AI & SOCIETY (2021) 36:783–796

https://doi.org/10.1007/s00146-020-01049-0

OPEN FORUM

Artificial intelligence in cyber physical systems

Petar Radanliev1 · David De Roure1 · Max Van Kleek2 · Omar Santos3 · Uchenna Ani4

Received: 19 April 2020 / Accepted: 10 August 2020 / Published online: 27 August 2020
© The Author(s) 2020

Abstract
This article conducts a literature review of current and future challenges in the use of artificial intelligence (AI) in cyber
physical systems. The literature review is focused on identifying a conceptual framework for increasing resilience with AI
through automation supporting both, a technical and human level. The methodology applied resembled a literature review
and taxonomic analysis of complex internet of things (IoT) interconnected and coupled cyber physical systems. There is an
increased attention on propositions on models, infrastructures and frameworks of IoT in both academic and technical papers.
These reports and publications frequently represent a juxtaposition of other related systems and technologies (e.g. Industrial
Internet of Things, Cyber Physical Systems, Industry 4.0 etc.). We review academic and industry papers published between
2010 and 2020. The results determine a new hierarchical cascading conceptual framework for analysing the evolution of
AI decision-making in cyber physical systems. We argue that such evolution is inevitable and autonomous because of the
increased integration of connected devices (IoT) in cyber physical systems. To support this argument, taxonomic methodol-
ogy is adapted and applied for transparency and justifications of concepts selection decisions through building summary
maps that are applied for designing the hierarchical cascading conceptual framework.

Keywords Artificial cognition · Industrial internet of things · Cyber physical systems · Industry 4.0 · Artificial intelligence ·
Anomaly detection

1 Introduction

Artificial intelligence (AI) is already changing our econ-

omy and society, and the increased AI decision making has
triggered debated on the potential harms and the need to
* Petar Radanliev make AI decision making more transparent (de Fine Licht
[email protected]
et al. 2020, forthcoming). Even with our current techno-
David De Roure logical progress, self-building technologies are possible
[email protected]
(Kammerer 2020, forthcoming). Cognitive architectures
Max Van Kleek representing truly intelligent human-like performance, that
[email protected]
includes ‘motivation, emotion, personality, and other rel-
Omar Santos evant aspects,’ are also possible (Sun 2020). Such findings
[email protected]
trigger concerns on the creation of collective ‘Borg–eye and
Uchenna Ani the We–I’ subjects, by merging the desires of many subjects,
[email protected]
e.g. though wearable connected devices, into a collective
1
Engineering Science Department, Oxford e-Research Centre, (Liberati 2020).
University of Oxford, 7 Keble Road, Oxford OX1 3QG, This articulates research questions on how the increased
England, UK AI decision making is changing our economy and soci-
2
Department of Computer Science, University of Oxford, ety, and how we can make AI decision making more
Oxford, England, UK transparent. These questions contribute to furthering the
3
Cisco Research Centre, Research Triangle Park, NC, USA discussion of this paper, especially in view of intrusive
4
Faculty of Engineering Science, STEaPP, University College self-building technologies that represent truly intelligent
London, London, England, UK

13
Vol.:(0123456789)

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

784 AI & SOCIETY (2021) 36:783–796

human-like performance, triggering the creation of collec- In this article, we refer to Artificial Intelligence (AI)
tive from desires of many subjects. not only as a technology for reasoning, planning, learning,
One such intrusive technology is the Industrial Inter- and processing, but we also refer to the ability to move and
net of Things (IIoT). Internet of Things (IoT) technology manipulate objects. This relates or research on Artificial
has become of considerable academic, government, and Intelligence (AI) with Cyber Physical Systems (CPS). By
industry interest in recent years. The IIoT can be explained Cyber Physical Systems (CPS), we refer to computer–human
as the use of internet of things technologies to improve networks, controlling physical processes, where physical
manufacturing and industrial processes. The IIoT term processes affect computations and vice versa. One modern
is closely related to the term Industry 4.0 (I4.0), which version of Cyber Physical Systems is the Internet of Things
represents at the same time: a paradigm shift in indus- (IoT). The Internet of Things (IoT) is one step forward in the
trial production, a generic designation for sets of strategic advancement of AI in machines and represents a system of
initiatives to boost national industries, a technical term interrelated computing devices, capable of operating with-
to relate to new emerging business assets, processes and out human-to-human or human-to-computer interaction. The
services, and a brand to mark a very particular historical Industrial Internet of Things (IIoT) in this study refers to
and social period. sensors and other devices networked with industrial appli-
Through reviewing a considerable academic, government cations, enabling data collection, exchange, and analysis,
and industry literature, specific research questions emerge with the objective for increase in productivity, efficiency
from the research gaps that the review has identified. There and economic benefits.
is a significant gap in current research on how the integration The review of such systems in this paper includes the
of complex and interconnected internet of things (IoT), cou- advancements in Cyber Physical Systems (CPS), the Inter-
pled in cyber physical systems (CPS), triggers inevitable and net of Things (IoT) in relation to Artificial Intelligence (AI)
autonomous evolution of artificial cognition. The literature autonomous evolution in Industry 4.0 (I4.0). In this context,
review and taxonomic analysis consider the significance of we propose the term CPS-IoT to refer to the integration of
these research gaps in the discussion on how technological cyber physical attributes into Industrial Internet of Things
advancement results with the inevitable and autonomous (IIoT) systems. This integration includes advances in real-
evolution of artificial cognition in complex, coupled and time processing, sensing, and actuation between IIoT sys-
interconnected socio-technical systems. tems and physical domains and provides capabilities for sys-
One example for Artificial Intelligence (AI) working tem analysis of the cyber and physical structures involved.
in combination with internet of things (IoT) devices is the We, therefore, focus here on artificial cognition, defined as
Tesla car. The car uses Artificial Intelligence (AI) to deter- the artificial intelligence in networked connection of people,
mine road conditions, optimal speed, weather, and to predict processes, data, and things. Therefore, artificial intelligence
pedestrians’ and cars’ movement. Another example, in the in this article represents a more inclusive and encompass-
context of Covid-19, is the use of smart buildings. While the ing concept that consolidates the cyber physical attributes
internet of things (IoT) can be used as sensors for switching of IIoT with the social aspects of the environment in which
on lights and opening doors, in combination with Artificial this technology is deployed and reflects the future cognitive
Intelligence (AI), it could also be used for predicting opti- makeup of IIoT/I4.0. The term artificial cognition in the con-
mal time for heating or cooling the building. In the future, text of this article is used to discuss effect from the evolving
Artificial Intelligence (AI) in cyber physical systems (CPS) IoT services and social networks of I4.0.
would include health and biomedical monitoring, robotics This article is structured as follows: Our methodology
systems, intelligent edge devices, among many other func- is described in Sect. 2. In Sect. 3, we discuss the findings
tions, and be used to correct natural disasters, human errors, drawn from the literature review including contributions
or malicious actions, etc. and gaps that form artificial cognition in CPS. Section 4
Hence, this is exercise is important, because with the produces a taxonomy for management techniques and their
increased number of internet of things (IoT) connected significance to the discussion on artificial cognition in I4.0.
devices, the role of cyber physical systems (CPS) has A Discussion section and a Conclusion section synthesise
changed and evolved. With the added element of Industrial our findings and ends the article.
Internet of Things (IIoT) increasing productivity, efficiency
and economic benefits, and the changing role of Artificial
Intelligence (AI) used for the creation of this new economic 2 Methods
benefits, the current five levels of cyber physical system
architecture seems obsolete. With considerations of these The methods applied in this study consist of systematic
new technologies, we focus on determining a new CPS literature review, taxonomies derived and follows existing
architecture. research studies on this topic that apply literature review

