodology. It is evident that the number of publications has
increased since 2018, with many queries demonstrating a significant correlation to ethical-driven
work, independently to the industrial sector or specific Trustworthy AI concepts.
The following rows in the table assess the implications for the industrial sector, targeting the pil-
lars associated with specific concepts of Industry 4.0 (i.e., 𝐺𝑆3 to 𝐺𝑆11). The queries were performed
with and without the inclusion of responsible AI or Trustworthy AI keywords related to ethical
concepts. Analyzing the incorporation of ethical concepts across Industry 4.0 technological enablers,
it can be observed that the overall percentage of publications incorporating these concepts was
over 0.90% of the total analyzed publications. However, when considering publications after 2018,
this percentage increased to up to 3.61%. The term Industry 4.0 and its corresponding pillars were
introduced in 2011 and gained popularity after 2016 [92]. The analysis reveals that all pillars, except
for Simulation, experienced similar periodic increases in attention from the scientific community.
As previously described, the broadness and generality of the term Industry 4.0 contribute to its
disconnection from other keywords. Similarly, while Robotics and Automation have been closely

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 13

associated with Industry 4.0, their trend in the general industrial sector (𝐺𝑆6) shows a different
pattern compared to other Industry 4.0 technological enablers, with a moderate correlation coeffi-
cient of 0.766. Notably, the development of the first digital and programmable robot dates back to
1954, and its incorporation across various domains has steadily increased since then.
Furthermore, based on these observations, an estimation of the relative numbers and their impact
on the field of Industry 4.0 can be studied. The analysis indicates that Additive Manufacturing,
Digital Twins, and Horizontal and Vertical Integration are the topics least covered in the literature.
It is important to consider that Additive Manufacturing may be referred to by different names
depending on the specific technology employed. Hence, it could be underrepresented unless a
comprehensive search is conducted for each relevant topic. On the other hand, the concept of
Digital Twins is gaining increasing interest in the industrial sector, with recent considerations being
made regarding its connection with AI [17].
Simulation and Digital Twins have some similarities in their use of virtual models and data-
driven analysis, but they differ in their focus and purpose. Simulations have a broader scope and are
utilized to understand and predict system behaviour. On the other hand, Digital Twins are specific
to individual assets or processes and offer real-time monitoring and optimization capabilities.
Although Simulations are widely known, the emergence of Digital Twins has gained significant
attention due to their ability to leverage data, analytics, and connectivity for optimizing industrial
processes, enhancing efficiency, and enabling human-machine collaboration. These concepts, along
with their impact on human-machine collaboration, life-cycle management, remote monitoring
and control, optimization of industrial systems, and improvements in efficiency and productivity,
have captured widespread interest, particularly within the context of Industry 5.0 and, thus, the
growing capture of interest as observed in GS4_2.
Finally, Horizontal and Vertical Integration may have limited considerations in terms of AI. These
topics are more closely associated with communication methods and the management of systems of
systems (e.g., Edge Computing, Cloud Computing), and their direct linkage to AI could be relatively
low.

3.4 Screening papers

A systematic analysis was conducted on the collected dataset with a primary focus on assessing
the relevance of the titles to the key facets identified previously. In cases where uncertainty arose,
a more detailed analysis was performed based on the information provided in the abstracts. This
evaluation process followed a structured pipeline approach, as illustrated in Figure 2. The main
objective of this rigorous screening was to ensure that each manuscript included at least one
Trustworthy AI requirement along with an additional topic related to either AI in industry or
Industry 4.0 technological enablers. Manuscripts solely addressing AI ethics were excluded from
consideration. Furthermore, manuscripts explicitly referencing Industry 5.0 were also excluded
from the analysis to maintain a focused examination of the relevant topics of interest.

3.5 Keywording using abstracts

The predefined set of keywords served as the basis for conducting the mapping and analysis
processes in the study. However, during the screening process, and together with the keywords
associated with Trustworthy AI requirements, three distinct technologies/methods emerged that
were deemed relevant to the investigation. These technologies/methods include Blockchain, Feder-
ated Learning, MLOps, and Big Data Stream Processing. These will be described and thoroughly
discussed in the forthcoming results and discussion section of the study.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
14 Castañé and Vyhmeister

Related to AI technology or Robotics?

Any service or application related to these?

Legend
Is there an Supress the paper
Industry 4.0 topic? Yes

No

Any trustworthy AI Any trustworthy AI Supress and archive

or human centric or human centric
consideration? consideration? Analyse and archive

Evaluate abstract Accept for evaluation Supress the paper

and conclusion for
relevance

Fig. 2. Screening process followed to select the papers based on the keywords

3.6 Data extraction and Mapping process

This section presents the findings of the systematic literature review following the rigorous screen-
ing process. The material provides comprehensive details on the analysis sets, including the sources
utilized for extracting research questions and findings. Additionally, the results were enhanced
with bibliometric analyses to improve the understanding of the queried information.
In each Research Question, a succinct discussion has been incorporated to establish a direct
correspondence with the information presented in the figures. This preliminary discussion serves
as a precursor to the subsequent section, where a detailed analysis of the findings, future directions,
and identified research gaps is presented.
The outcomes are depicted in grid charts in the upcoming sections. The values used represent
the counts of a specific topic. These numbers on the intersections are the individual manuscripts
where the topics intersected were detected. The first row and last column represent the cumulative
count where all topics are aggregated in the specific facet.
Furthermore, two additional categories were added to the mapping on each facet: general
overview, and technique specific. General overview refers to manuscripts with a generalization of
the aspects of AI and, thus even when they can be relevant as a research topic, do not provide insights
on topics related to AI. There were also considered in this category those focused on Trustworthy
requirements or applications of AI without a thorough applicability within the Industry 4.0 enablers.
Technique-specific refers to manuscripts covering the application of a concrete AI technique (e.g.
Random Forest, 2D Convolutional Neural Network) or discussion of the implementation of different
AI life-cycles (e.g. Training) within the industrial sector but not in a specific industrial application
(e.g. Process planning).
Table 7, summarizes the number of mentions within each of the evaluated manuscripts with
respect to the total. Given the size limitations of the paper, the topics have been consolidated within
nine different options for each one of the columns as described in the caption.

RQ1 - What are the key Trustworthy AI requirements prioritized in Industry 4.0
technological enablers and their AI applications, and how can they be further promoted?
The Trustworthy AI topics that have garnered significant attention are those that exhibit strong
associations with technical components, as depicted in Figure 3. Among these concepts, Robustness
and Safety have received the highest interest (27.37%), followed by Transparency (22.63%), and
Privacy and Data Governance (19.3%). Notably, Human Agency and Oversight occupy the fourth

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 15

Table 7. Total Numbers of References in Each Topic

Research Trustworthy AI
Topic Industry 4.0
Type AI Application
Topic 1 75 75 36 76
Topic 2 137 126 1 2
Topic 3 10 108 12 7
Topic 4 51 104 25 12
Topic 5 29 18 126 15
Topic 6 6 28 117 115
Topic 7 - 23 81 36
Topic 8 - - 24 100
Topic 9 - - 48 14
Topics from top to bottom Research Type: Solution, Validation, Evaluation, Phylosophical, Review, Experience.
Trustworthy AI: Human Agency & Oversight, Robustness & Safety, Privacy & Data Governance, Transparency, Diversity,
Non-Disc. & Fairness, Societal & Env. Wellbeing, Accountability. Industry 4.0: General Industry, Additive Manufacturing,
Augmented Reality, Simulation, Autonomous System, IOT & Big Data, Cloud Computing, Cybersecurity, Horizontal &
Vertical Integration. AI Application: Abstract AI, Process Planning, Quality Control, Predictive Maintenance, Logistics,
Robotics, Assistance & Learning Systems, Concrete AI, Process Control & Optimisation

position (16.3%). This trend can be attributed to the widespread integration of human-in-the-loop,
human-on-the-loop, and human-in-control considerations in Autonomous Systems. In Robotics, a
substantial proportion of manuscripts focus on human interaction and emphasize robustness and
safety considerations (36% on 67 identified manuscripts across 182 mentions on Robotics). Note-
worthy findings in this context can be observed in the domain of autonomous vehicles [57, 82, 121].
Furthermore, the transportation industry has displayed a growing interest in other Trustworthy
AI topics, such as transparency [50, 104]. However, as expected in industries with high utiliza-
tion of Autonomous Systems, the primary focus revolves around Technical Robustness, Safety
[12, 22, 31], and topics related to human-centric approaches, human-machine collaboration, and
decision-making [9, 55, 100].

NUM. REF. AI APPLICATION 143 3 17 17 33 182 63 22 152

ACCOUNTABILITY 11 1 1 3 6 2 1 6 31

SOCIETAL & ENV WELLBEING 12 2 1 1 6 5 1 7 35

DIVERSITY, NON-DISCRIMINATION & FAIRNESS 11 1 2 1 2 1 1 6 25

TRANSPARENCY 33 2 4 6 13 35 13 4 33 143

PRIVACY & DATA GOVERNANCE 36 1 3 3 8 13 4 2 52 122

ROBUSTNES & SAFETY 27 4 3 5 67 20 8 39 173

HUMAN AGENCY & OVERSIGHT 13 2 1 2 53 18 5 9 103

W NG
L CE S CS S N IC AI
IE NI RO AN ST
IC TI EM IO IF Y
RV NT BO ST AT EC TH
E AN EN GI SY IS SP OR
OV PL CO INT LO RO G TIM E W
L S Y P T
ES IT MA NI
N QU
RA C AL &
O
NI US
NE O QU VE AR L TR
GE PR TI LE RO CH F.
IC & NT TE RE
ED CE M.
PR N CO U
A S N
ST ES
SI OC
AS PR

