Digital engineering transformation with trustworthy AI towards industry 4.0: emerging paradigm shifts

4.1 Trustworthiness of systems

Trustworthiness of a system is defined in (Schneider, 1999) as “assurance that a system deserves to be trusted—that it will perform as expected despite environmental disruptions, human and operator error, hostile attacks, and design and implementation errors.” In short (Avizienis, Laprie, Randell, & Landwehr, 2004), systems trustworthiness is the “assurance that a system will perform as expected.” Trustworthiness is a relative concept and is dependent on what is expected. Trustworthiness is often characterized as a collection of the expected properties of a system to be trusted, as illustrated in Fig. 5. The “assurance” can be achieved by various trust mechanisms implemented in engineering systems and in societal systems such as audits, certifications, regulations, and law enforcement.

OPEN IN VIEWER

Fig. 5

The evolving trustworthiness properties as the evolution of systems with technologies and the sources of concerns.

In the following, we discuss the expected properties of trustworthiness of different generations of engineering systems and the sources of the concerns or threats to trustworthiness. Traditional systems are typically stand-alone systems (a physical system, a computer system, or a system embedded with computer systems) or just connected logically and internally within an organization rather than the Internet. As shown in the bottom part of Fig. 5, the trustworthiness properties concerned with those traditional systems are mainly reliability, maintainability, availability, integrity, robustness, safety, and usability. Those properties together are referred to as dependability. The sources of concerns or the threats to the trustworthiness properties are mainly non-malicious faults caused by (1) errors introduced in systems design and development; (2) errors introduced in manufacturing and maintenance; (3) errors happened in system operations; (4) environmental disruptions, including natural disasters, environmental pollution, and gradual but irreversible environment change associated with climate change.

In the middle part of Fig. 5, enabled by Internet and wireless mobile communication, the networked industrial systems evolve to cyber-physical systems (CPS) (Lee, 2008; NSF, 2008; Rajkumar, Lee, Sha, & Stankovic, 2010), which use networked computing components to monitor, coordinate, control, and integrate physical systems and processes. The connectedness of systems naturally led to great challenges with respect to security and privacy, and correspondingly they become the central theme of systems trustworthiness (Avizienis et al., 2004; Schneider, 1999). The threats to security and privacy are obviously from various malicious attacks and incompetent security defence.

As illustrated in the upper part of Fig. 5, trustworthiness for the systems in the 4IR involves concerns with trustworthy AI, pervasive security, and technical dependability. The issues in an AI system could trigger security issues and systems dependability issues; of course, security issues could also trigger systems dependability issues. The sources causing the systems trustworthiness issues are in three folds:

The pervasive security issues caused by the pervasive connectedness facilitated by IoT: the pervasive connectedness facilitated by IoT pervasively extends the security and privacy concerns to every corner of the world where humans work and live. Without assured technologies and societal mechanisms, the Internet of Things can turn into “the Internet of Risky Things” (S. Smith, 2017), even the Internet of dangerous things. The threats to security and privacy in 4IR include malicious attacks and incompetent security defence as before but with a much larger attack/exposure surface. Additionally, the threats also include certain relevant societal mechanisms. For example, as Smith pointed out (S. Smith, 2017), some smart devices will last longer than their manufacturers and suppliers, and who will keep updating the systems or their embedded computing components? Without updates, a digital system will be vulnerable in the complex digital operating environment.

In the last decade, AI achieved remarkable advances. Just to name a few, AlphaGo (Silver et al., 2018), using deep neural networks-based reinforcement learning, won the world number one player of Go game, which is regarded as the most challenging board game for computers to win. AlphaGo’s achievement marks a new milestone of machine intelligence towards superhuman intelligence on specific tasks. Although it is still a pilot project and has no broader adoption yet, Waymo One, the commercial taxi service operating with level 4 autonomous vehicles, has been offered on the streets in a city (Waymo One, 2020). This service marks a new level of integrated intelligent capabilities of an AI system on a type of complex tasks which previously only humans could do. Another remarkable advance is Generative Adversarial Networks (GANs) (Creswell et al., 2018; Goodfellow et al., 2020). A GAN consists of a pair of deep neural networks, a generator, and a discriminator; together, they can generate realistic images and other contents as requested (see examples at https://thispersondoesnotexist.com/). GANs lead to many potential applications but also open the door for deepfake.