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 785

with taxonomy (Milano et al. 2020), in pursuit of narratives AND is changed to OR, then the data records change to hun-
(O’Hara 2020). Academic literature and practical studies dreds of thousands, but its relevance to the correlated topics
are consulted intensively to discuss the IoT technologies and diminishes, and focus is placed on the one topic searched
their relation to the I4.0. While the mainstream academic with the Boolean: OR. We repeated the same search with
literature offers limited insights regarding existing and Google Scholar, with all topics TOPIC: (artificial intelli-
emerging cognitive developments, we use summary maps gence) AND TOPIC: (industrial internet of things) AND
to showcase recent developments in this field. TOPIC: (internet of things) AND TOPIC: (cyber physical
Our rationale is that—as the landscape of artificial cog- systems) AND TOPIC: (industry 4.0). The same search on
nition develops and changes very quickly—merely relying Google Scholar produced 20,700 data records. Hence, to
on journal publications provides too narrow a view of the ensure the relevance to all of the topics we investigated, of
present situation. We used the analytical target cascading, our selected data records, we used both the Web of Science
combined with the grounded theory approach (Glaser and Core Collection and Google Scholar, but since the number of
Strauss 1967), to construct a conceptual cascading model for articles was much greater on Google Scholar, we used pre-
the future integration of cognition in the I4.0. These models dominately the Google Scholar search engine for analysing
then inform a qualitative empirical study for the new cog- the greater volume of data records. Since both databases
nitive feedback mechanism approach. The chosen method contain articles from the same journals, and Google Scholar
for conducting systematic literature review represented search engine is more effective in search queries on many
the following: (1) searching established journal databases topics, using Booleans, we considered this as valid argument
and updating the findings with cross checking with google for selecting the most relevant data records.
scholar search engine; (2) creating a table of search terms Since the existing CPS architecture that we reviewed and
and article inclusion criteria such as relevance, peer review, tried to update was published in 2015, we tried to include
data of publication (less than 10 years), and design of stud- literature predominately from the time period between 2015
ies. (3) we also considered ethical issues in relations to how and 2020. However, some of the most important literature
data was obtained, reported, and protected. For example, from 2010 to 2020 is also included, and for inclusiveness,
we did not include any non-peer-reviewed studies that were a very few articles from before 2010 are included in the
critical of different nations or organisations. We also did not review. Considering the purpose of this review was to update
include any literature where data sources were not included. our understanding of CPS architecture, we did not conduct
For example, studies that claim individual company and/or a historic analysis of all relevant literature. Instead, we
nation CPS or IoT performance was better (e.g. Huawei vs considered that the CPS architecture from 2015 included
Ericson vs Nokia) were not included if the data were not knowledge from historic literature, and our aim was to
included in the study, or if we were unable to verify the update that knowledge with the most recent findings on CPS
results. architecture.
Concepts that are recognised as most prominent are cat-
egorised following the grounded theory approach for catego-
3 Literature review on cyber risk analytics rising emerging concepts (Glaser and Strauss 1967). This
and artificial intelligence process is detailed in the ‘Methods’ chapter. As a result of
following the (Glaser and Strauss 1967) research approach
The literature review is focused on identifying the most arguing that ‘all you see is data’, the categorising of most
prominent concepts present in current models, infrastruc- prominent concepts identified from over 90 different sources,
tures and frameworks, from over 90 academic, government the emerging categories of concepts are diverse in research
and industry papers, reports, and technical notes, published nature. Throughout the paper, the reader meets terms related
predominately between 2010 and 2020. In our search for to: (1) economic potential; (2) cognitive design; (3) risk
data records, we used predominately Google Scholar and the engineering; (4) correlation effect; (5) cognitive feedback;
Web of Science Core Collection. For selecting the academic (6) ‘unrecognised and outdated data. These six terms are
literature, we found Google Scholar more flexible when add- just examples of the plethora of different terms and concepts
ing more search terms. For example, when adding multi- that emerge from our literature review on the topic of cyber
ple terms in the Web of Science Core Collection, with the physical system architecture. We categorised these terms
Boolean: AND, the search results are limited. We searched and concepts, and redesigned the exiting five levels of cyber
for TOPIC: (artificial intelligence) AND TOPIC: (industrial physical system architecture—or 5C (Fig. 1).
internet of things) AND TOPIC: (internet of things) AND The grounded theory method is applied to categorise
TOPIC: (cyber physical systems) AND TOPIC: (industry these diverse terms and concepts to the existing architec-
4.0). This search on the Web of Science Core Collection ture that comprises five levels of cyber physical systems
produced only 25 data records. If only one of the Booleans: or 5C (Fig. 1). The grounded theory approach is used to

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

786 AI & SOCIETY (2021) 36:783–796

Fig. 1  The 5 levels cyber physical system architecture—commonly referred to as 5C architecture

categorise these new terms and concepts, emerging form intelligence, control of physical systems, as well as their
the literature review, and organised into cascading hierar- interaction with humans.
chies of actions (in Table 1), presented as summary maps. Such approach is especially needed for not only develop-
The importance of these diverse concepts and the rela- ing nations that lack an I4.0 strategies, but also for more
tionship between seemingly unrelated concepts, is what developed countries—such as the UK and USA. The UK has
coheres to the design of the proposed hierarchical cascad- been ranked as the overall global cyber superpower followed
ing approach (in Fig. 2). by the US (Allen and Hamilton 2014). It is also reported that
The complexity of the literature coherent design becomes the UK and US are strongly protected to withstand digital
more explicit with examples that are presented throughout infrastructure cyber-attacks, which is crucial in developing a
the paper. The examples place the paper within the expe- resilient digital economy. However, in the index quantifying
riential and cultural practice of engineering. Here we pre- industrial applications in digital infrastructure key sectors,
sent one explicit example of how the research questions that the UK drops down to the 5th place and the US to the 3rd
are drawn from the literature review are then included to place. This seems to be partly due to the UK and US lag-
drive new finding and contributions on the identified gaps ging behind other countries in terms of harnessing economic
in existing literature. The first example is used to drive con- value from the I4.0 (Allen and Hamilton 2014). This could
ceptual and theoretical underpinnings of the research gap. be caused by the lack of cognitive abilities in the Internet of
This example from literature derives findings that the exact Things (IoT) deployment (Radanliev et al. 2020a).
economic impact of cognitive CPS infrastructure remains The literature review continues with identifying, catego-
to be determined (Leitão et al. 2016) although cognitive rising and relating emerging concepts to the conceptual and
CPS systems will represent a large percentage of future ICT theoretical underpinnings of the arguments that cohere to the
application in industry (Marwedel and Engel 2016). This conceptual framework design.
situation presented in this example requires a new approach
for integrating the physical and cyber subsystems of cogni-
tive CPS. The new approach needs to provide an overall
understanding of the design, development, and evolution of
cognition in CPS, and needs to integrate theories of artificial