Fig. 3. Trends in AI application vs Trustworthy AI

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
16 Castañé and Vyhmeister

Another area of high interest in the literature within the context of Industry 4.0 is the combination
of IoT, Big Data, and Cloud Computing. The integration of Big Data analytics and AI in the industrial
sector is a common topic of discussion [87, 99]. However, these technologies also raise concerns
regarding privacy, particularly when personal data is collected and utilized. The use of sensitive
data, such as process data, corporate data, healthcare data, and financial data, introduces additional
security and privacy challenges. Furthermore, various communication channels, including social
media platforms and Edge/Cloud services, can pose potential privacy threats. This consideration
becomes even more significant and complex within the Industry 4.0 framework, which fosters the
development of AI-driven intelligent applications within a network interconnected and real-time-
based system.
On the other hand, the Trustworthy requirements that have received relatively less attention
in relation to Industry 4.0 technological enablers are Diversity, Non-discrimination, and Fairness
(3.95%), Accountability (4.9%), and Societal and Environmental Well-being (5.56%).
Within diversity, non-discrimination, and Fairness requirement, the topic of Fairness is the main
focus of research [54, 65, 68, 115, 126]. The relevance of diversity and non-discrimination within the
industrial sector may be relatively lower compared to other topics. However, as these aspects are
further explored and integrated into other domains – e.g. healthcare, human labour, and education
–, their consideration and implementation within Industry 4.0 can be expected to be promoted.
Accountability involves establishing mechanisms to ensure responsibility and accountability
for AI systems and their outcomes. As defined in the Trustworthy AI guidelines, accountability
encompasses topics such as auditability, minimization and reporting of negative impacts, redress,
and specification of trade-offs between different Trustworthy AI requirements. Applied to AI
Applications, only the 4.9% is identified split in between general overview (7.7%), assistance and
learning systems (3.23%), and technique specific (3.95%). The accountability field is in between a
scientific and management area and therefore requires of heterogeneous skills to have a broad
understanding to innovate on it. However, this is a key requirement for future legislation. Some
considerations are related to AI standardisation – existing and future.
The management process plays a crucial role in effectively handling accountability and can
be facilitated through the use of management tools or components to aid in the implementation,
development, and utilization of AI systems. Moreover, considering the risks associated with AI
systems, the use of risk management tools can further enhance the development of new approaches.
For instance, various frameworks and standards are currently being developed in this regard
[116, 117]. By contrast, recent efforts were made to establish standards that promote accountability
and cover other Trustworthy AI topics. For example, the IEEE 7001-2021 standard for Transparency
of Autonomous Systems [48]. This standard is applicable to Autonomous Systems, including
automated driving systems and assisted living robots, and addresses issues including recommender
systems and chatbots used for medical diagnosis. It specifically focuses on the potential to cause
physical, psychological, societal, economic, environmental, or reputation harm.
Finally, regarding societal and environmental well-being (5.56%), as mentioned in the IEEE
standard for transparency [48], the impact on society and the environment can be associated with
indirect access to AI data or processes, such as unauthorized data access. Therefore, implementing
robust privacy and data management processes can contribute to societal and environmental well-
being. As different regulatory conditions are developed and implemented for AI assets, the potential
harm resulting from specific risks can be minimized. Consequently, as policies are established, the
industry will regulate itself, reducing these risks.
Overall, when examining the trends of Trustworthy AI across different AI applications of the
Trustworthy requirements that have received the most attention are those associated with technical
components – Transparency, Privacy & Data Governance, and Robustness & Safety. Additionally,

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 17

Human agency and oversight also exhibit a relatively high level of coverage, primarily contributed
by Robotics AI assets that involve considerations of human-in-the-loop, human-on-the-loop, and
human-in-command. Given the significance of IoT, Autonomous Systems, and Cloud Computing, it
is not surprising that the most relevant AI applications are closely linked to these drivers and, at
the same time, demonstrate relevance to Trustworthy AI considerations, such as Robotics, Machine
Learning, and assistance and learning systems. Conversely, the AI applications that exhibit the
lowest relevance, such as process planning, quality control, logistics, and predictive maintenance
are already well-established topics in the industrial setting. However, the results clearly indicate
that Trustworthy AI has not yet permeated these areas, highlighting the need to focus attention on
understanding the impact of Trustworthy AI requirements on each of these specific AI applications.

RQ2 - What are the current trends in the relationship between Industry 4.0 enablers and
AI applications?
Next Figure 4 provides a classification of AI applications in industrial domains based on the Industry
4.0 Technology Enablers. The selection process for the papers ensured that manuscripts addressing
Trustworthy AI requirements, even indirectly, were included.

NUM. REF. AI APPLICATION 117 6 19 30 39 144 51 24 186

HORIZONTAL AND VERTICAL INTEGRATION 16 2 4 3 4 2 3 3 22 59

CYBERSECURITY 10 2 1 1 12 26

CLOUD COMPUTING 18 1 3 8 7 4 3 3 49 96

IOT & BIG DATA 33 2 6 10 15 5 6 3 61 141

AUTONOMOUS SYSTEM 8 2 3 3 115 25 9 24 189

SIMULATION 5 1 1 2 4 10 6 4 7 40

AUGMENTED REALITY 2 2 3 5 5 3 1 4 25

ADDITIVE MANUFACTURING 1 1 1 1 4

GENERAL INDUSTRY 25 4 7 364

L E S CS S RS
IE
W NGRO NC IC TI EM ON IF
IC
NI ST TI LE
ER
V
AN NT NA GI BO ST SA EC AB
OV PL TE
CO LO RO SY MI SP EN
Y IN G TI UE
AL E SS LIT
M A
N IN OP Q 4 .0
R A I
NE OC QU VE AR L
& N
TR
Y
GE PR TI LE RO CH
IC TE US
ED
& NT ND
PR CE CO .I
TAN SS R EF
S E
SI OC
M
AS PR NU

Fig. 4. Trends in AI application vs industry 4.0 technological enabler

Without considering general overview papers and technique specific (19% and 30.2% respectively),
Robotics (23.28%) and assistance and learning systems (8.28%) are the main areas that attracted
significant research interest in AI applications. It is noticeable the amount of work that can be
found in general descriptions and road mapping as well as in techniques that are not into any
concrete industry enabler. Furthermore, most of the research on Robotics applications focuses on
Autonomous Systems, with minimal contributions to other Industry 4.0 Technology Enablers.
The adoption of Robotics has substantially increased in industrial environments with applications
ranging from welding and disassembly to pick and place and transportation. Cooperative robots,
known as Cobots, have emerged as a new trend in Autonomous Systems, having physical interac-
tions with humans [107]. Safety considerations play a crucial role in human-robot interactions, and
they are integral to Trustworthy requirements. As depicted in Figure 3, major efforts in Robotics
are around robustness & safety is addressed with a 36.8%, Human Agency & Oversight with a
29.1%, and Transparency a 19.2%. Ensuring safety involves measures to prevent direct physical
contact between humans and robots, considerations of decision responsibility, human agency and

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
18 Castañé and Vyhmeister

oversight in interactive systems, and transparency to improve worker adoption. Moreover, several
studies have focused on novel techniques for ensuring human safety during interactions with
robots[26, 29, 105]. Often these are based on predicting the engagement state of humans and robots,
improvements in these models, including predictions of human intentions, contribute to enhancing
security [2].
In contrast to Robotics, technique-specific papers are a big representative in the classification,
published across different Industry 4.0 drivers, particularly remarkable in information distribution
such as IoT & Big Data (37.8 %) and Edge-Cloud Computing (26.3 %).
AI adoption aims to improve the robustness and quality of industrial processes. However, in-
corporating AI and ML applications presents the challenge of user acceptance. Consequently,
efforts to unveil the black-box models have gained significant attention in the sector [84]. This is
illustrated in Figure 3 on the ML techniques encompassed within technique-specific with respect
to the fields of Robustness (22.5%), Transparency (23.1%) and Privacy and Data Governance (42,6%),
in the industrial sector and about 27.3%, 22.6%, and 19.3% of the overall interest in Trustworthy
requirements. Thus, the importance of ML within technique-specific in different AI applications,
considering Trustworthy considerations, is evident. In addition, new trends in Reinforcement
learning processes have shown promise in enhancing Safety and Human Agency and Oversight in
complex environments [10, 49, 64, 82].
Among the AI applications, Augmented Reality (2.64%) and Additive Manufacturing (0.42%)
exhibit lower importance. Augmented Reality is primarily applied in human-robot collaboration
tasks, maintenance, and assembly [97], with minor contributions to manufacturing, training, and
logistics in Industry 4.0. However, as pointed out by [97], Industry 4.0 is currently focused on
resolving technical challenges, and aspects such as ergonomics, technological complexity, and
application usability is able to incorporate Augmented Reality.
In the case of Additive Manufacturing, the limited number of findings could be attributed to
different nomenclatures. Nonetheless, the application of Additive Manufacturing with Trustworthy
AI considerations requires overcoming various challenges, including data complexity, heterogeneity,
explainability, limited data availability, cyber-physical security and privacy, human agency and
oversight, and the integration of Augmented Reality for quality processing tasks such as interac-
tion with 3D geometry for quality inspection [69]. These challenges represent a convergence of
previously mentioned limitations within a specific AI application.

RQ3 - What innovations can facilitate the incorporation of Trustworthy AI guidelines

within Industry 4.0 towards Industry 5.0?
To assess the extent to which Trustworthy requirements have been integrated into AI applications
and Industry 4.0 Technology Enablers, an analysis of the readiness of these applications was
conducted using the framework proposed by Petersen et al. [89] as shown in Table 8. To these
categories, a new facet – Opinion Paper – is proposed to encompass the manuscripts with a higher
level of abstraction than Evaluation, such as the surveys.
The quantitative measure used for the incorporation of Trustworthy AI requirements within
Industry 4.0 is the Technology Readiness Level (TRL) grade, used to evaluate the manuscripts. The
TRL scale ranges from TRL 0 with the first principles, to TRL 9 where the system is implemented or
deployed in real environments. The intermediate levels represent different stages of development
and validation according to [61]. More specifically, the TRL scale is used as follows: solution proposal
(TRL 1 to 2), validation (TRL 3 to 4), and evaluation (greater than 5). On the case of philosophical
paper, opinion paper and experience paper, the TRL scale is not applicable as described in the Table.
It is important to note that, based on our analysis, no work with a TRL higher than 6 was identified
in the literature.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 19

Furthermore, in the context of ML components, the TRL was explored in a study by Lavin et
al. [61]. This study used TRL to assess the readiness of ML algorithms and their associated data
management processes for implementation in specific domains.

Table 8. Research Type Facet based on [89]

Category Description TRL

Solution Proposal An innovative or significantly extended solution is presented 1 and 2
to address a specific problem. The solution not only highlights
its uniqueness but also provides a compelling rationale or
illustrative examples to demonstrate its potential benefits.
Validation Novel techniques, methods, and algorithms are proposed 3 and 4
and validated through theoretical analysis, without
being implemented in practical use cases.
Evaluation The implemented techniques are applied in real-world >= 5
scenarios, accompanied by a thorough evaluation process.
Philosophical Paper Propose a fresh perspective on existing concepts by organizing N.A.
the field using a taxonomy or conceptual framework.
Opinion / Review Paper Present an individual viewpoint concerning a technique, N.A.
method, or algorithm, highlighting the potential for
enhancements or modifications. Avoid relying on prior
research methodologies or related work in the discussion.
Experience Papers Provide a detailed account of the practical implementation, N.A.
outlining the specific actions taken and their execution.
It is essential for the author to possess personal
experience as a prerequisite for this description.

Figures 5 to 7 show the trends of the research type vs the Trustworthy requirements, the Industry
4.0 Technology Enablers, and the AI applications domain, respectively.