For many decades, the AI community’s focus has been on extending machine intelligence capabilities. Although the advances of AI today are still domain-specific (or narrow AI) rather than general AI, the remarkable achievements, the disruptive impacts, and the fast growth of applications of AI systems triggered concerns and research on trustworthy AI systems. The concerns are mainly from two perspectives. (1) Ethical AI, focusing on social effects, mainly ethical considerations, essentially, the profound effects AI systems bring to humans, human groups, and human society, and advocating using AI to benefit humans as a central principle. (2) Reliable AI, focusing on the technical performance, or dependability, concerning the competency of AI systems on technical matters. There is a broad range of expected properties of trustworthiness in AI systems.

Ethical AI is an emerging interdisciplinary research field, gathering together researchers from a wide range of areas, including computer science, philosophy, sociology, anthropology, public policy, law, and others. Some thought leaders and researchers conducted pioneering work in the field. The Association for the Advancement of Artificial Intelligence (AAAI) organized a panel of leading AI researchers that conducted “Asilomar Study of Long-Term AI Futures” in 2008. The study explored a wide range of topics on societal impacts and guidance of AI research (AAAI Presidential Panel on Long-Term AI Future, 2009). The study assessed concerns and perspectives about disruptive outcomes of superhuman intelligence, explored ethical and legal issues associated with autonomous systems, and identified near-term AI research challenges and opportunities, including enhancing people’s privacy, enhancing human-AI collaboration and interaction, making ML and reasoning transparent to people, and preventing using AI for malevolent purpose. Continuing the effort to guide AI research for good, the AI100 2016 report (Stone et al., 2016) addressed that AI research is “shifting from simply building systems that are intelligent to building intelligent systems that are human-aware and trustworthy.” The report collectively presents the view and prospects of a panel of experts on the opportunities and challenges of AI and policy recommendations in eight selected AI application domains targeting people’s lives in a typical North American city in 2030. They addressed challenges regarding how to build safe and reliable systems, gain public trust concerning safety, security, and privacy, make AI systems behave ethically and overcome bias, and how AI systems smoothly interact with humans. In the long term, ethical AI concerns the fundamental relations between AI systems and humans. There are concerns about “technological singularity” or “intelligence explosion,” partially reflected by the dystopia depicted in fiction. Experts in the AI field believe those radical outcomes remain fictional and are not immediate threats; well, some thought leaders suggest “avoid strong assumptions regarding upper limits on future AI capabilities” (Future of Life Institute, 2017). Given AI’s profound impacts on human society, it is necessary to review AI systems’ purposes, usages, and impacts and create principles and regulations for guiding AI research and development to avoid harming humans and human society. The 23 Asilomar principles (Future of Life Institute, 2017) and the EU HELG ethical AI guide (EU AI HLEG, 2019) reflect a broad range of concerns about ethical issues of AI systems. EU AI HELG defined trustworthy AI as three components: lawful AI, ethical AI, and robust AI. The guide proposed four ethical principles: (1) Respect for human autonomy; (2) Prevention of harm; (3) Fairness; (4) Explicability, covering transparency, auditability, traceability, and explainability. On the ground of these four principles, the guide further proposed a list of seven requirements: (1) Human agency and oversight; (2) Technical robustness and safety, including security, accuracy, reliability, and reproducibility; (3) Privacy and data governance; (4) Transparency, including traceability, explainability, and communication; (5) Diversity, non-discrimination and fairness; (6) Societal and environmental wellbeing; (7) Accountability. The EU guide also discussed technical and non-technical methods to implement the requirements and proposed a list of questions used to assess the trustworthiness of AI systems.

The ethics of AI studies what is right/wrong regarding the purpose and usages of AI, based on the profound effects AI systems bring to humans and human society. The central principle of ethical AI is about using AI to benefit humans and human society. In the upper part of Fig. 5, ethicality is about the extent to which an AI system complies with this principle.

As addressed by the AI100 2021 report (Littman et al., 2021), “AI systems and humans have complementary strengths;” thus, “combined, they can accomplish more than either alone.” Technically, it remains a challenge regarding how to team up humans and AI systems effectively. For the fundamental ethical principles of trustworthy AI (EU AI HLEG, 2019), no doubt human-AI teaming is the right direction to go, not only for maximizing capability and performance but also for ethical consideration. Cooperability is about the extent to which an AI system facilitates and supports human-machine teaming for complex problem-solving, including the channels or methods to enhance interactions and collaborations between humans and machines. Cooperability covers controllability; the latter is about the ability of an AI system that allows humans to monitor and control the system.