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 787

Table 1  Summary map—table of technologies that drive artificial autonomous AI will evolve in economic and societal deci-
cognition in CPS sion making.
Taxonomy of key elements that drive AI This process is already in motion, triggered by the enor-
mous economic potential for hyper-connected economy.
CPS—cognitive communities
Literature recognises that important future business oppor-
Cyber physical systems CPS
tunities lay in the networking potential of digital economy
Internet of everything IoE
(Nicolescu et al. 2018). The infrastructure for smart man-
5 level CPS architecture 5C
ufacturing technology could create large cost savings for
Agent-oriented architecture AoA
manufacturers (Anderson 2016) and enable faster develop-
Object-oriented architecture OoA
ment of economies of scale (Brettel et al. 2016). Industrial
Cloud optimised virtual object architecture VOA
Internet, or ‘Industry 4.0,’ supports a finer granularity and
Virtual engineering objects VEO
control to meet individual customer requirements, creates
Virtual engineering processes VEP
value opportunities (Hermann et al. 2016; Shafiq et al.
Model-driven manufacturing systems MDMS
2015; Stock and Seliger 2016; Wang et al. 2016), increases
Service oriented architecture SoA
resource productivity, and provides flexibility in business
Dynamic intelligent swamps DIS
processes (Hussain 2017). The integration of cognitive
CPS—cognitive processes
cyber-physical capabilities into IIoT arguably requires a
Connected devices and networks CDN
new process for integrating physical and cyber subsystems—
Compiling for advanced analytics CfAA
including an overall understanding of the cognitive design,
Business processes and services BPS
development, and evolution of CPS and IIoT. Gaining such
Cloud distributed process planning DPP
understanding may require consolidation of IIoT theories
Physical and human networks PHN
for control of physical systems and the interaction between
CPS—cognitive societies
humans and CPS (Marwedel and Engel 2016; Roure et al.
Internet of things IoT
2019; Banks 2019).
Web of things WoT
On the other hand, the US National Institute of Stand-
Social manufacturing SM
ards and Technology (US NIST) deliberately stays away
Internet of people IoP
from formalising any process model in this space (Barrett
Internet of services IoS
et al. 2017; NIST 2018). Instead, their recent Framework
Systems of systems SoS
for Cyber Physical Systems proposes sets of artefacts and
CPS—cognitive platforms
activities that could be considered by organisations in the
Internet protocol version 6 IPv6
deployment of CPS. These proposals are the result of formal
Internet-based system and service platforms ISP
ontologies of digital artefacts and their interactions with the
Model-based development platforms MBDP
exterior world. The US NIST identifies three main views on
Knowledge development and applications KDoA
CPS that encompass identified responsibilities in the sys-
Real-time distribution RtD
tems engineering process: conceptualisation, realisation, and
assurance. Each of these three views corresponds to funda-
mental processes in the life of cognitive CPS, respectively:
3.1 Values and risks from intrusive autonomous (1) Models of CPS (design), (2) the CPS itself (implemen-
self‑building connected technologies (IoT, edge tation), and (3) CPS Assurance (validation). The trade-offs
computing) in cyber physical systems between different instantiations of these processes as well
as between critical aspects such as Security, Safety, Busi-
One of the main drives for artificial intelligence in cyber ness, and Privacy need to be understood. In this context,
physical systems is value creation. Our society is driven by Risk Engineering is proposed as an activity embedded in
social-economic values. Organisational goals are always the design, development, and lifecycle of the future CPS
based on some form of values. For example, governmental and IoT systems (Radanliev et al. 2020b). This assumes that
and non-governmental sectors are driven by the development cyber risk is just one instantiation of risk for an organisation
of societal values. Private organisations are often driven by or product and, therefore, should be subject to the higher
economic values. One of the main drives for value creation processes of compliance and regulation in each domain.
is the emerging new data streams that enable understand- Building on this understanding of risk, a cognitive feedback
ing of new events in real-time, and predicting future events. approach is needed for formalising compositional ways to
This new and emerging data come at volumes that only reason about cyber risks in an I4.0 context. For example,
AI can process with low-latency. Since this value emerges what we could do to understand and measure the systemic
from cyber physical systems, it becomes inevitable that IIoT risk is to create a requirement for automatic sharing of

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

788 AI & SOCIETY (2021) 36:783–796

Fig. 2  Hierarchical cascading framework design, describing how artificial intelligence is evolving in CPS

cyber-attacks data records between IIoT supply chains. If et al. 2017) and the impact of media coverage (Tanczer et al.
IoT connected devices are reporting the standalone risks of 2018).
a sole company, this would enable supply chain participants
to understand and differentiate between stand alone and sys- 3.3 Review on existing cyber risk analytics
temic cyber risk. However, when IoT connected devices start
reporting on standalone risks of a sole company, this could The Cyber Value at Risk (CvaR) model (World Economic
create duplicate data records, collection of irrelevant data Forum 2015), represents an attempt to understand the eco-
records, and many other complications. Hence, the cyber- nomic impact of cyber risk for individual organisations.
attack reporting needs to include an element of cognition, CVaR provides cyber risk measurement units, value anal-
possibly in the fog computing layer, because it would be ysis methods related to the cost of different cyber-attacks
challenging to implement cognition in the edge computing type (Roumani et al. 2016), and proof of concept methods
systems. that are based on data assumptions. Given the lack of data
needed to validate the CvaR model, these studies calculate
3.2 Argument for cognitive analytics the economic impact based on organisations’ ‘stand-alone’
cyber risk and, therefore, ignore the correlation effect of
The arguments for cognitive feedback approach emerge from sharing infrastructure and information and the probability
the inherent risk in integrating the physical with the cyber of cascading impacts, which represents a crucial element
world, where cyber risk environment is constantly changing of I4.0. These limitations of the CvaR model are of great
(Radanliev et al. 2018), and estimated loss of cybercrime concern, e.g. in sharing cyber risk in critical infrastructure
varies greatly (Biener et al. 2014; DiMase et al. 2015). The (Zhu et al. 2011). Critical infrastructures are vital for strong
real impact of cyber risk remains unknown (Shackelford digital economies, but issues of synchrony, components fail-
2016), mainly due to lack of suitable probabilistic data and ures, and increasing complexity demand development and
lack of a universal, standardised impact assessment frame- elaboration of new rigorous CPS methods (Rajkumar et al.
work (Radanliev et al. 2020b; Koch and Rodosek 2016). To 2010). In the absence of a common reference point of cyber
develop such a framework, accumulated risk needs to be risks, existing cyber risk assessment methodologies have led
quantified in real-time and shared across technology plat- to inconsistencies in measuring risk (Agyepong et al. 2019),
forms (Ruan 2017). This requires a dynamic understand- which negatively affects the adaptation of I4.0. Assessment
ing of the network risk. In addition, new risk elements that of IIoT cyber risk in I4.0 should be based on a system that
require cognitive analytics also need to be quantified, such enables cognitive assessment of the cyber network risk, not
as intellectual property of digital information (Anthonysamy only the stand-alone cyber risks (Craggs and Rashid 2017)
of a sole company (Radanliev 2014).