NUM REF. INDUSTRY 4.0 ENABLERS 43 1 12 24 129 118 83 25 47

NUM. REF. TRUSTWORTHY AI 78 129 112 113 19 29 24

EXPERIENCE 2 3 1 1 1 8
EXPERIENCE 1 1 2 2 6

6 15 13 16 2 5 5 62 REVIEW 4 1 5 4 8 17 12 5 3 59
REVIEW

PHILOSOPHICAL 14 19 27 27 5 8 6 106 PHILOSOPHICAL 13 3 17 17 8 5 6 69

EVALUATION 3 3 3 4 13 1 1 2 5 3 1 4 17
EVALUATION

VALIDATION 34 63 49 32 7 7 6 198 9 5 13 71 49 41 11 22 221

VALIDATION

SOLUTION 20 28 18 32 5 9 7 119
SOLUTION 14 2 3 28 29 18 2 12 108

HT TY CE CY SS G Y PE
IG FE AN EN NE EIN ILIT TY
RS SA RN AR IR LB AB H Y G IT
Y ON EM TA G TY N PE
E SP FA EL NT RC TR IN TI ST DA IN RI IO TY
OV
& VE EA US UR AL UT AT
NE
S
GO AN & W OU RE LA SY G CU GR
H
&
TR
N
EN
V C ES IN
D CT MU BI MP SE RC
CY ST TA AT
IO AC .R L FA TE
D
SI US & CO ER TE EA
EN BU DA & EF RA NU MO T IN ES
AG RO
IN AL R EN IO D YB L
Y
& IM T M. NE MA M N O O U C
IC
A
F.
R
AN AC CR CIE NU GE VE
G TO CL
M DIS SO TI AU RT RE
HU PR
IV
N- DI AU VE M.
NO AD AN
D NU
Y,
SIT AL
ER O NT
DIV RI
Z
HO

Fig. 5. Facet vs Trustworthy AI Requirement Fig. 6. Facet vs Industry 4.0 technological enabler

From a general perspective, the analysis of the Figures reveals a common trend in the papers
based on their TRL. The majority of papers fall into the validation category, accounting for 39.3% for
Trustworthy AI, 46% for Industry 4.0 Technology Enablers, and 45.9% for Trends in AI applications.
This is followed by the solution category with 23.6%, 22.5% and 22.2% in the same order as earlier
described. Finally, the evaluation category with 2.6%, 3.5% and 3.9% respectively. It is evident

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
20 Castañé and Vyhmeister

UM. REF. AI APP. 84 2 7 11 15 118 36 14 99

EXPERIENCE 2 3 1 6

REVIEW 13 1 3 2 6 8 2 2 9 46

PHILOSOPHICAL 22 1 4 13 5 1 10 56

EVALUATION 3 1 1 1 1 2 1 5 15

VALIDATION 19 1 3 1 66 16 10 61 177

SOLUTION 25 1 1 5 3 27 11 13 86

L E S CS S PE
EW NG RO NC IC TI EM ON IC
VI NI ST TI IF TY
ER AN NT NA GI BO ST SA EC CH
PL CO TE LO RO SY MI SP AR
OV S Y IN G PTI E E
R AL ES A LIT M A
N IN O QU RE
S
NE OC QU VE AR L
& NI F.
PR TI LE RO CH RE
GE IC TE
ED CE
& NT M.
PR CO NU
TAN S
SI
S ES
AS OC
PR

Fig. 7. Facet vs Trends in AI applications

that TRL 3 is the most prevalent, indicating that the analyzed work primarily consists of proof of
concept and validation stages. Consequently, it is still in an early stage of development and not
yet operational for practical implementation in companies and industries. It is also important to
notice that the numbers reflect the lower adoption in industry on some of these topics and that
they are not yet ready to be used in some of the real scenarios. This semi-qualitative assessment
provides a general understanding of the progress towards the transition to Industry 5.0. However,
a more comprehensive analysis can be conducted for each specific topic within the Trustworthy AI
requirements.
Another noteworthy trend observed during the analysis is the inclusion of ethical and trust
considerations in models to make them operational and linked to different stages of development.
These models serve the purpose of enhancing the understanding of specific requirements and
providing metrics for managing ethical risks. Some notable findings include the development of
a surrogate model for explainability [110], a robot-to-human trust model [119], and a model for
transparency between emerging technologies and humanitarian logistics sustainability [56]. As
these models continue to advance, a better understanding of the collaboration between AI-driven
assets and human counterparts can be achieved.

4 DISCUSSION
4.1 Trends and enabler technologies to transition to Industry 5.0
During the assessment of the manuscripts, novel trends were identified as facilitators for the
transition from Industry 4.0 to Industry 5.0. These emerging trends encompass Blockchain, Federated
Learning, ML operations (MLOps), and Big Data Stream Processing.
These emerging trends are shown in the following figures following the same approach as
previously presented with a specific focus on comparing the four major emergent topics with the
Trustworthy AI requirements in Figure 8; Industry 4.0 Technology Enablers in Figure 9; and AI
applications in Figure 10. The relations will be described next in the Section.
Figure 8 shows the relation of the emergent trends and its relation with Trustworthy AI require-
ments. Blockchain and Federated Learning have similar trends with major contributions to Privacy
& Data Governance, and Robustness & Safety. The main difference between them is Transparency,
where Blockchain defines different benefits. MLOps only had a few contributions as described in
the literature but with a strong potential given the need for software standard methodologies in

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 21

NUM REF. INDUSTRY 5.0 ENABLERS 69 55 3 3

ACCOUNTABILITY 4 1 1 6

SOCIETAL & ENV WELLBEING 3 1 4

DIVERSITY, NON-DISCRIMINATION & FAIRNESS 2 2 4

TRANSPARENCY 19 3 2 24

PRIVACY & DATA GOVERNANCE 28 35 2 65

ROBUSTNES & SAFETY 12 13 1 26

HUMAN AGENCY & OVERSIGHT 1 1

AI
N G S NG AI
H N IN OP SI Y
KC AR ML ES TH
OC LE OC OR
BL ED PR TW
AT AM US
ER RE TR
D ST F.
FE RE
TA M.
DA NU
G
BI

Fig. 8. Emergent Trends in Trustworthy AI Requirements

the industrial sector. In the same manner, Big Data Stream Processing presents only a few works
however it has the potential to be a key enabler in the transition as will be explained later in the
Section.

NUM REF. INDUSTRY 5.0 ENABLERS 91 90 4 13

HORIZONTAL AND VERTICAL INTEGRATION 17 11 2 30

CYBERSECURITY 6 5 11

CLOUD COMPUTING 22 34 1 6 63

IOT & BIG DATA 33 37 1 6 77

AUTONOMOUS SYSTEM 2 1 3

SIMULATION 4 4

AUGMENTED REALITY 5 2 1 8

ADDITIVE MANUFACTURING 1 1

GENERAL INDUSTRY 1 1

N NG S NG S
AI OP R
H NI SI LE
KC AR ML ES AB
OC LE OC EN
BL ED PR 4.
0
AT A M Y
ER RE TR
FE
D ST D US
TA IN
DA F.
G RE
BI M
NU

Fig. 9. Emergent trends in Industry 4.0 Technology Enablers

Figure 9 depicts the mentioned methods’ contribution to the Industry 4.0 Technology Enablers.
Blockchain and Federated Learning describe similar contributions in which the communication
and data transfer/managing systems are considered (i.e. IoT, Cloud or Edge Computing, Horizontal
& Vertical Integration, and Cybersecurity).
Figure 10 presents the analysis of the contributions of AI applications within the industrial
domain. Since these methods deal with approaches of accessing and managing distributed data
and information for AI training, the Figure shows that the implementation of specific techniques
and the discussed emergent technologies grow in interest and therefore will have a big impact on
industry as drivers for the transition to Industry 5.0.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
22 Castañé and Vyhmeister

NUM REF. INDUSTRY 5.0 ENABLERS 45 44 2 7

TECHNIQUE SPECIFIC 21 31 1 1 54

PROCESS CONTROL & OPTIMISATION 11 1 1

ASSISTANCE & LEARNING SYSTEMS 2 2

ROBOTICS 2 1 3

LOGISTICS 8 1 9

PREDICTIVE MAINTENANCE 1 2 4 7

QUALITY CONTROL 2 1 3

PROCESS PLANNING 1 1

GENERAL OVERVIEW 8 7 1 2 18
4

N NG S NG N
AI OP IO
H NI SI AT
KC AR ML ES C
OC LE OC PLI
BL ED PR AP
AT I
AM F.
A
ER RE
D ST RE
FE M.
TA
DA NU
G
BI

Fig. 10. Emergent Trends in AI Applications in Industry 4.0

4.1.1 Blockchain. As highlighted by Mushtaq et al. [81], cryptographic techniques were widely
employed to safeguard sensitive information. Originally developed for information security pur-
poses, these techniques have begun to permeate various domains. Notably, Blockchain technology,
initially devised for the Bitcoin cryptocurrency, has emerged as a decentralized transaction and
data management solution [81]. Since its conceptualization in 2008, Blockchain has garnered sig-
nificant attention owing to its inherent features that ensure security, data integrity, anonymity, and
reliability. The industrial sector has witnessed a substantial increase in the adoption of Blockchain
technology within the framework of Industry 4.0. It offers numerous advantages, including the
support for autonomous decision-making processes and real-time utilization of streaming informa-
tion [42, 75, 80, 81]. The integration of intelligent sensors and actuators for enhancing industrial
operations is facilitated by the Industrial IoT, which leverages smart devices and real-time analytics
to harness the data collected from sensors.
The analytics and communication also deal with sensitive data, which should be protected against
any information security-related attacks. Thus Blockchain and other mechanisms should be used
to safely and securely process information. Furthermore, the inherently decentralized nature of
Blockchain technology provides significant advantages in mitigating system failure-related issues,
particularly in the context of autonomous system control. By leveraging Blockchain, the state
condition of deployed tools and components can be accurately measured, thereby ensuring the
robustness and reliability of the system. A seamless integration of Blockchain with AI further
enhances its utility. For instance, Blockchain technology ensures the integrity and security of
extensive data indexes originating from diverse sources. This secure foundation enables AI systems
to effectively and comprehensively learn from the data, facilitating more accurate and reliable
decision-making processes. It also facilitates Accountability, a pillar mainly driven by policies
over the AI assets that will be transferred to industry. Therefore, an emerging trend of AI-based
Blockchain data processing is paving the way for streamlined integration between these two
cutting-edge technologies [103].
4.1.2 Federated Learning. Building upon the aforementioned insights in previous sections, as
depicted in Figure 3, there is a prominent trend where the domains of IoT & Big Data, and Cloud
Computing prominently align with the critical aspects of Privacy and Data Governance require-
ments. This convergence highlights the substantial emphasis placed on safeguarding privacy and