Fairness is about whether an AI system fairly treats people of different groups regarding race, gender, age, cultural background, and others. Fairness received much attention in recent years when AI systems started being used for some life-changing scenarios, such as hiring decisions, financial credit evaluation, and judicial decisions (Mehrabi, et al., 2021). Fairness is a highly challenging topic for complex human societal reasons. Technically, fairness can be treated as bias-free. The recent deep learning progress reflected by GANs can be used to create real-like samples for balanced data in ML.

Accountability is the availability and integrity of the identity of an entity that performed an operation in the AI system of concern. In simple, it is who (human operators or autonomous components in an AI system) did what and when and the responsibility for that. In the scenarios of a security incident, an accident, an error, or a system failure, accountability helps to identify the causes, make responsibility clear, and avoid future mistakes.

Transparency is about the extent to which how the AI system operates is transparent to various stakeholders, such as operators, business partners, auditors, and users. Transparency is a basis for ensuring ethicality, fairness, and privacy and facilitating controllability and cooperability.

Explainability is the ability to explain the outcomes and processes of an AI system to humans. The major challenge is that in ML, neural networks have low-level coding for feature representation which is inherently hard to be explained in high-level knowledge. A very large number of parameters and complex structures of deep neural networks make the interpretability further harder.

Traceability is about the ability to collect and document the provenance of the data used and produced, the models used and trained, and the operating processes of an AI system. Traceability is the basis for transparency and supports explainability.

Reproducibility is about whether a model instance can be rebuilt with the same AI algorithm and data and whether an experiment or, more generally, a process running with an AI model can be reproduced. Reproducibility is essential to science and engineering.

Reliable AI reflects people’s concern about systems’ technical performance when more and more AI systems are used or embedded in a large engineering system in the 4IR. Naturally, this aspect of concern brings the focused attention partially back to more classical trustworthiness properties of engineering systems but with a focus on AI systems or AI components and their impact on the larger systems. This article uses the term “dependability” to cover all expected properties (for trustworthiness) on the technical performance aspect. The concept of dependability here is broader than what was defined in (Avizienis et al., 2004). Reliability is the probability of a system functioning without failure for a given period of time. Maintainability is about how easy to make a system maintain healthy, updatable, and upgradeable. Resilience is about the ability of a system to restore it to a working state when it is damaged in situations such as natural disaster events or cyber-attacks. Safety is about how safe to humans a system is. The broad use of AI components in engineering systems makes safety a significant concern. Accuracy is the measure of errors made by a model or system; in the context of ML, it is critical to have high accuracy for new data beyond the data used for training the model. In Systems Engineering, robustness is defined as “the degree to which a system or component can function correctly in the presence of invalid inputs or stressful environmental conditions”(ISO/IEC/IEEE, 2010). In ML, robustness is about the stable outcomes in the presence of perturbations in inputs and could be measured with sensitivity. In deep learning, lack of robustness is a critical cause for the possible deepfake by GANs. The general meaning of robustness is similar to resilience. In robustness, the perturbations are on a small scale. Usability is how easy a system is to be used, and usability is beyond human-machine interaction. Poor usability can lead to failures in many other properties in modern systems, including security and safety.

Security is defined as the well-known definition of the CIA triad: Confidentiality (Prevention of unauthorized access to the protected resources or disclosure of the protected information), Integrity (absence of unauthorized alterations), and Availability (Readiness for correct services for authorized users). Some other interesting security properties can be defined on top of the CIA triad. For example, authenticity is the integrity of information content and its provenance (origin). Obviously, in the digital and connected environment, failures in security have broad impacts and can compromise other trustworthiness properties. Privacy is another big concern in AI. Since data is the fuel for AI systems, privacy concerns about whether the data gathering, holding, processing, usage, sharing, and governance respect people’s privacy.

As illustrated by Fig. 5, the trustworthiness of AI systems is the new collection of concerns when the systems evolve into Industry 4.0 systems. Interestingly, on the other hand, the digitalization of engineering artifacts, processes, and enterprises in the 4IR could support achieving the expected trustworthiness properties of AI systems.

In digital engineering transformation, it is an excellent opportunity for the engineering design community at large to bring the new capabilities of AI and the trustworthy AI principles together in various engineering systems design for human society to leverage the power of AI and at the same time to avoid or minimize the potential negative impacts.