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 789

3.4 Review of financial assessment of cyber risk 4.1 Key technologies for self‑adapting system
from CPS
Before conducting the taxonomic categorisations
In early literature, existing financial models have been pro- in the summary tables (see chapter 5) in this final sec-
posed to assess information security investment (Anderson tion we compress the rationale for the categorisations
and Moore 2006; Gordon and Loeb 2002; Rodewald and (see Table 1). This section also details the four categories,
Gus 2005). However, cyber risk covers more elements than which is one less category than the five levels of CPS pre-
information security financial cost, such as brand reputation sented in Fig. 1. Our CPS architecture does not include the
(Lee et al. 2019a) or intellectual property (Lee et al. 2019b). ‘cyber’ level, which was considered as a separate level in
In terms of modelled economic and financial impact of mas- the previous architecture. We argue that cyber is far more
sive cyber-attacks, additional questions emerge in relation than an individual level: we argue that cyber is part of all
to the impact on public sector, rethinking of business pro- levels of the CPS architecture.
cesses, growth in liability risk, and mitigation options (Ruf- The academic literature we analysed outlines the evo-
fle et al. 2014). Such economic evaluations trigger a debate lution of CPS into the more inclusive and encompassing
between limited economic lifespans of digital assets and system that brings together people, process, data, and
value in inheriting ‘out of date’ data (Tan et al. 2008). In things—making networked connections and transactions
an I4.0 context, cyber risks are not only simply associated more valuable to individuals, organisations, and things.
with machines and products that store their knowledge and Hence, by applying grounded theory for categorising
create a virtual living representation in the network (Drath the literature analysed, the following key feedback man-
and Horch 2014) but also to the global flows and markets agement technologies predominated: (a) integration of
they are part of. physical flows, information flows, and financial flows; (b)
innovative approaches to managing operational processes;
(c) exploiting the industrial digitisation to gain competi-
4 Taxonomy of management technologies tiveness; (d) and utilisation of Big Data to improve the
and methodologies on AI‑enabled efficiency of production and services. From the extensive
methods literature reviewed on this topic, the requirements for
cognitive feedback are categorised in Table 1 as: follows
This section redefines the Fig. 1—5C architecture (5 lev- domain communities, processes, societies, and platforms.
els of CPS architecture) and creates a taxonomy from the These domains represent how the changing roles of inno-
chapter 3—literature review. The taxonomy represents a vation, production, logistics, and the service processes
list of focal points, listed in a summary table (Table 1), for require CPS advancements in the following: (a) domain
visualising and focusing the direction for a new CPS archi- communities; (b) internet-based system and service
tecture. To define the contribution from this study, before platforms; (c) business processes and services, and (d)
we present the new cognitive feedback mechanism, we first dynamic real-time data from physical and human networks
explain the existing 5C architecture in Fig. 1 as described (perceived as data from intelligent swamps). This intro-
in (Lee et al. 2015). The purpose of including Fig. 1 was to duces the approach used for the taxonomic categorisations
discuss the weaknesses of the current understanding of CPS and the summary tables in chapter 5.
architecture.
From Fig. 1, we can see that the current five levels cyber
physical system architecture (5C) includes one level for
cyber elements. With the rise of connected devices—IoT 5 Summary of the taxonomic analysis:
and IIoT, and AI in human–computer interactions, the cyber building summary maps
level is obsolete, because each level contains various cyber
elements. In this study, we seek for improved understanding Although we described the process in the previous section,
of CPS architecture and we seek that though a taxonomy of we wanted to explain further that this section—chapter 5—
recent literature. is summarising the findings from the literature review in
The new cognitive feedback mechanism builds upon the chapter 4 and categorises the emerging terms and con-
existing recommendations that CPS needs to adapt quickly cepts into actions and activities, presented as hierarchical
(Niggemann et al. 2015), to create multi-vendor and modular cascades of activities in a summary map (Table 1). The
production systems (Weyer et al. 2015). Requiring under- summary map in Table 1 is the first step in building a
standing of multi-discipline testing (Balaji et al. 2015), sys- new theory and improving the current five levels of CPS
tem sociology (Dombrowski and Wagner 2014), and social architecture with a new and more up-to-date architecture.
networks (Wan et al. 2015; Roure et al. 2015). Before presenting the summary map, we briefly discuss

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

790 AI & SOCIETY (2021) 36:783–796

the categorisations as described in the previous section 5.2 Summary map of emerging terms
(domain communities, processes, societies, and plat- and concepts—presented as actions
forms.) and refer to their origin—with references from and activities
the literature review.
The categories of key elements for artificial cognition in
5.1 Taxonomic categorisations for advancing CPS are presented in Table 1. The relationships of these
the existing 5C architecture elements to CPS is grouped with the grounded theory into
the following categories: CPS—cognitive communities,
The domain communities, processes, societies, and plat- CPS—cognitive processes, CPS—cognitive societies and
forms, are expanded into (1) domain communities; (2) CPS—cognitive platforms. These categories and the syn-
internet-based system and service platforms; (3) business ergies between the elements lead to artificial cognition in
processes and services, and (4) dynamic real-time data from CPS for self-aware process are categorised in Table 1. The
physical and human networks. taxonomic analysis of the literature reviewed is applied to
Domain communities include the following: Agent-ori- structure closely related concepts higher and looser rela-
ented Architecture (Ribeiro et al. 2010), Object-oriented tionships lower within each category in the Table 1 sum-
Architecture (Thramboulidis 2015), Cloud optimised Vir- mary maps. These communities, processes, societies, and
tual Object Architecture (Giordano et al. 2016), supported platforms emerged from categorising the literature review
with Virtual Engineering Objects and Virtual Engineering findings. The taxonomic interpretation of the relationships
Processes with Internet Protocol version 6 (IPv6) connected between these concepts is built upon the literature findings
devices and networks (Wahlster et al. 2013). and represent the backbone of theoretical development and
Internet-based system and service platforms (La and Kim its understanding of interconnected concepts in this paper.
2010) are used to model CPS through the Web of Things We created the taxonomic categorisations in Table 1 to seek
(Dillon et al. 2011), with compiling of data, processes, improvement and update of the existing CPS architecture
devices, and systems for cognitive analytics and connection (see Fig. 1). In the taxonomy, we relate the merging concepts
to cognitive model-driven (robot-in-the-loop) manufactur- to the original concept in Fig. 1, but we do not include the
ing systems (Jensen et al. 2011; Shi et al. 2011; Wang et al. ‘cyber’ layer. We considered cyber to be an integral part
2014). Internet-based system and service platforms can be of all layers in CPS architecture. Hence, the taxonomy in
used to promote model-based development platforms, such Table 1 contains four levels of CPS architecture.
as behaviour modelling of robotic systems, e.g. Automata In brief, the summary table (Table 1) can also be seen as
(Ringert et al. 2015). Internet-based systems and service multiple cascading hierarchies of actions—found in litera-
platforms can enable the development of social manufac- ture as terms and concepts. The six terms mentioned in the
turing and interconnect with the Internet of People to create introduction of chapter 3—literature review, are dissected in
CPS collaborative communities (Lee et al. 2014). greater detail, with more specific focus on presenting actions
Business processes and services need to be intercon- and activities, not desired objectives. For example, from the
nected into industrial value chains to integrate machine six terms, we used the first term ‘(1) economic potential;’
information into decision making and be connected to the and in the literature review, we investigated for actions and
Internet of Services for service oriented CPS architecture activities that are related to this term. In the summary table
(Wang et al. 2015) and Cloud distributed planning manu- (Table 1), we can see new terms and concepts, e.g. Busi-
facturing. Business processes and services in CPS can also ness processes and services; Model-driven manufacturing
promote knowledge development of business areas and systems; etc. The wording in these terms and concepts is
applications. structured in a more actionable form. For example, the term
Dynamic real-time data from physical and human net- (1) ‘economic potential’ does not provide any guidance on
works (perceived as dynamic intelligent swamps) of modules how this economic potential can be achieved. We just dis-
connected to physical and human networks, can operate as covered that ‘economic potential’ was strongly present in
systems of systems, and can act as mechanisms for real-time literature on cyber physical system architecture. So we used
distribution (Kang et al. 2012) and feedback directly from this term as one of the six guidance terms in the introduction
users and markets. of chapter 3. But in the summary table (Table 1), we can see
these terms as actionable concepts, e.g. ‘Model-driven man-
ufacturing systems’ that explain what needs to be done to
reach the ‘economic potential’. The summary table (Table 1)
presents multiple cascading hierarchies of actions that are
used in the design of the cascading hierarchy framework in
(in Fig. 2). Instead of presenting these actions and activities