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 23

ensuring robust Data Governance, with these aspects constituting a significant proportion of the
overall focus with 14.9% and 10.2%, respectively. A promising approach to enhance privacy in
the utilization of AI, particularly in the context of IoT, Cloud Computing, and Edge services, is
Federated Learning. Federated Learning is an innovative AI technique that allows systems to train
models locally, leveraging local updates to collaboratively build a global AI asset without the need
to transfer sensitive data externally [70, 124]. This decentralized approach enables the preservation
of data privacy while enabling effective model training and knowledge sharing across distributed
systems.
In Federated Learning, each participating device, such as IoT devices, trains its local model using
its own local data, thereby keeping the data within its control and avoiding the need to transmit it
externally. Instead, only the locally updated model parameters are communicated to a central server
or aggregator, where they are aggregated to construct a global model [70, 124]. This distributed
learning paradigm offers significant advantages in terms of data privacy and security, as it reduces
the risk of data exposure and unauthorized access to sensitive information.
Another option is Split Learning. Split Learning is an emerging distributed ML paradigm that
addresses privacy preservation challenge by keeping the data on the client devices [66, 120]. In
Split Learning, the training process is divided between client devices and a central server. The
client devices perform initial computations and transmit the extracted features to the server for
further model training. In contrast, Federated Learning involves training models on decentralized
devices (such as mobile devices or Edge nodes) without transferring the raw data to a central server.
The model is trained locally on each device, and only the model updates are aggregated on the
server. Split Learning offers several advantages, including enhanced privacy preservation, reduced
communication overhead, and improved scalability. It finds applications in various domains such
as healthcare, IoT, and Edge Computing, where data privacy and resource constraints are critical
concerns. Privacy-preserving mechanisms, such as differential privacy, can be integrated into Split
Learning to provide rigorous privacy guarantees. Secure aggregation protocols are employed to
protect the integrity of aggregated model updates. In summary, Split Learning offers a mechanism
to collectively train models without directly exchanging raw data, preserving the privacy of the
contributing parties. The relationship between Split Learning and Federated Learning is also
explored to address privacy and security concerns [112].
The integration of Federated Learning and Split Learning with IoT, Cloud, and Edge services offers
a compelling solution for privacy-sensitive applications. By keeping data local and performing
model updates collaboratively, these techniques enable privacy-preserving AI applications in
scenarios where data confidentiality is of utmost importance. This approach facilitates the training
of accurate and robust models while ensuring compliance with privacy regulations and addressing
concerns related to Data Governance. The implementation of an appropriate distribution policy
further enhances the integration of systems and models. For instance, when multiple AI processes
are employed on equal operational units under similar operational conditions but in different
locations, a suitable distribution policy facilitates the coordination and collaboration of IoT devices.
Through this collaborative effort, the sensed information from IoT devices is effectively coordinated
and utilized, leading to the development of a shared global model that demonstrates high prediction
accuracy. This collaborative learning process can be performed either at the Edge or in the Cloud,
depending on the specific requirements and constraints of the application.
However, it is important to note that collaborative learning techniques also pose several chal-
lenges. These include addressing communication and bandwidth limitations, dealing with hetero-
geneous and unevenly distributed data, ensuring model Fairness and Accountability, and managing
the coordination and synchronization of local model updates across devices [70, 124]. Overcoming
these challenges requires careful design and optimization of the learning process, as well as the

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
24 Castañé and Vyhmeister

development of appropriate mechanisms for data aggregation, model evaluation, and privacy preser-
vation. According to Yang et al. [124], advanced ML learning techniques in Federated Learning can
effectively mitigate learning latency by eliminating the need to offload data to a centralized data
center. This approach allows for local model updates on selected devices, thereby minimizing the
communication overhead associated with transmitting data to a central server. However, despite
the potential benefits, the process of updating the selected devices can pose challenges in terms of
communication bottlenecks, particularly in intelligent system applications.
Another challenge is the additional Trustworthy AI risks related to Accountability and Fairness.
These risks arise from the involvement of multiple stakeholders and devices, as well as the potential
for uneven or heterogeneous distribution of data across these entities. Ensuring Accountability
in distributed learning processes requires mechanisms for tracking and verifying the actions and
decisions of all participants involved in the collaborative learning process. Similarly, achieving
Fairness needs the development of strategies to mitigate biases and ensure equitable representation
and treatment of data across different devices and stakeholders. Addressing these Trustworthy
AI risks in such environments requires careful consideration and innovative solutions. Future
research efforts should focus on developing robust frameworks and methodologies that enhance
Accountability and Fairness in Federated Learning while preserving the privacy and security of
participant data.
Lo et al. [70] propose a methodology that combines a Blockchain-based Trustworthy Federated
Learning architecture to mitigate the risks associated with these Trustworthy AI considerations. The
integration of Blockchain technology provides a transparent and immutable record of participant
actions, ensuring Accountability and traceability in the Federated Learning process. This approach,
along with other advancements in the field, aims to enhance the Privacy and Data Governance
aspects of Federated Learning.
Furthermore, other researchers have also explored the fusion of Blockchain and Federated
Learning to promote privacy and Data Governance. Fan et al. [39] propose a hybrid Federated
Learning framework that integrates Blockchain technology to ensure secure and privacy-preserving
data sharing among participants. Zhang et al. [125] present a Blockchain-based approach for
Federated Learning, focusing on data integrity, privacy, and incentive mechanisms to encourage
active participation and cooperation among the Federated Learning participants.
These works demonstrate the growing interest in integrating Blockchain and Federated Learning
to address Accountability, Fairness, Privacy, and Data Governance concerns in collaborative AI
settings. Further research is needed to explore the full potential of these approaches and develop
comprehensive frameworks that effectively balance the diverse requirements of Trustworthy AI in
Federated Learning scenarios.
In order to establish Accountability in distributed learning, Lo et al. [70] propose the adoption
of a smart contract-based data-model provenance registry. This registry ensures a transparent
and immutable record of the data and model updates contributed by each participant in the
Federated Learning process. Furthermore, the authors introduce a weighted fair data sampler
that promotes Fairness during the training process by addressing the potential bias introduced by
varying data distributions among participants. Their approach demonstrates promising results,
showing improved generalization and accuracy compared to traditional Federated Learning settings.

4.1.3 Machine Learning Operations - MLOps. In addition to the use of smart contracts and fair
data sampling, Accountability in Federated Learning can also be enhanced through the adoption
of MLOps practices. MLOps encompasses a set of principles and techniques aimed at effectively
deploying, managing, and monitoring ML models in production. It involves collaboration between

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 25

software development (DevOps) and data science teams to ensure reliable and efficient ML model
operations.
While the literature on MLOps in the context of Federated Learning is still limited, its potential
for enhancing Accountability and Transparency in the involved processes is evident. By defining
proper processes and workflows, MLOps can provide an extension to various AI applications,
enabling better governance and Accountability throughout the AI life-cycle. Further research and
exploration of MLOps are warranted to fully realize its benefits and establish best practices in this
evolving field.
Furthermore, Fairness and societal and environmental well-being presents an opportunity for
contributions on Trustworthy AI but not on Industry 4.0. In addition to Figure 7, in Figure 3
an important gap is shown in topics related to process planning, quality control and predictive
maintenance. While in the latter it is complex to find a relation, the other two start presenting
some work related to equitable treatment of individuals or groups, ensuring that the benefits and
opportunities derived from Industry 4.0 technologies are distributed without bias or discrimination
[46]. Fairness plays a pivotal role in addressing societal concerns and promoting inclusivity in
the deployment of emerging technologies. By incorporating Fairness principles into the design,
development, and implementation of Industry 4.0 systems, equal access can strive for, unbiased
decision-making, and the elimination of discriminatory practices. Achieving Fairness in Industry 4.0
requires the identification and mitigation of potential biases in data collection, algorithmic decision-
making, and human-computer interactions, that will have an impact on the production systems
and quality control processes. It requires sandboxing environments to test the processes in a no-
harming manner, followed by robust regulatory frameworks, ethical guidelines, and Accountability
mechanisms to ensure that the benefits of industry are shared equitably across diverse populations
and that the inclusion of these processes does not impact the ongoing productions. By integrating
Fairness as a core principle, industry can contribute to a more just and inclusive society. This can
be achieved by exploiting on different phases of development [16], and MLOps techniques are a
quite useful methodology for understanding the phases where the developers should take care to
avoid undesired bias by using sandboxing environments.

4.1.4 Big Data Stream Processing. As depicted in Table 2, the attributes intrinsic to the domain of
Big Data Stream Processing exhibit a direct nexus with the fundamental tenets of Trustworthy AI.
The integration of Big Data and stream processing with Trustworthy AI has the potential to provide
a solid foundation for achieving data privacy and robustness & safety requirements. The paradigm
can effectively address these through the judicious implementation of robust data anonymization,
encryption, and access control mechanisms. Privacy-preserving techniques can be implemented
during real-time data processing. This integration can help to ensure that personal information
remains secure while extracting insights from streaming data, keeping access control mechanisms.
In addition, the Big Data Stream Processing facilitates real-time data pre-processing, cleansing,
and filtering processes. These processes serve to augment the quality and dependability of the data
reservoir exploited by AI systems. It provides expeditious processing, visualization, surveillance,
and real-time analysis of streaming data. This covers a pivotal need to endow AI systems with the
competence to effectuate prompt and precision-guided decisions and can address human agency and
oversight by implementing human-in-the-loop and human-in-control processes, thus enhancing the
trustworthiness of AI systems. In addition, with respect to Transparency and Accountability, key
requirements to address the black-box nature of AI algorithms and models deployed in industry, Big
Data Stream Processing can facilitate the mechanisms to capture and process data in a transparent
and auditable manner[102].

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
26 Castañé and Vyhmeister

Although Big Data Stream Processing has the potential to contribute significantly to various
aspects of Trustworthy AI, the existing literature lacks comprehensive coverage of its impact on
ensuring Trustworthy AI. In fact, we could only locate a single manuscript that partially addresses
the consideration of this technology’s influence on Trustworthy AI, as mentioned in [18]. Despite
the potential benefits that the technology has, it focuses on the broader impacts of IoT and Big
Data, lacking specific discussions related to Trustworthy AI.

4.2 Mapping between Industry 4.0 and Industry 5.0

Figure 11 illustrates the interconnection between Industry 4.0 and its Technology Enablers with var-
ious techniques and concepts that have the potential to contribute to the fulfilment of Trustworthy
AI requirements.

Additive
Manufacturing Human-in-the loop
Human-on-the loop Human agency
Human-in-command and Oversight
Autonomous
Systems
Big Data Technical
Stream Robustness and
Processing Safety
Augmented
Reality
Privacy and
Blockchain Data
Governance
Simulation
Industry 4.0 Federated Industry 5.0
Transparency
Learning
IoT &
Big Data
AI components Diversity non
and discrimination
algorithms and Fairness
Cloud &
Edge
AI Accountability
Standards &
Cybersecurity Frameworks
Environmental
and Societal
Horizontal & Well-being
Vertical
Integrations

Fig. 11. Fostering Industry 5.0 - linkage of technology concepts with new topics and Trustworthy AI require-
ments

The primary objective of this map is to provide a comprehensive overview rather than offering
prescriptive guidance on how to ensure AI trustworthiness. By highlighting the potential connec-
tions between Industry 4.0, its Technology Enablers, and various techniques and concepts, this
map aims to facilitate a broader understanding of the landscape and stimulate further exploration
and research in the pursuit of Trustworthy AI.
On the left side of the diagram, the Industry 4.0 pillars are listed in a column. On the right part,
the Trustworthy AI requirements and their linkage to Industry 5.0.
The middle column shows the specific techniques and concepts as Technical components, Al-
gorithms and Standards, and Frameworks, which have a transversal impact on the Technology
Enablers of Industry 4.0. These components serve as foundational elements that can be lever-
aged to address Trustworthy AI challenges. Blockchain technology, known for its decentralized
and immutable nature, enables secure and transparent transactions and data management. By
leveraging Blockchain, Industry 4.0 can enhance the traceability and auditability of AI processes,
thereby addressing the trustworthiness aspect of AI systems. Similarly, Federated Learning, an

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 27

AI technique that allows training models locally on individual devices without sharing raw data,
preserves data privacy and confidentiality. By adopting Federated Learning, Industry 4.0 systems
can achieve higher levels of privacy and data security, mitigating concerns related to data leakage
and unauthorized access.
The inclusion of MLOps in the map is limited due to insufficient specification and documentation
in the existing literature. However, MLOps encompasses the practices and processes involved
in deploying and managing ML models having the potential to be applied to any Industry 4.0
pillar to operationalise these through the use of AI. It enhances Accountability, Transparency, and
reproducibility in AI systems. However, further research and development are required to elucidate
the specific contributions of MLOps to Trustworthy AI within the context of Industry 4.0.
Big Data Stream Processing has a direct connection with the human-in-the-loop, human-on-the-
loop and human-in-command that are enablers for Additive Manufacturing, Autonomous Systems,
Simulation and Digital Twins and Augmented Reality. Furthermore, the nature of Big Data Stream
Processing also have strong connections with deployment environments, Cloud, Edge and IoT, Big
Data, and Cybersecurity. In addition, the enabler for Big Data Stream Processing goes for technical
Robustness, Privacy and Data Governance, and Transparency.
The AI components and algorithms, that is core for the drivers, has a strong link with all
the components in Industry 4.0. Not depicted on the diagram to avoid an additional crosslink.
Furthermore, it provides direct input into AI standards and frameworks needed for Accountability
and the legal part of environmental and societal well-being required for Industry 5.0.