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 791

as categories related to the six terms, we used the recom- Table 1. Our aim was not to confirm that the embodiment of
mendations from the literature where we found the related AI in the IIoT is leading to a transformation in AI; we con-
actions and activities. We wanted to determine if these six sider that as a given—postulate from the beginning of this
terms are true representation of all the terms and concepts in study. Our aim in this section was to present advancement
literature, or was a different structure more relevant. to the current 5 levels CPS architecture (5C) as presented
The summary maps in Table 1 confirm that a notion of in Fig. 1 and to integrate the plethora of emerging concepts
artificial cognition in CPS goes beyond machine to machine from our literature review, which are not included in the
(M2M) (Wan et al. 2013; Stojmenovic 2014), and beyond current 5C architecture—(in Fig. 1).
the proposed 3 level CPS, which are (1) services, (2) cloud, The summary maps in Table 1 should be seen from a
and (3) physical object layers. Artificial cognition in CPS conceptual standpoint, and not from engineering perspective
also goes beyond the existing knowledge of 5C architec- on the definition of terms. If seen on a standalone bases, the
ture (as seen in Fig. 1). When artificial cognition in CPS is summary maps in Table 1 could be seen as concepts that rep-
combined with intelligent manufacturing equipment (Posada resent a diverse set of different terms. From reading the sum-
et al. 2015), then a new set of communities, processes, socie- mary maps categorisations in Table 1, the Internet Protocol
ties, and platforms (categorised in Table 1) emerge. When v6 is categorised as a platform, while from an engineering
combined, these new machineries represent intrusive self- perspective IPv6 is a networking protocol. There are mul-
building technologies, triggering an inevitable and autono- tiple categorisations of this type. To reduce the categories
mous evolution of artificial cognition in CPS. and themes in our pursuit of deeper understanding of these
This evolution goes beyond the existing description of categories, the grounded theory approach used the Pugh-
5C architecture (in Fig. 1). The new description of artifi- controlled convergence and, in the process, themes are asso-
cial cognition in CPS (as seen in Table 1) is based on the ciated with the ‘best fit’ categories. The rationale for this cat-
integration of artificial intelligence (AI), machine learning, egorisation is as follows: Protocol (e.g. the Internet Protocol
the cloud, and IoT, creating systems of machines capable of v6) is the official procedure or system of rules governing the
interacting with humans (Carruthers 2014). For example, communication or activities of programs and/or industries.
the application of behaviour economics into CPS already Platform on the other hand refers to the technologies that
enables market speculation on human behaviour (Rutter are used as a base upon which other applications, processes
2015), and even neuromarketing (Lewis and Brigder 2004), or technologies are developed. A CPS in the context of this
to determine consumer purchasing behaviour. We can expect categorisation is a platform, while the languages it uses to
to see autonomous CPS adopting the use of these methods communicate (e.g. IPv6) with software are the protocol.
to predetermine human behaviour. Further clarification as why such categorisations have
Technologies described in Table 1 that would enable arti- been made by applying the Pugh-controlled convergence to
ficial cognition in CPS include software defined networks reduce the number of categories is that we can consider a
(Kirkpatrick 2013) and software-defined storage (Ouyang platform as a software, while protocol is more like a theory,
et al. 2014), built upon the following: protocols and enter- or theoretical model which a platform can be based on. In
prise grade cloud hosting; AI, machine learning, and data the interest of keeping the cascading hierarchy design to a
analytics (Kambatla et al. 2014; Pan et al. 2015); and mesh level that can easily be understood, the presented categorisa-
networks and peer-to-peer connectivity (Wark et al. 2007). tions have been associated in abbreviated form in Table 2.
Without cognitive risk analytics, the embedded control of
CPS is creating security and risk management vulnerabilities
from integrating less secured systems, triggering questions
Table 2  Emerging 4 levels CPS architecture
regarding risk management and liability for breaches and
damages (Boyes et al. 2018). Without cognitive risk analyt- Artificial cognition in CPS
ics, many other technical challenges can be foreseen in the CPS—cog- CPS— CPS—cognitive societies CPS—
CPS vital domains—especially in the design, construction, nitive com- cognitive cognitive
and verification of CPS (Anthi et al. 2019). munities processes platforms

5C: AoA, CDN IoT IPv6

OoA,
6 From summary maps to conceptual VOA,
framework VEO,
VEP
MDMS CfAA WoT, SM, IoP ISP, MBDP
In this section, we use the hierarchical cascading method,
SoA BPS, DPP IoS KDoA
with categorical coding (Radanliev 2014), to build a concep-
DIS PHN SoS RtD
tual framework based on the findings in the summary map

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

792 AI & SOCIETY (2021) 36:783–796

The cascading hierarchy design in Table 2 represents correlated in a hierarchical framework in Fig. 2, and cor-
the first step in the conceptual framework design in Fig. 2. respond with the taxonomy (in Table 2). These are estab-
The similarities between Table 2 and Fig. 2 are clear. The lished models for decomposing and reverse engineering
differences between the new understanding of artificial design processes. The hierarchical cascading in Fig. 2
intelligence in CPS in Fig. 2 are also clear and very dis- explores the potential for automated and semi-automated
tinguishable from the existing understanding of artificial methods that could be applied to ascertain and accelerate
intelligence in CPS as seen in Fig. 1. Our approach for (and start to automate) the evolution of autonomous arti-
building the conceptual framework (Fig. 2) is based on ficial intelligence in CPS. The concepts and the hierarchi-
an extensive review of literature that included multiple cal structure in Fig. 2 originate from the summary map
systems, models, and methodologies from over 90 leading and the hierarchical cascading of actions in Table 1. The
articles on this topic. Concepts that reappeared in multiple abbreviations are present in Table 1, and categorisation
articles were selected as the most prominent, and the rela- of taxonomic imperatives is first presented in Table 2. In
tionships were recorded from each article. This enables a Fig. 2, we apply the findings to build a conceptual diagram
new approach to building the conceptual framework, based for visualising the updated CPS architecture.
on complex socio-economic, organisation goals and policy The conceptual diagram in Fig. 2 originates from the
issues that were identified in over 90 leading articles in re-evaluated five levels in the original 5C architecture,
this field, published in the past decade. but without the ‘cyber’ level—described in Fig. 1. The
The taxonomy of abbreviations in Table 2 was derived remaining four levels are updated with concepts emerg-
from the taxonomy of literature in Table 1, which catego- ing from recent literature on this topic, with a time span
rises the emerging concepts into a structure for artificial between 2010 and 2020. We identified from literature (in
cognition in CPS. The structure relates the cognition in the summary map Table 1) new and emerging concepts
CPS with IIoT, bringing together the IoP and IoS, along related to CPS architecture that are not included in the
with the process and transaction of IoT data. For example, current 5C architecture. In Fig. 2, we present a hierarchi-
the IoT data from DIS (see Tables 1 and 2 for definitions cal integration of these new and emerging concepts and
of abbreviations) connected to IoP and IoS, (representing present an updated 4C architecture—four levels of CPS
systems of systems) enhance the cyber risk avoidance with architecture.
real-time distribution and feedback directly from users and Since this review paper is built upon the notion of
markets. updating the existing 5C architecture in Fig. 1, we used
Thus, the evolution of cognition in CPS space adds a new the same conceptual order, but we integrated the improve-
perspective to the existing cyber risk avoidance mechanisms. ments found in recent literature. We identified a lot of
The inter-relationships between these elements are crucial new terms, springing up between 2010 and 2020, and we
for defining dynamic cyber risk analytics with real-time wanted to put them in conceptual order. We anticipated
probabilistic data. The current approaches taken for cyber this to be a required first step and a real service to future
risk analytics assume development of IoP and IoS and reli- studies aiming to build a diagnostic architecture. The
ability of IoT. A deeper understanding of the relationship conceptual diagram in Fig. 2 derives new understanding
between IoT and I4.0, following the categories presented in on why cognitive evolution in cyber physical systems is
Table 1, is required to develop a dynamic cyber risk analyt- inevitable and autonomous with the increased integration
ics structure. of connected devices (IoT). The hierarchical cascading in
Furthermore, Table 2 shows that cognitive CPS capa- Fig. 2 is designed using the grounded theory approach for
bilities are related to the integration of cyber physical relating emerging concepts. The emerging concepts identi-
capabilities into the industrial value chains. Hence, the fied in the literature review are first presented in the sum-
proposed structure for cognitive CPS uses principles of mary maps, and then taxonomic approach is used to relate
IoT and integrates network intelligence, providing conver- the categories and organise in a hierarchy of most closely
gence, orchestration, and visibility across otherwise dis- to most distantly related. Conceptual design is then used
parate systems. The proposed cognitive CPS also provides to cascade the hierarchy in a framework. The framework in
a structure for the operation and management of multi- Fig. 2 explores how automated and semi-automated meth-
ple CPS-related elements in the context of I4.0. Figure 2 ods are accelerating (starting to automate) the evolution of
shows the inter-relationship between different cognitive autonomous artificial intelligence in CPSs. The framework
CPS communities, processes, societies, and platforms. The in Fig. 2 represents a new mechanism and prototype of a
integration of cyber physical capabilities into the cognitive hierarchical structure that facilitates deeper understanding
CPS involves the integration of IoT, WoT, SM, IoP, and of interconnected concepts—both being crucial given that
IoS into SoS. With the use of Grounded Theory and Pugh- there is no direct reference in literature to artificial cogni-
controlled convergence, the categories (from Table 1) are tion in CPS and cyber risk analytics.

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 793

Fig. 3  Emerging CPS architec-

ture—4 levels

For final comparison, we present a visualisation in Fig. 3 support system for prioritising workload, with a single focus
that compares the emerging four levels of CPS architecture. on industrial processes. The new conceptual framework pre-
In Fig. 3, we can see that CPS as a concepts has evolved sented in Fig. 2 includes social machines, connected devices,
significantly since the 5 levels CPS architecture in Fig. 1. and knowledge developments, among new concepts such as
In Fig. 3, we can see how the taxonomy from the sum- internet of services and internet of people.
mary map in Table 1 has been integrated in the four levels The differences between the new framework in Fig. 2 and
CPS architecture. We can also compare Figs. 1 and 3 to the earlier framework as seen in Fig. 1 mean that AI is evolv-
visualise the differences between the CPS architectures from ing at a much faster rate than industrial understanding of this
2015 and 2020. Although Fig. 2 presents the same informa- process. The new framework in Fig. 2 captures the changes
tion, in a conceptual diagram, we designed Fig. 3 for easier in connected devices generating vast amounts of data, cap-
comparison. tured and stored in different heterogenous formats (e.g.
high-dimensional data, real-time data, translytic data, spa-
tiotemporal data). The new framework in Fig. 2 details the
7 Discussion process of how the new data are captured, stored, processed,
analysed, and used in near real-time, with low-latency. This
The updated four levels—CPS architecture in Fig. 2 offers is a very different process than our past understanding of
a new and important step in updating our understanding of CPS cognitive decision-making tasks, as seen in Fig. 1.
how CPS operate in 2020. Since the existing 5 levels CPS The main point of discussion from the new conceptual
architecture (see Fig. 1) is few years old, and there has been framework is that CPS are capable of much more than we
many changes in connected systems since its creation, we describe in existing frameworks on CPS cognition in Fig. 1.
considered this update timely and of relevance. We also With the availability of new types of data from IoT devices,
argue that with the rise in new IoT and IIoT, complex, cou- CPS are becoming more automated. For example, with the
pled and connected systems, such updates should occur at new translytic data, CPS can transact and analyse data. With
much faster intervals. This paper adopted the argument that spatiotemporal data, CPS can map the demand in real-time.
AI should ‘be programmed with a virtual consciousness and And with the complexities of high-dimensional data, CPS
conscience’ (Meissner 2020), because we are in the middle can understand the relationships between seemingly unre-
of a new AI revolution that is changing our economy and lated events and create new services and products. These
society. There are studies investigating whether AI can cre- new data streams are highly complex, and only AI can
ate ‘novel though’ (Fazi 2019). The mechanism in this paper analyse such data and derive predictions with low-latency.
describing how AI is evolving into CPS is based on grouping Hence the evolution of AI in CPS is inevitable, autonomous,
of future and present techniques and presenting the design and it is already happening.
process through a new hierarchical cascading design for a The arguments presented in this research are focused
conceptual framework. on understanding how the increased computational power
The conceptual framework in Fig. 2 details significant of connected devices, has created intrusive self-building
advancements over the past 5 years that can be seen in the CPS, that represent human-like performance, triggering the
most closely related framework on this topic in Fig. 1. For creation of collective intelligence. The aggregated knowl-
example, cognition in Fig. 1 is based solely on decision edge synthesised from recent literature, created a more

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

794 AI & SOCIETY (2021) 36:783–796

comprehensive understanding of the current evolution of Complex interconnected and coupled systems can evolve
AI in CPS. We should not wait another 5 years before a new automatically with the continuous technological upgrades
framework is designed to explain how AI is evolving with in existing CPS. The new hierarchical cascading framework
the emergence of new data formats, analysed with increased in this paper identifies approaches to model imperative
computational powers in connected devices. mechanisms within complex interconnected and coupled
systems. In important environments for AI, such as IoT, we
can model the connections and interdependencies between
8 Conclusion components to both external and internal IoT services and
CPS in summary map. The summary map identifies the
In this paper, we have produced a hierarchical cascading imperative categories for the evolution of artificial cognition
framework for analysing the evolution of AI decision-mak- in CPS. By applying established engineering design models,
ing in cyber physical systems. The significance of the new the summary map is advanced in a hierarchical structure
framework is the findings that (1) such evolution is auton- for artificial intelligence in CPS. However, more empirical
omous because of the increased integration of connected and philosophical research is needed on this topic before we
devices (IoT) in cyber physical systems; (2) such evolution can argue a comprehensive understanding on how artificial
is inevitable, because only AI can analyse the volume of data intelligence is evolving behind complex interconnected and
generated in low-latency, near real-time, hence, only AI can coupled systems.
create value from new and emerging forms of big data. Nev-
ertheless, we argue that the main value of the new 4 levels of
CPS architecture, is the perception of cyber-physical systems Compliance with ethical standards
as physical and human networks, where cognition emerges
from the cyber-physical ‘societies’ and ‘communities’ (see Conflict of interest On behalf of all authors, the corresponding author
states that there is no conflict nor competing interest.
Table 1). Our interpretation of CPS architecture perceives
cyber-physical systems as social machines, and we place Open Access This article is licensed under a Creative Commons Attri-
value in human interaction with such systems. In previous bution 4.0 International License, which permits use, sharing, adapta-
CPS architecture (Fig. 1), we can see that human interven- tion, distribution and reproduction in any medium or format, as long
tion is predominated in the configuration level and the CPS as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes
depend on human cognition and there is a separate layer for were made. The images or other third party material in this article are
‘cyber’. In our 4 levels of CPS, we integrated the ‘cyber’ included in the article’s Creative Commons licence, unless indicated
in all levels, and we argue that there is a value for artificial otherwise in a credit line to the material. If material is not included in
intelligence to learn from human–computer interactions. the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
Instead of relying only on feedback from connected devices, need to obtain permission directly from the copyright holder. To view a
in some scenarios, human input is of much greater value. copy of this licence, visit http://creat​iveco​mmons​.org/licen​ses/by/4.0/.
We have seen this in the current efforts to monitor a fast
spreading pandemic—Covid-19. All contact tracing apps are
based on human–computer input. Relying on computer data
alone, was considered too slow and ineffective. We use this References
as a final example to rationalise our argument for perceiving
future cyber-physical systems operating as social machines, Agyepong E, Cherdantseva Y, Reinecke P, Burnap P (2019) Challenges
obtaining input from humans and connected devices. and performance metrics for security operations center analysts: a
systematic review. J Cyber Secur Technol 4(1):1–28
To present this rationale, in the taxonomic methodology, Allen, Hamilton (2014) Cyber Power Index: findings and methodology.
we adapted the summary map method, for transparency and McLean, Virginia
justifications for concept selection and we used the litera- Anderson G (2016) The economic impact of technology infrastructure
ture review for decisions on the design of the hierarchical for smart manufacturing. US Department of Commerce, National
Institute of Standards and Technology
cascading. Out attempt in this paper was to contribute to the Anderson R, Moore T (2006) The economics of information security.
cultural practice of engineering discussion, on how artificial Sci AAAS 314(5799):610–613
intelligence is evolving over time, by presenting a snapshot Anthi E, Williams L, Slowinska M, Theodorakopoulos G, Burnap P
in time on this topic. It is evident that our analysis builds on (2019) A supervised intrusion detection system for smart home
IoT devices. IEEE Internet Things J 6(5):9042–9053
considerable assumptions and guesses based on findings and Anthonysamy P, Rashid A, Chitchyan R (2017) Privacy requirements:
taxonomic categorisation of existing literature. Our discus- present & future. In: 2017 IEEE/ACM 39th International confer-
sion in this paper points to the importance of understand- ence on software engineering: software engineering in society
ing the impact of connecting complex and coupled systems. track (ICSE-SEIS), pp 13–22