5 CONCLUSIONS
A general perspective of the readiness of Industry 4.0 to Industry 5.0 was analyzed in the present
work, examining the advancements, challenges, and emerging trends in the realm of industrial
automation and AI. Through the analysis of existing literature, we have gained valuable insights
into the key technologies and concepts driving this transition and their implications for the future
of industrial systems.
Industry 4.0 has revolutionized traditional manufacturing processes by leveraging technologies
such as IoT, Big Data analytics, Cloud Computing, and cyber-physical systems. These advancements
have enabled increased automation, connectivity, and data-driven decision-making, leading to
enhanced productivity, efficiency, and competitiveness in industrial sectors. However, as we move
towards the era of Industry 5.0, new opportunities and challenges arise, necessitating the integration
of AI and intelligent systems into the industrial landscape.
The emergence of AI in Industry 5.0 brings about significant transformations in the way in-
dustrial processes are managed and optimized. AI technologies such as ML, Deep Learning, and
Natural Language Processing empower industrial systems to autonomously learn, adapt, and make
intelligent decisions. This paradigm shift towards intelligent and Autonomous Systems has the
potential to unlock new levels of efficiency, flexibility, and customization in industrial operations.
This document sheds light on several Technology Enablers that are driving the transition from
Industry 4.0 to Industry 5.0. Blockchain, with its decentralized and transparent nature, offers
enhanced security, data integrity, and privacy for industrial systems. Federated Learning, as an
AI technique, enables collaborative and privacy-preserving model training across distributed IoT
devices, ensuring data privacy and security Big Data processing is a key enabler for Transparency,
Robustness, Privacy and Data Governance by enabling the fast process of data to be comprehensively
visualised by humans. These technologies, along with others like MLOps, are poised to play a pivotal
role in fostering Trustworthy AI and addressing the challenges associated with Accountability,
Fairness, and Transparency in industrial systems.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
28 Castañé and Vyhmeister

Furthermore, it was revealed the importance of considering the ethical, legal, and societal
implications of the transition to Industry 5.0. As AI becomes increasingly embedded in industrial
processes, it is crucial to ensure responsible and ethical deployment, taking into account issues
such as bias, Fairness, and human-machine collaboration. Additionally, the regulatory frameworks
and standards governing AI in industrial settings need to evolve to address the unique challenges
and opportunities presented by Industry 5.0.
In conclusion, the transition from Industry 4.0 to Industry 5.0 represents a significant evolution
in industrial automation, driven by the convergence of AI, IoT, and advanced analytics. The study
shows that the most developed Trustworthy AI requirements are directly linked to technical
components. At the same time, the lowest is not directly dependent on the industrial sector since
they require a clear definition of protocols, regulations, and implementation approaches (standards)
that facilitate the transition towards Industry 5.0. Finally, attention was driven towards technological
approaches such as Blockchain, Federated Learning, MLOps, and ontological approaches to foster
different Trustworthy AI requirements. This survey has provided an overview of the current
state of the field, highlighting the key technologies, challenges, and potential solutions in the
journey towards Industry 5.0. As we navigate this transformative phase, it is crucial for researchers,
industry practitioners, and policymakers to collaborate and address the technical, ethical, and
societal considerations to ensure the successful and responsible integration of AI in the industrial
landscape. By embracing these advancements and leveraging the potential of Industry 5.0, we
can unlock unprecedented levels of productivity, sustainability, and innovation in the future of
manufacturing and beyond.

6 ACKNOWLEDGEMENTS
This paper has been partially supported by the European Commission by funding the ASSISTANT
project (no. 101000165), AI4Europe project (no 101070000) and the Science Foundation Ireland
under Grant No. 12/RC/2289-P2 for funding the Insight Centre of Data Analytics, co-funded under
the European Regional Development Fund. We would like to personally thank Mr. Shaun Gavigan
for the proofreading of this paper.

REFERENCES
[1] Ovidiu Vermesan and Joël Bacquest (Eds.). [n. d.]. Next Generation Internet of Things: distributed intelligence at the
edge and human machine-to-machine cooperation (Gistrup, Denmark, 2018). River Publishers. OCLC: 1080080699.
[2] Ahmed A. Abdelrahman, Dominykas Strazdas, Aly Khalifa, Jan Hintz, Thorsten Hempel, and Ayoub Al-Hamadi.
[n. d.]. Multimodal Engagement Prediction in Multiperson Human–Robot Interaction. 10 ([n. d.]), 61980–61991.
https://doi.org/10.1109/ACCESS.2022.3182469
[3] Aidi Ahmi, Hany Elbardan, and Raja Haslinda Raja Mohd Ali. [n. d.]. Bibliometric Analysis of Published Literature
on Industry 4.0. In 2019 International Conference on Electronics, Information, and Communication (ICEIC) (Auckland,
New Zealand, 2019-01). IEEE, 1–6. https://doi.org/10.23919/ELINFOCOM.2019.8706445
[4] Imran Ali and Huy Minh Phan. [n. d.]. Industry 4.0 technologies and sustainable warehousing: a systematic literature
review and future research agenda. 33, 2 ([n. d.]), 644–662. https://doi.org/10.1108/IJLM-05-2021-0277
[5] Luís Alves, Estrela Ferreira Cruz, Sérgio I Lopes, Pedro M Faria, and António Miguel Rosado da Cruz. [n. d.]. Towards
circular economy in the textiles and clothing value chain through blockchain technology and IoT: A review. 40, 1
([n. d.]), 3–23. https://doi.org/10.1177/0734242X211052858
[6] Mohammadhossein Amini, Shing I. Chang, and Prahalada Rao. [n. d.]. A cybermanufacturing and AI framework for
laser powder bed fusion (LPBF) additive manufacturing process. 21 ([n. d.]), 41–44. https://doi.org/10.1016/j.mfglet.
2019.08.007
[7] Angelos Angelopoulos, Emmanouel T. Michailidis, Nikolaos Nomikos, Panagiotis Trakadas, Antonis Hatziefremidis,
Stamatis Voliotis, and Theodore Zahariadis. [n. d.]. Tackling Faults in the Industry 4.0 Era—A Survey of Machine-
Learning Solutions and Key Aspects. 20, 1 ([n. d.]), 109. https://doi.org/10.3390/s20010109
[8] Zareef Askary and Ravinder Kumar. [n. d.]. Cloud Computing in Industries: A Review. In Recent Advances in Mechanical
Engineering (Singapore, 2020) (Lecture Notes in Mechanical Engineering), Harish Kumar and Prashant K. Jain (Eds.).

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 29

Springer, 107–116. https://doi.org/10.1007/978-981-15-1071-7_10

[9] David Baiden and Oleg Ivlev. [n. d.]. Human-robot-interaction control for orthoses with pneumatic soft-actuators
— Concept and initial trails. In 2013 IEEE 13th International Conference on Rehabilitation Robotics (ICORR)
(Seattle, WA, 2013-06). IEEE, 1–6. https://doi.org/10.1109/ICORR.2013.6650353
[10] Osbert Bastani. [n. d.]. Safe Reinforcement Learning with Nonlinear Dynamics via Model Predictive Shielding. In
2021 American Control Conference (ACC) (New Orleans, LA, USA, 2021-05-25). IEEE, 3488–3494. https://doi.org/10.
23919/ACC50511.2021.9483182
[11] Peter M. Bednar and Christine Welch. [n. d.]. Socio-Technical Perspectives on Smart Working: Creating Meaningful
and Sustainable Systems. 22, 2 ([n. d.]), 281–298. https://doi.org/10.1007/s10796-019-09921-1
[12] Michael Beetz, Georg Bartels, Alin Albu-Schaffer, Ferenc Balint-Benczedi, Rico Belder, Daniel Bebler, Sami Haddadin,
Alexis Maldonado, Nico Mansfeld, Thiemo Wiedemeyer, Roman Weitschat, and Jan-Hendrik Worch. [n. d.]. Robotic
agents capable of natural and safe physical interaction with human co-workers. In 2015 IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS) (Hamburg, Germany, 2015-09). IEEE, 6528–6535. https://doi.org/
10.1109/IROS.2015.7354310
[13] P. Brosset, S. Patsko, A. Khadikir, A.L. Thieullent, J. Buvat, Y. Khemka, and A. Jain. [n. d.]. Scaling AI in Manufacturing
Op[erations: A Practitioners’ Perspective. Capgemini Research Institute. https://www.capgemini.com/wp-content/
uploads/2019/12/AI-in-manufacturing-operations.pdf
[14] Davide Brunetti, Cristina Gena, and Fabiana Vernero. [n. d.]. Smart Interactive Technologies in the Human-Centric
Factory 5.0: A Survey. 12, 16 ([n. d.]), 7965. https://doi.org/10.3390/app12167965
[15] Javaid Butt. [n. d.]. A Strategic Roadmap for the Manufacturing Industry to Implement Industry 4.0. 4, 2 ([n. d.]), 11.
https://doi.org/10.3390/designs4020011
[16] Roberta Calegari, Gabriel G. Castañé, Michela Milano, and Barry O’Sullivan. 2023. Assessing and Enforcing Fairness
in the AI Lifecycle. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI-
23, Edith Elkind (Ed.). International Joint Conferences on Artificial Intelligence Organization, 6554–6562. https:
//doi.org/10.24963/ijcai.2023/735 Survey Track.
[17] G. Castañé, A. Dolgui, N. Kousi, B. Meyers, S. Thevenin, E. Vyhmeister, and P-O. Östberg. [n. d.]. The ASSISTANT
project: AI for high level decisions in manufacturing. ([n. d.]), 1–19. https://doi.org/10.1080/00207543.2022.2069525
[18] Hamza Chahed, Muhammad Usman, Ayan Chatterjee, Firas Bayram, Rajat Chaudhary, Anna Brunstrom, Javid Taheri,
Bestoun S Ahmed, and Andreas Kassler. 2023. AIDA–A holistic AI-driven networking and processing framework for
industrial IoT applications. Internet of Things (2023), 100805.
[19] Prosper Chimunhu, Erkan Topal, Ajak Duany Ajak, and Waqar Asad. [n. d.]. A review of machine learning applications
for underground mine planning and scheduling. 77 ([n. d.]), 102693. https://doi.org/10.1016/j.resourpol.2022.102693
[20] Michael Chui, Martin Harrysson, James Manyka, Roger Roberts, Rita Chung, Pieter Nel, and Ashley van Heteren. [n. d.].
Applying AI for social good | McKinsey. https://www.mckinsey.com/featured-insights/artificial-intelligence/applying-
artificial-intelligence-for-social-good
[21] Grégory Claeys, Simone Tagliapietra, Georg Zachmann, et al. 2019. How to make the European Green Deal work.
JSTOR.
[22] George Clark, Michael Doran, and William Glisson. [n. d.]. A Malicious Attack on the Machine Learning Policy
of a Robotic System. In 2018 17th IEEE International Conference On Trust, Security And Privacy In Computing And
Communications/ 12th IEEE International Conference On Big Data Science And Engineering (TrustCom/BigDataSE) (New
York, NY, USA, 2018-08). IEEE, 516–521. https://doi.org/10.1109/TrustCom/BigDataSE.2018.00079
[23] European Commission. [n. d.]. Regulation of the European Parliament and of the Council; Laying Down Harmonised
Rurles on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts. https://eur-
lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021PC0206&from=EN
[24] European Commission. 2021. Laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and
amending certain union legislative acts.
[25] EU Commission et al. 2022. Industry 5.0: towards more sustainable, resilient and human-centric indus-
try. https://ec.europa.eu/info/news/industry-50-towards-more-sustainable-resilient-and-human-centric-industry-
2021-jan-07_en
[26] Marco Costanzo, Giuseppe De Maria, Gaetano Lettera, and Ciro Natale. [n. d.]. A Multimodal Approach to Human
Safety in Collaborative Robotic Workcells. 19, 2 ([n. d.]), 1202–1216. https://doi.org/10.1109/TASE.2020.3043286
[27] S. K. Devitt, R. Horne, Z. Assaad, E. Broad, H. Kurniawati, B. Cardier, A. Scott, S. Lazar, M. Gould, C. Adamson, C.
Karl, F. Schrever, S. Keay, K. Tranter, E. Shellshear, D. Hunter, M. Brady, and T. Putland. [n. d.]. Trust and Safety.
([n. d.]). https://doi.org/10.48550/ARXIV.2104.06512 Publisher: arXiv Version Number: 1.
[28] European Commission (EC). 2020. European Skills Agenda. (2020).
[29] Kerstin Eder, Chris Harper, and Ute Leonards. [n. d.]. Towards the safety of human-in-the-loop robotics: Challenges
and opportunities for safety assurance of robotic co-workers’. In The 23rd IEEE International Symposium on Robot and