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

AI & SOCIETY (2021) 36:783–796 795

Balaji B, Al Faruque MA, Dutt N, Gupta R, Agarwal Y (2015) Models, Kang W, Kapitanova K, Son SH (2012) RDDS: a real-time data distri-
abstractions, and architectures. In: Proceedings of the 52nd annual bution service for cyber-physical systems. IEEE Trans Ind Infor-
design automation conference on—DAC’15, pp 1–6 matics 8(2):393–405
Banks J (2019) A perceived moral agency scale: development and Kirkpatrick K (2013) Software-defined networking. Commun ACM
validation of a metric for humans and social machines. Comput 56(9):16
Human Behav 90:363–371 Koch R, Rodosek G (2016) In: Proceedings of the 15th European con-
Barrett M, Marron J, Yan Pillitteri V, Boyens J, Witte G, Feldman L ference on cyber warfare and security: ECCWS 2016: hosted by
(2017) Draft NISTIR 8170, the cybersecurity framework: imple- Universität der Bundeswehr, Munich, 7–8 July 2016
mentation guidance for federal agencies, Maryland La HJ, Kim SD (2010) A service-based approach to designing cyber
Biener C, Eling M, Wirfs JH (2014) Insurability of cyber risk 1. The physical systems. In: 2010 IEEE/ACIS 9th international confer-
Geneva Association, Geneva, pp 1–4 ence on computer and information science, pp 895–900
Boyes H, Hallaq B, Cunningham J, Watson T (2018) The industrial Lee J, Kao HA, Yang S (2014) Service innovation and smart analytics
internet of things (IIoT): an analysis framework. Comput Ind for industry 4.0 and big data environment. Procedia CIRP 16:3–8
101:1–12 Lee J, Bagheri B, Kao HA (2015) A cyber-physical systems architec-
Brettel M, Fischer FG, Bendig D, Weber AR, Wolff B (2016) Enablers ture for industry 4.0-based manufacturing systems. Manufacturing
for self-optimizing production systems in the context of industrie Lett 3:18–23
4.0. Procedia CIRP 41:93–98 Lee B, Cooper R, Hands D, Coulton P (2019) Value creation for IoT:
Carruthers K (2016) Internet of things and beyond: cyber-physical sys- challenges and opportunities within the design and development
tems—IEEE internet of things. IEEE Internet of Things, Newslet- process. In: Living in the internet of things (IoT 2019). IET, living
ter, 2014 in the internet of things 2019, London, pp 1–8
Craggs B, Rashid A (2017) Smart cyber-physical systems: beyond Lee B, Cooper R, Hands D, Coulton P (2019) Design drivers: a critical
usable security to security ergonomics by design. In: 2017 IEEE/ enabler to meditate value over the NPD process within Internet
ACM 3rd international workshop on software engineering for of Things. In: 4d conference proceedings: meanings of design
smart cyber-physical systems (SEsCPS), pp 22–25 in the next era. DML (Design Management Lab), Ritsumeikan
de Fine Licht K, de Fine Licht J (2020) Artificial intelligence, transpar- University, Osaka, pp 96–107
ency, and public decision-making. AI Soc 1–10 (forthcoming) Leitão P, Colombo AW, Karnouskos S (2016) Industrial automation
De Roure D, Hooper C, Page K, Tarte S, Willcox P (2015) Observing based on cyber-physical systems technologies: prototype imple-
social machines part 2: how to observe? In: Proceedings of the mentations and challenges. Comput Ind 81:11–25
2015 ACM web science conference, pp 1–5 Lewis D, Brigder D (2004) Market researchers make increasing use of
De Roure D, Page KR, Radanliev P, Van Kleek M (2019) Complex brain imaging. Adv Clin Neurosci Rehabil 5(3):36–37
coupling in cyber-physical systems and the threats of fake data. In: Liberati N (2020) The Borg–eye and the We–I. The production of
Living in the internet of things (IoT 2019) 2019 page, p 11 (6 pp) a collective living body through wearable computers. AI Soc
Dillon TS, Zhuge H, Wu C, Singh J, Chang E (2011) Web-of-things 35(1):39–49
framework for cyber-physical systems. Concurr Comput Pract Exp Marwedel P, Engel M (2016) Cyber-physical systems: opportunities,
23(9):905–923 challenges and (some) solutions. Springer International Publish-
DiMase D, Collier ZA, Heffner K, Linkov I (2015) Systems engineer- ing, Cham, pp 1–30
ing framework for cyber physical security and resilience. Environ Meissner G (2020) Artificial intelligence: consciousness and con-
Syst Decis 35(2):291–300 science. AI Soc 35(1):225–235
Dombrowski U, Wagner T (2014) Mental strain as field of action in the Milano S, Taddeo M, Floridi L (2020) Recommender systems and
4th industrial revolution. Procedia CIRP 17:100–105 their ethical challenges. AI Soc 1–11 (forthcoming)
Drath R, Horch A (2014) Industrie 4.0: hit or hype? [Industry Forum]. Nicolescu R, Huth M, Radanliev P, De Roure D (2018) Mapping the
IEEE Ind Electron Mag 8(2):56–58 values of IoT. J Inf Technol 33(4):345–360
Fazi M (2019) Beatrice, “Can a machine think (anything new)? Auto- Niggemann O, Biswas G, Kinnebrew JS, Khorasgani H, Volgmann
mation beyond simulation”. AI Soc 34(4):813–824 S, Bunte A (2015) Data-driven monitoring of cyber-physical
Giordano A, Spezzano G, Vinci A (2016) A smart platform for large- systems leveraging on big data and the internet-of-things for
scale cyber-physical systems. Springer International Publishing, diagnosis and control. In: International workshop on the prin-
Cham, pp 115–134 ciples of diagnosis (DX), pp 185–192
Glaser BG, Strauss AL (1967) The discovery of grounded theory: strat- NIST (2018) NIST Version 1.1, National Institute of Standards and
egies for qualitative research. Routledge, Abingdon Technology, U.S. Department of Commerce. [Online]. https​://
Gordon LA, Loeb MP (2002) The economics of information security www.nist.gov/news-event​s/news/2018/04/nist-relea​ses-versi​on-
investment. ACM Trans Inf Syst Secur 5(4):438–457 11-its-popul​ar-cyber​secur​ity-frame​work
Hermann M, Pentek T, Otto B (2016) Design principles for industrie O’Hara K (2020) The contradictions of digital modernity. AI Soc
4.0 scenarios. In: 2016 49th Hawaii international conference on 35(1):197–208
system sciences (HICSS), pp 3928–3937 Ouyang J, Lin S, Jiang S, Hou Z, Wang Y, Wang Y, Ouyang J, Hou
Hussain F (2017) Internet of things: Building blocks and business mod- Z, Wang Y, Wang Y (2014) SDF: software-defined flash for
els, (No. 978-3, pp 319-55404) Springer International Publishing web-scale internet storage systems. In Proceedings of the 19th
Jensen JC, Chang DH, Lee EA (2011) A model-based design meth- international conference on Architectural support for program-
odology for cyber-physical systems. In: 2011 7th international ming languages and operating systems—ASPLOS ’14, vol 42,
wireless communications and mobile computing conference, pp no. 1, pp 471–484
1666–1671 Pan M, Sikorski J, Kastner CA, Akroyd J, Mosbach S, Lau R, Kraft
Kambatla K, Kollias G, Kumar V, Grama A (2014) Trends in big data M (2015) Applying industry 4.0 to the Jurong Island Eco-indus-
analytics. J Parallel Distrib Comput 74:2561–2573 trial Park. Energy Procedia 75:1536–1541
Kammerer F (2020) Self-building technologies AI Soc Posada J, Toro C, Barandiaran I, Oyarzun D, Stricker D, de Amicis
1–15 (forthcoming) R, Pinto EB, Vallarino I (2015) Visual computing as a key ena-
bling technology for industrie 4.0 and industrial internet. IEEE
Comput Graph Appl 35(2):26–40

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

796 AI & SOCIETY (2021) 36:783–796

Radanliev P (2014) A conceptual framework for supply chain sys- Stojmenovic I (2014) Machine-to-machine communications with in-
tems architecture and integration design based on practice and network data aggregation, processing, and actuation for large-scale
theory in the North Wales slate mining industry. British Library cyber-physical systems. IEEE Internet Things J 1(2):122–128
Radanliev P, De Roure D, Nicolescu R, Huth M, Montalvo RM, Sun R (2020) Potential of full human–machine symbiosis through truly
Cannady S, Burnap P (2018) Future developments in cyber risk intelligent cognitive systems. AI Soc 35(1):17–28
assessment for the internet of things. Comput Ind 102:14–22 Tan Y, Goddard S, Pérez LC (2008) A prototype architecture for cyber-
Radanliev P, De Roure D, Page K, Nurse JR, Montalvo Mantilla R, physical systems. ACM Sigbed Review 5(1):1–2
Santos O, Maddox LT, Burnap P (2020) Cyber risk at the edge: Tanczer LM, Steenmans I, Elsden M, Blackstock J, Carr M (2018)
current and future trends on cyber risk analytics and artificial Emerging risks in the IoT ecosystem: Who’s afraid of the big bad
intelligence in the industrial internet of things and industry 4.0 smart fridge? In: Living in the internet of things: cybersecurity
supply chains. Cybersecurity 3(13):1–21 of the IoT, p 33 (9 pp)
Radanliev P, De Roure DC, Nurse JR, Montalvo RM, Cannady S, Thramboulidis K (2015) A cyber–physical system-based approach for
Santos O, Madox LT, Maple C (2020) Future developments in industrial automation systems. Comput Ind 72:92–102
standardisation of cyber risk in the Internet of Things (IoT). SN Wahlster W, Helbig J, Hellinger A, Stumpf MAV, Blasco J, Galloway
Appl Sci 169(2):1–16 H, Gestaltung H (2013) Recommendations for implementing the
Rajkumar R, Lee I, Sha L, Stankovic J (2010) Cyber-physical sys- strategic initiative INDUSTRIE 4.0. Federal Ministry of Educa-
tems: the next computing revolution. In: Proceedings of the 47th tionand Research
design automation conference on—DAC’10, p 731 Wan J, Yan H, Li D, Zhou K, Zeng L (2013) Cyber-physical systems
Ribeiro L, Barata J, Ferreira J (2010) An agent-based interaction- for optimal energy management scheme of autonomous electric
oriented shop floor to support emergent diagnosis. In: 2010 8th vehicle. Comput J 56(8):947–956
IEEE international conference on industrial informatics, pp Wan J, Cai H, Zhou K (2015) Industrie 4.0: enabling technologies.
189–194 In: Proceedings of 2015 international conference on intelligent
Ringert JO, Rumpe B, Wortmann A (2015) Architecture and behav- computing and internet of things, pp 135–140
ior modeling of cyber-physical systems with MontiArcAutoma- Wang L, Wang XV, Gao L, Váncza J (2014) A cloud-based approach
ton. arXiv preprint arXiv​:1509.04505​ for WEEE remanufacturing. CIRP Ann Manuf Technol
Rodewald G, Gus (2005) Aligning information security investments 63(1):409–412
with a firm’s risk tolerance. In: Proceedings of the 2nd annual Wang L, Törngren M, Onori M (2015) Current status and advance-
conference on information security curriculum development— ment of cyber-physical systems in manufacturing. J Manuf Syst
InfoSecCD ’05, p 139 37:517–527
Roumani MA, Fung CC, Rai S, Xie H (2016) Value analysis of cyber Wang S, Wan J, Li D, Zhang C (2016) Implementing smart factory of
security based on attack types. ITMSOC Trans Innov Bus Eng industrie 4.0: an outlook. Int J Distrib Sens Netwk 12(1):1–10
1:34–39 Wark T, Corke P, Sikka P, Klingbeil L, Guo Y, Crossman C, Valen-
Ruan K (2017) Introducing cybernomics: a unifying economic frame- cia P, Bishop-Hurley G (2007) Transforming agriculture through
work for measuring cyber risk. Comput Secur 65:77–89 pervasive wireless sensor networks. IEEE Pervasive Comput
Ruffle SJ, Bowman G, Caccioli F, Coburn AW, Kelly S, Leslie B, Ralph 6(2):50–57
D (2014) Stress test scenario: sybil logic bomb cyber catastrophe. Weyer S, Schmitt M, Ohmer M, Gorecky D (2015) Towards indus-
Cambridge Risk Framew Ser Cent Risk Stud Univ Cambridge try 4.0—standardization as the crucial challenge for highly
Rutter T (2015) The rise of nudge—the unit helping politicians to modular, multi-vendor production systems. IFAC-PapersOnLine
fathom human behavior. Guard 7(23):2015 48(3):579–584
Shackelford SJ (2016) Protecting intellectual property and privacy in World Economic Forum (2015) Partnering for cyber resilience towards
the digital age: the use of national cybersecurity strategies to miti- the quantification of cyber threats, Geneva
gate cyber risk. Chapman Law Rev 19:412–445 Zhu Q, Rieger C, Basar T (2011) A hierarchical security architecture
Shafiq SI, Sanin C, Szczerbicki E, Toro C (2015) Virtual engineer- for cyber-physical systems. In: 2011 4th international symposium
ing object/virtual engineering process: a specialized form of on resilient control systems, pp 15–20
cyber physical system for industrie 4.0. Procedia Comput Sci
60:1146–1155 Publisher’s Note Springer Nature remains neutral with regard to
Shi J, Wan J, Yan H, Suo H (2011) A survey of cyber-physical systems. jurisdictional claims in published maps and institutional affiliations.
In: 2011 International conference on wireless communications
and signal processing (WCSP), pp 1–6
Stock T, Seliger G (2016) Opportunities of sustainable manufacturing
in industry 4.0. Procedia CIRP 40:536–541

13

Content courtesy of Springer Nature, terms of use apply. Rights reserved.

Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:

1. use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
2. use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
4. use bots or other automated methods to access the content or redirect messages
5. override any security feature or exclusionary protocol; or
6. share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at

[email protected]