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
30 Castañé and Vyhmeister

Human Interactive Communication (Edinburgh, UK, 2014-08). IEEE, 660–665. https://doi.org/10.1109/ROMAN.2014.

6926328
[30] Thomas F. Edgar and Juergen Hahn. [n. d.]. Process Automation. In Springer Handbook of Automation, Shimon Y. Nof
(Ed.). Springer Berlin Heidelberg, 529–543. https://doi.org/10.1007/978-3-540-78831-7_31
[31] Amar El Maadi and Mohand Said Djouadi. [n. d.]. Suspicious motion patterns detection and tracking in crowded
scenes. In 2013 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR) (Linkoping, Sweden,
2013-10). IEEE, 1–6. https://doi.org/10.1109/SSRR.2013.6719327
[32] Kerem Elibal and Eren Özceylan. [n. d.]. A systematic literature review for industry 4.0 maturity modeling: state-of-
the-art and future challenges. 50, 11 ([n. d.]), 2957–2994. https://doi.org/10.1108/K-07-2020-0472
[33] European Commission. Directorate General for Communications Networks, Content and Technology. and High
Level Expert Group on Artificial Intelligence. [n. d.]. Ethics guidelines for trustworthy AI. Publications Office.
https://data.europa.eu/doi/10.2759/346720
[34] European Commission. Directorate General for Research and Innovation. [n. d.]. Enabling Technologies for Industry 5.0:
results of a workshop with Europe’s technology leaders. Publications Office. https://data.europa.eu/doi/10.2777/082634
[35] European Commission. Directorate General for Research and Innovation. [n. d.]. Industry 5.0: human centric, sustainable
and resilient. Publications Office. https://data.europa.eu/doi/10.2777/073781
[36] European Commission. Joint Research Centre. [n. d.]. AI watch, defining artificial intelligence 2.0: towards an operational
definition and taxonomy for the AI landscape. Publications Office. https://data.europa.eu/doi/10.2760/019901
[37] European Parliament. Directorate General for Internal Policies of the Union. [n. d.]. The white paper on artificial
intelligence. Publications Office. https://data.europa.eu/doi/10.2861/614816
[38] Simon Fahle, Christopher Prinz, and Bernd Kuhlenkötter. [n. d.]. Systematic review on machine learning (ML) methods
for manufacturing processes – Identifying artificial intelligence (AI) methods for field application. 93 ([n. d.]), 413–418.
https://doi.org/10.1016/j.procir.2020.04.109
[39] Sizheng Fan, Hongbo Zhang, Yuchen Zeng, and Wei Cai. [n. d.]. Hybrid Blockchain-Based Resource Trading System
for Federated Learning in Edge Computing. 8, 4 ([n. d.]), 2252–2264. https://doi.org/10.1109/JIOT.2020.3028101
[40] Stefan Ferber. [n. d.]. Industry 4.0: Agility in production? https://blog.bosch-si.com/industry40/industry-4-0-agility-
in-production/
[41] Samuel Fosso Wamba, Shahriar Akter, Andrew Edwards, Geoffrey Chopin, and Denis Gnanzou. [n. d.]. How ‘big
data’ can make big impact: Findings from a systematic review and a longitudinal case study. 165 ([n. d.]), 234–246.
https://doi.org/10.1016/j.ijpe.2014.12.031
[42] Paula Fraga-Lamas and Tiago M. Fernandez-Carames. [n. d.]. A Review on Blockchain Technologies for an Advanced
and Cyber-Resilient Automotive Industry. 7 ([n. d.]), 17578–17598. https://doi.org/10.1109/ACCESS.2019.2895302
[43] Enzo Morosini Frazzon, Jens Hartmann, Thomas Makuschewitz, and Bernd Scholz-Reiter. [n. d.]. Towards Socio-
Cyber-Physical Systems in Production Networks. 7 ([n. d.]), 49–54. https://doi.org/10.1016/j.procir.2013.05.009
[44] Ryohei Fujimaki. [n. d.]. The 6 Challenges of Implementing AI in Manufacturing. https://www.americanmachinist.
com/enterprise-data/article/21149328/the-6-challenges-of-implementing-ai-in-manufacturing-dotdata
[45] Sandra Grabowska and Sebastian Saniuk. [n. d.]. Business Models in the Industry 4.0 Environment—Results of Web
of Science Bibliometric Analysis. 8, 1 ([n. d.]), 19. https://doi.org/10.3390/joitmc8010019
[46] Na He, Zhong-Zhong Jiang, Jian Wang, Minghe Sun, and Guanghe Xie. 2021. Maintenance optimisation and
coordination with fairness concerns for the service-oriented manufacturing supply chain. Enterprise Information
Systems 15, 5 (2021), 694–724.
[47] Julian P. T Higgins, Thomas , James, Jackie Chandler, Miranda Cumpston, Tianjing Li, Matthew J Page, Welch ,
Vivian A, and Cochrane Collaboration. [n. d.]. Cochrane handbook for systematic reviews of interventions. https:
//doi.org/10.1002/9781119536604 OCLC: 1167575221.
[48] IEEE. [n. d.]. IEEE SA - Standards Store | IEEE 7000-2021. https://www.techstreet.com/ieee/standards/ieee-7000-
2021?product_id=2109271-2021&utm_medium=aem&utm_source=ieeesa&utm_campaign=ais-2021
[49] Alessandro Iucci, Alberto Hata, Ahmad Terra, Rafia Inam, and Iolanda Leite. [n. d.]. Explainable Reinforcement
Learning for Human-Robot Collaboration. In 2021 20th International Conference on Advanced Robotics (ICAR) (Ljubljana,
Slovenia, 2021-12-06). IEEE, 927–934. https://doi.org/10.1109/ICAR53236.2021.9659472
[50] Joanna Jaworek-Korjakowska, Aleksander Kostuch, and Pawel Skruch. [n. d.]. SafeSO: Interpretable and Explainable
Deep Learning Approach for Seat Occupancy Classification in Vehicle Interior. In 2021 IEEE/CVF Conference on
Computer Vision and Pattern Recognition Workshops (CVPRW) (Nashville, TN, USA, 2021-06). IEEE, 103–112. https:
//doi.org/10.1109/CVPRW53098.2021.00020
[51] Anna Jobin, Marcello Ienca, and Effy Vayena. [n. d.]. The global landscape of AI ethics guidelines. 1, 9 ([n. d.]),
389–399. https://doi.org/10.1038/s42256-019-0088-2 Number: 9 Publisher: Nature Publishing Group.
[52] Sachin S. Kamble, Angappa Gunasekaran, and Shradha A. Gawankar. [n. d.]. Sustainable Industry 4.0 framework: A
systematic literature review identifying the current trends and future perspectives. 117 ([n. d.]), 408–425. https:

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 31

//doi.org/10.1016/j.psep.2018.05.009
[53] Staffs Keele et al. 2007. Guidelines for performing systematic literature reviews in software engineering.
[54] Patrik Joslin Kenfack, Daniil Dmitrievich Arapov, Rasheed Hussain, SM Ahsan Kazmi, and Adil Khan. 2021. On the
fairness of generative adversarial networks (gans). In 2021 International Conference" Nonlinearity, Information and
Robotics"(NIR). IEEE, 1–7.
[55] Shanker Keshavdas and Geert-Jan M. Kruijff. [n. d.]. Functional mapping for human-robot collaborative exploration.
In 2012 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR) (College Station, TX, USA,
2012-11). IEEE, 1–6. https://doi.org/10.1109/SSRR.2012.6523884
[56] Muhammad Khan, Gohar Saleem Parvaiz, Abbas Ali, Majid Jehangir, Noor Hassan, and Junghan Bae. [n. d.]. A Model
for Understanding the Mediating Association of Transparency between Emerging Technologies and Humanitarian
Logistics Sustainability. 14, 11 ([n. d.]), 6917. https://doi.org/10.3390/su14116917
[57] Md Habib Ullah Khan and Md Mamun Howlader. [n. d.]. Design of An Intelligent Autonomous Accident Prevention,
Detection And Vehicle Monitoring System. In 2019 IEEE International Conference on Robotics, Automation, Artificial-
intelligence and Internet-of-Things (RAAICON) (Dhaka, Bangladesh, 2019-11). IEEE, 40–42. https://doi.org/10.1109/
RAAICON48939.2019.6263505
[58] Liane Mahlmann Kipper, Leonardo Bertolin Furstenau, Daniel Hoppe, Rejane Frozza, and Sandra Iepsen. [n. d.].
Scopus scientific mapping production in industry 4.0 (2011–2018): a bibliometric analysis. 58, 6 ([n. d.]), 1605–1627.
https://doi.org/10.1080/00207543.2019.1671625
[59] Asif Ali Laghari, Kaishan Wu, Rashid Ali Laghari, Mureed Ali, and Abdullah Ayub Khan. [n. d.]. A Review and State
of Art of Internet of Things (IoT). 29, 3 ([n. d.]), 1395–1413. https://doi.org/10.1007/s11831-021-09622-6
[60] Dave Lauer. [n. d.]. You cannot have AI ethics without ethics. 1, 1 ([n. d.]), 21–25. https://doi.org/10.1007/s43681-
020-00013-4
[61] Alexander Lavin, Ciarán M. Gilligan-Lee, Alessya Visnjic, Siddha Ganju, Dava Newman, Atılım Güneş Baydin, Sujoy
Ganguly, Danny Lange, Amit Sharma, Stephan Zheng, Eric P. Xing, Adam Gibson, James Parr, Chris Mattmann, and
Yarin Gal. [n. d.]. Technology Readiness Levels for Machine Learning Systems. https://doi.org/10.48550/arXiv.2101.
03989 arXiv:2101.03989 [cs] Number: arXiv:2101.03989.
[62] Tava Lennon Olsen and Brian Tomlin. [n. d.]. Industry 4.0: Opportunities and Challenges for Operations Management.
([n. d.]). https://doi.org/10.2139/ssrn.3365733
[63] Daphne Leprince-Ringuet. [n. d.]. AI’s big problem: Lazy humans just trust the algorithms too much. https://www.
zdnet.com/article/ai-needs-to-be-controlled-but-lazy-humans-may-not-be-up-to-the-job/
[64] Jinning Li, Liting Sun, Jianyu Chen, Masayoshi Tomizuka, and Wei Zhan. [n. d.]. A Safe Hierarchical Planning
Framework for Complex Driving Scenarios based on Reinforcement Learning. In 2021 IEEE International Conference
on Robotics and Automation (ICRA) (Xi’an, China, 2021-05-30). IEEE, 2660–2666. https://doi.org/10.1109/ICRA48506.
2021.9561195
[65] Meng Li, Donghui Hu, Chhagan Lal, Mauro Conti, and Zijian Zhang. 2020. Blockchain-enabled secure energy trading
with verifiable fairness in industrial Internet of Things. IEEE Transactions on Industrial Informatics 16, 10 (2020),
6564–6574.
[66] Qinbin Li, Yiqun Diao, Quan Chen, and Bingsheng He. 2022. Federated learning on non-iid data silos: An experimental
study. In 2022 IEEE 38th International Conference on Data Engineering (ICDE). IEEE, 965–978.
[67] Yongxin Liao, Fernando Deschamps, Eduardo de Freitas Rocha Loures, and Luiz Felipe Pierin Ramos. [n. d.]. Past,
present and future of Industry 4.0 - a systematic literature review and research agenda proposal. 55, 12 ([n. d.]),
3609–3629. https://doi.org/10.1080/00207543.2017.1308576
[68] Bo Liu, Ming Ding, Tianqing Zhu, Yong Xiang, and Wanlei Zhou. 2019. Adversaries or allies? Privacy and deep
learning in big data era. Concurrency and Computation: Practice and Experience 31, 19 (2019), e5102.
[69] Chenang Liu, Wenmeng Tian, and Chen Kan. [n. d.]. When AI meets additive manufacturing: Challenges and emerging
opportunities for human-centered products development. ([n. d.]), S0278612522000619. https://doi.org/10.1016/j.
jmsy.2022.04.010
[70] Sin Kit Lo, Yue Liu, Qinghua Lu, Chen Wang, Xiwei Xu, Hye-Young Paik, and Liming Zhu. [n. d.]. Towards Trustworthy
AI: Blockchain-based Architecture Design for Accountability and Fairness of Federated Learning Systems. ([n. d.]),
1–1. https://doi.org/10.1109/JIOT.2022.3144450 Conference Name: IEEE Internet of Things Journal.
[71] Anna Lokrantz, Emil Gustavsson, and Mats Jirstrand. [n. d.]. Root cause analysis of failures and quality deviations in
manufacturing using machine learning. 72 ([n. d.]), 1057–1062. https://doi.org/10.1016/j.procir.2018.03.229
[72] Jean-Loup Loyer, Elsa Henriques, Mihail Fontul, and Steve Wiseall. [n. d.]. Comparison of Machine Learning
methods applied to the estimation of manufacturing cost of jet engine components. 178 ([n. d.]), 109–119. https:
//doi.org/10.1016/j.ijpe.2016.05.006
[73] Liyong Ma, Wei Xie, and Yong Zhang. [n. d.]. Blister Defect Detection Based on Convolutional Neural Network for
Polymer Lithium-Ion Battery. 9, 6 ([n. d.]), 1085. https://doi.org/10.3390/app9061085

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
32 Castañé and Vyhmeister

[74] Somayya Madakam, Vihar Lake, Vihar Lake, Vihar Lake, et al. 2015. Internet of Things (IoT): A literature review.
Journal of Computer and Communications 3, 05 (2015), 164. https://doi.org/10.4236/jcc.2015.35021
[75] S. Madumidha, P. Siva Ranjani, S. Sree Varsinee, and P.S. Sundari. [n. d.]. Transparency and Traceability: In Food
Supply Chain System using Blockchain Technology with Internet of Things. In 2019 3rd International Conference on
Trends in Electronics and Informatics (ICOEI) (Tirunelveli, India, 2019-04). IEEE, 983–987. https://doi.org/10.1109/
ICOEI.2019.8862726
[76] Claudia Mazziotti, Wayne Holmes, Michael Wiedmann, Katharina Loibl, Manolis Mavrikis, Alice Hansen, and Beate
Grawemeyer. [n. d.]. Robust student knowledge: Adapting to individual student needs as they explore the concepts
and practice the procedures of fractions. 2 ([n. d.]), 9.
[77] N. N. Misra, Yash Dixit, Ahmad Al-Mallahi, Manreet Singh Bhullar, Rohit Upadhyay, and Alex Martynenko. [n. d.].
IoT, Big Data, and Artificial Intelligence in Agriculture and Food Industry. 9, 9 ([n. d.]), 6305–6324. https://doi.org/
10.1109/JIOT.2020.2998584
[78] Mojtaba Mozaffar, Shuheng Liao, Xiaoyu Xie, Sourav Saha, Chanwook Park, Jian Cao, Wing Kam Liu, and Zhengtao
Gan. [n. d.]. Mechanistic artificial intelligence (mechanistic-AI) for modeling, design, and control of advanced
manufacturing processes: Current state and perspectives. 302 ([n. d.]), 117485. https://doi.org/10.1016/j.jmatprotec.
2021.117485
[79] Pranab K. Muhuri, Amit K. Shukla, and Ajith Abraham. [n. d.]. Industry 4.0: A bibliometric analysis and detailed
overview. 78 ([n. d.]), 218–235. https://doi.org/10.1016/j.engappai.2018.11.007
[80] Marcel Muller, Sandro Rodriguez Garzon, Martin Westerkamp, and Zoltan Andras Lux. [n. d.]. HIDALS: A Hybrid IoT-
based Decentralized Application for Logistics and Supply Chain Management. In 2019 IEEE 10th Annual Information
Technology, Electronics and Mobile Communication Conference (IEMCON) (Vancouver, BC, Canada, 2019-10). IEEE,
0802–0808. https://doi.org/10.1109/IEMCON.2019.8936305
[81] Anum Mushtaq and Irfan Ul Haq. [n. d.]. Implications of Blockchain in Industry 4.O. In 2019 International Conference
on Engineering and Emerging Technologies (ICEET) (Lahore, Pakistan, 2019-02). IEEE, 1–5. https://doi.org/10.1109/
CEET1.2019.8711819
[82] Abu Jafar Md Muzahid, Md. Abdur Rahim, Saydul Akbar Murad, Syafiq Fauzi Kamarulzaman, and Md Arafatur
Rahman. [n. d.]. Optimal Safety Planning and Driving Decision-Making for Multiple Autonomous Vehicles: A
Learning Based Approach. In 2021 Emerging Technology in Computing, Communication and Electronics (ETCCE)
(Dhaka, Bangladesh, 2021-12-21). IEEE, 1–6. https://doi.org/10.1109/ETCCE54784.2021.9689820
[83] Luciano Novais, Juan Manuel Maqueira, and Ángel Ortiz-Bas. [n. d.]. A systematic literature review of cloud computing
use in supply chain integration. 129 ([n. d.]), 296–314. https://doi.org/10.1016/j.cie.2019.01.056
[84] Iulian Ogrezeanu, Anamaria Vizitiu, Costin Cius, del, Andrei Puiu, Simona Coman, Cristian Boldis, or, Alina Itu, Robert
Demeter, Florin Moldoveanu, Constantin Suciu, and Lucian Itu. [n. d.]. Privacy-Preserving and Explainable AI in
Industrial Applications. 12, 13 ([n. d.]), 6395. https://doi.org/10.3390/app12136395
[85] High-Level Expert Group on Artificial Intelligence. [n. d.]. Ethics Guidelines for Trustworthy AI. ([n. d.]).
[86] Ercan Oztemel and Samet Gursev. [n. d.]. Literature review of Industry 4.0 and related technologies. 31, 1 ([n. d.]),
127–182. https://doi.org/10.1007/s10845-018-1433-8
[87] Irina Pencheva, Marc Esteve, and Slava Jankin Mikhaylov. [n. d.]. Big Data and AI – A transformational shift for
government: So, what next for research? 35, 1 ([n. d.]), 24–44. https://doi.org/10.1177/0952076718780537
[88] Ricardo Silva Peres, Andre Dionisio Rocha, Paulo Leitao, and Jose Barata. [n. d.]. IDARTS – Towards intelligent data
analysis and real-time supervision for industry 4.0. 101 ([n. d.]), 138–146. https://doi.org/10.1016/j.compind.2018.07.
004
[89] Kai Petersen, Robert Feldt, Shahid Mujtaba, and Michael Mattsson. 2008. Systematic mapping studies in software
engineering. In 12th International Conference on Evaluation and Assessment in Software Engineering (EASE) 12. 1–10.
[90] Kai Petersen, Sairam Vakkalanka, and Ludwik Kuzniarz. 2015. Guidelines for conducting systematic mapping studies
in software engineering: An update. Information and software technology 64 (2015), 1–18.
[91] Mark Petticrew and Helen Roberts. 2008. Systematic reviews in the social sciences: A practical guide. John Wiley &
Sons.
[92] Thomas Philbeck and Nicholas Davis. [n. d.]. THE FOURTH INDUSTRIAL REVOLUTION: SHAPING A NEW ERA. 72,
1 ([n. d.]), 17–22. https://www.jstor.org/stable/26588339 Publisher: Journal of International Affairs Editorial Board.
[93] Wolter Pieters. [n. d.]. Explanation and trust: what to tell the user in security and AI? 13, 1 ([n. d.]), 53–64. https:
//doi.org/10.1007/s10676-010-9253-3
[94] Julia M. Puaschunder. [n. d.]. The Legal and International Situation of AI, Robotics and Big Data With Attention to
Healthcare. https://doi.org/10.2139/ssrn.3472885
[95] Thomas P Quinn, Manisha Senadeera, Stephan Jacobs, Simon Coghlan, and Vuong Le. [n. d.]. Trust and medical AI:
the challenges we face and the expertise needed to overcome them. 28, 4 ([n. d.]), 890–894. https://doi.org/10.1093/
jamia/ocaa268

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
When Industry meets Trustworthy AI 33

[96] Slavko Rakic, Marko Pavlovic, and Ugljesa Marjanovic. [n. d.]. A Precondition of Sustainability: Industry 4.0 Readiness.
13, 12 ([n. d.]), 6641. https://doi.org/10.3390/su13126641
[97] Vule Reljić, Ivana Milenković, Slobodan Dudić, Jovan Šulc, and Brajan Bajči. [n. d.]. Augmented Reality Applications
in Industry 4.0 Environment. 11, 12 ([n. d.]), 5592. https://doi.org/10.3390/app11125592
[98] A Renda, S Schwaag Serger, D Tataj, A Morlet, D Isaksson, F Martins, and E Giovannini. 2022. Industry 5.0, a
transformative vision for Europe: governing systemic transformations towards a sustainable industry. https://doi.org/10.
2777/17322
[99] Yuji Roh, Geon Heo, and Steven Euijong Whang. [n. d.]. A Survey on Data Collection for Machine Learning: A Big
Data - AI Integration Perspective. 33, 4 ([n. d.]), 1328–1347. https://doi.org/10.1109/TKDE.2019.2946162
[100] Leonel Rozo, Sylvain Calinon, and Darwin G. Caldwell. [n. d.]. Learning force and position constraints in human-robot
cooperative transportation. In The 23rd IEEE International Symposium on Robot and Human Interactive Communication
(Edinburgh, UK, 2014-08). IEEE, 619–624. https://doi.org/10.1109/ROMAN.2014.6926321
[101] Mark Ryan. [n. d.]. The social and ethical impacts of artificial intelligence in agriculture: mapping the agricultural AI
literature. ([n. d.]). https://doi.org/10.1007/s00146-021-01377-9
[102] Radhya Sahal, John G Breslin, and Muhammad Intizar Ali. 2020. Big data and stream processing platforms for Industry
4.0 requirements mapping for a predictive maintenance use case. Journal of manufacturing systems 54 (2020), 138–151.
[103] Ramiz Salama and Fadi Al-Turjman. [n. d.]. AI in Blockchain Towards Realizing Cyber Security. In 2022 International
Conference on Artificial Intelligence in Everything (AIE) (Lefkosa, Cyprus, 2022-08). IEEE, 471–475. https://doi.org/10.
1109/AIE57029.2022.00096
[104] A. Schenk and U. Clausen. [n. d.]. Creating Transparency in the Finished Vehicles Transportation Process Through
the Implementation of a Real-Time Decision Support System. In 2020 IEEE International Conference on Industrial
Engineering and Engineering Management (IEEM) (Singapore, Singapore, 2020-12-14). IEEE, 1017–1021. https:
//doi.org/10.1109/IEEM45057.2020.9309978
[105] Candace L. Sidner, Cory D. Kidd, Christopher Lee, and Neal Lesh. [n. d.]. Where to look: a study of human-robot
engagement. In Proceedings of the 9th international conference on Intelligent user interface - IUI ’04 (Funchal, Madeira,
Portugal, 2004). ACM Press, 78. https://doi.org/10.1145/964442.964458
[106] Luca Silvestri, Antonio Forcina, Vito Introna, Annalisa Santolamazza, and Vittorio Cesarotti. [n. d.]. Maintenance
transformation through Industry 4.0 technologies: A systematic literature review. 123 ([n. d.]), 103335. https:
//doi.org/10.1016/j.compind.2020.103335
[107] Mithun Sivan, Guruprasad Hagari, and Eshan Baranwal. [n. d.]. Cobots for FinTech. In 2021 International Conference
on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME) (Mauritius, Mauritius, 2021-10-07).
IEEE, 1–4. https://doi.org/10.1109/ICECCME52200.2021.9591113
[108] Nikos Smyrnaios. [n. d.]. L’effet GAFAM : stratégies et logiques de l’oligopole de l’internet. 188, 2 ([n. d.]), 61–
83. https://www.cairn.info/revue-communication-et-langages1-2016-2-page-61.htm Bibliographie_available: 1
Cairndomain: www.cairn.info Cite Par_available: 0 Publisher: NecPlus.
[109] A. Sumesh, K. Rameshkumar, K. Mohandas, and R. Shyam Babu. [n. d.]. Use of Machine Learning Algorithms for
Weld Quality Monitoring using Acoustic Signature. 50 ([n. d.]), 316–322. https://doi.org/10.1016/j.procs.2015.04.042
[110] Hanxiao Tan and Helena Kotthaus. [n. d.]. Surrogate Model-Based Explainability Methods for Point Cloud NNs. In
2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (Waikoloa, HI, USA, 2022-01). IEEE,
2927–2936. https://doi.org/10.1109/WACV51458.2022.00298
[111] Zacharie Tazrout. [n. d.]. GAFAM: a look back at their AI strategies and acquisitions in 2020. https://www.actuia.com/
english/gafam-a-look-back-at-their-ai-strategies-and-acquisitions-in-2020/
[112] Chandra Thapa, Pathum Chamikara Mahawaga Arachchige, Seyit Camtepe, and Lichao Sun. 2022. Splitfed: When
federated learning meets split learning. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 36.
8485–8493.
[113] Carsten Ullrich, Matthias Aust, Niklas Kreggenfeld, Denise Kahl, Christopher Prinz, and Simon Schwantzer. [n. d.].
Assistance- and knowledge-services for smart production. In Proceedings of the 15th International Conference on
Knowledge Technologies and Data-driven Business (Graz Austria, 2015-10-21). ACM, 1–4. https://doi.org/10.1145/
2809563.2809574
[114] Publications Office of the European Union. [n. d.]. Enabling Technologies for Industry 5.0 : results of a workshop
with Europe’s technology leaders. http://op.europa.eu/en/publication-detail/-/publication/8e5de100-2a1c-11eb-9d7e-
01aa75ed71a1/language-en ISBN: 9789276220480 Publisher: Publications Office of the European Union.
[115] Muhammad Habib ur Rehman, Ahmed Mukhtar Dirir, Khaled Salah, Ernesto Damiani, and Davor Svetinovic. 2021.
TrustFed: A framework for fair and trustworthy cross-device federated learning in IIoT. IEEE Transactions on Industrial
Informatics 17, 12 (2021), 8485–8494.
[116] Eduardo Vyhmeister, Gabriel Castane, P-O Östberg, and Simon Thevenin. 2023. A responsible AI framework: pipeline
contextualisation. AI and Ethics 3, 1 (2023), 175–197.

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.
34 Castañé and Vyhmeister

[117] Eduardo Vyhmeister, Gabriel Gonzalez-Castane, and P-O Östberg. 2023. Risk as a driver for AI framework development
on manufacturing. AI and Ethics 3, 1 (2023), 155–174.
[118] Junliang Wang, Chuqiao Xu, Jie Zhang, and Ray Zhong. 2022. Big data analytics for intelligent manufacturing systems:
A review. Journal of Manufacturing Systems 62 (2022), 738–752.
[119] Qiao Wang, Dikai Liu, Marc G. Carmichael, Stefano Aldini, and Chin-Teng Lin. [n. d.]. Computational Model
of Robot Trust in Human Co-Worker for Physical Human-Robot Collaboration. 7, 2 ([n. d.]), 3146–3153. https:
//doi.org/10.1109/LRA.2022.3145957
[120] Jiajun Wu, Steve Drew, Fan Dong, Zhuangdi Zhu, and Jiayu Zhou. 2023. Topology-aware Federated Learning in Edge
Computing: A Comprehensive Survey. arXiv preprint arXiv:2302.02573 (2023).
[121] Zhenyu Wu, Chuanyi Li, Jiaying Chen, and Hongbo Gao. [n. d.]. Learning driving behavior for autonomous vehicles
using deep learning based methods. In 2019 IEEE 4th International Conference on Advanced Robotics and Mechatronics
(ICARM) (Toyonaka, Japan, 2019-07). IEEE, 905–910. https://doi.org/10.1109/ICARM.2019.8834039
[122] Li Da Xu and Lian Duan. [n. d.]. Big data for cyber physical systems in industry 4.0: a survey. 13, 2 ([n. d.]), 148–169.
https://doi.org/10.1080/17517575.2018.1442934
[123] Xun Xu, Yuqian Lu, Birgit Vogel-Heuser, and Lihui Wang. [n. d.]. Industry 4.0 and Industry 5.0—Inception, conception
and perception. 61 ([n. d.]), 530–535. https://doi.org/10.1016/j.jmsy.2021.10.006
[124] Kai Yang, Yuanming Shi, Yong Zhou, Zhanpeng Yang, Liqun Fu, and Wei Chen. [n. d.]. Federated Machine Learning for
Intelligent IoT via Reconfigurable Intelligent Surface. 34, 5 ([n. d.]), 16–22. https://doi.org/10.1109/MNET.011.2000045
[125] Weishan Zhang, Qinghua Lu, Qiuyu Yu, Zhaotong Li, Yue Liu, Sin Kit Lo, Shiping Chen, Xiwei Xu, and Liming Zhu.
[n. d.]. Blockchain-Based Federated Learning for Device Failure Detection in Industrial IoT. 8, 7 ([n. d.]), 5926–5937.
https://doi.org/10.1109/JIOT.2020.3032544
[126] Yanqi Zhao, Yong Yu, Yannan Li, Gang Han, and Xiaojiang Du. 2019. Machine learning based privacy-preserving fair
data trading in big data market. Information Sciences 478 (2019), 449–460.
[127] Ting Zheng, Marco Ardolino, Andrea Bacchetti, and Marco Perona. [n. d.]. The applications of Industry 4.0 technologies
in manufacturing context: a systematic literature review. https://www.tandfonline.com/doi/epub/10.1080/00207543.
2020.1824085?needAccess=true ISSN: 0020-7543.
[128] Tiago Zonta, Cristiano André da Costa, Rodrigo da Rosa Righi, Miromar José de Lima, Eduardo Silveira da Trindade,
and Guann Pyng Li. [n. d.]. Predictive maintenance in the Industry 4.0: A systematic literature review. 150 ([n. d.]),
106889. https://doi.org/10.1016/j.cie.2020.106889

ACM Comput. Surv., Vol. 1, No. 1, Article . Publication date: March 2022.