5.1 What is the pattern of paradigm shifts?

As discussed in section 2, digitalization in the 4IR is leading to paradigm shifts in almost every aspect of our society. What are the emerging new engineering paradigms? Engineering uses scientific principles to design, build, and operate engineering systems for solving real-world problems or meeting humans’ needs in a specific field, thus essentially sharing the paradigms in scientific research. A general pattern of scientific paradigm shifts can be illustrated as the following Fig. 6.

OPEN IN VIEWER

Fig. 6

A view of paradigm shifts, based on Kuhn’s structure of scientific revolutions (Kuhn, 1962).

According to Kuhn’s structure of scientific revolutions (Kuhn, 1962), in the period of “normal research” or “normal science,” a scientific community uses a dominant paradigm to conduct research and produce the main body of knowledge in that field. Then, the dominant paradigm may have crises for its deficiencies or limitations facing the new observations and/or new problems coming from the changing environment. To battle the crises and meet the new need(s), the disciple enters a period of “extraordinary research” (as named by Kuhn). In this period, new concepts, models, tools, methods, among others, will be created. If the new disciplinary components are incremental and can be integrated into the current paradigm, the paradigm will be updated, thus being a scientific evolution. Otherwise, the new components developed in “extraordinary research,” possibly together with the components from other disciplines, will contribute to the development of a new paradigm in a period of “pre-paradigm” (again, named by Kohn). In the “pre-paradigm” period, one or multiple paradigms will be formed and compete. Finally, the most accepted paradigm(s) will become the discipline’s new dominant paradigm(s). This “paradigm shift” (Kohn) is a scientific revolution. After the transformation, the discipline enters again “normal research” period and starts a new life cycle.

The above discussed pattern is revealed in the context of scientific research. It appears to be general, as we treat a scientific paradigm as a knowledge system that evolves in the environment.

Jim Gray had a vision that scientific research is shifting to the fourth paradigm (data-intensive paradigm) after the empirical, theoretical, and computational paradigms (Gray, 2007; Hey, Tansley, & Tolle, 2009), as illustrated in Fig. 7. The empirical paradigm is characterized as finding and describing patterns based on the observations of real-world phenomena, and the theoretical paradigm is characterized as building theories on mathematical models. The computational paradigm uses computer simulations to tackle the difficulty when the theoretical models become too complex to derive and prove propositions.

OPEN IN VIEWER

Fig. 7

Four Paradigms in Scientific Research, Slide from (Gray, 2007; Hey et al., 2009).

The fourth paradigm reflects the scientific research in the new research environment with the big data produced from instruments and simulations, high demands of computing for knowledge discovery from big data, and emerging demands of data publishing and research work reproducibility. In the fourth paradigm, scientific research significantly focuses on knowledge discovery from big data and reproducible models and datasets sharing.

It is essential to point out that a new paradigm does not mean entirely replacing existing paradigms. Instead, a new paradigm is established on the ground of existing paradigms. A new paradigm represents the focus shift and the new approach to address new problems in a new environment. Theoretical models are built on the ground of observations, and computational simulations are based on theoretical models. The fourth paradigm enhances and unifies the empirical, theoretical, and computational paradigms, on the ground of an unprecedented big data environment and the new technology for knowledge discovery from big data.

Inspired by Jim Gray’s fourth paradigm, with the fast growth of big data in the last decade, data science has emerged as a new discipline. A NIST publication (NIST Big Data Public Working Group, 2019) pointed out that “data science is the fourth paradigm of science.” However, it has been heavily focused on knowledge discovery from data. From the library and information science perspective, data curation and knowledge curation are among data science topics. Still, the needs emerging from the fourth paradigm are beyond just traditional curation and go much further into knowledge and data sharing with traceability, explainability, accountability, reproducibility, and interoperability.

By the fourth paradigm, data-intensive science/engineering will emerge in various disciplines of science/engineering. (Data-intensive engineering is the manifestation of data science in the engineering field.) In this paradigm shift, data science is the common core, shared by many domain-specific science/engineering disciplines that need knowledge discovery from big data and knowledge sharing. The combination of data science (the fourth paradigm) with each specific discipline will form many domain-specific data-intensive science/engineering disciplines.

“Science is the systematic description of phenomena” (Richards, 1928). Science focuses on discovering the essential laws of nature. Engineering is “the application of science to the optimum conversion of the resources of nature to the uses of humankind” (R J. Smith, 2022). On the one hand, there is an intersection between science and engineering, as engineering includes scientific research for discovering applied knowledge for engineering purposes. From this perspective, data-intensive engineering is the manifestation of the data-intensive paradigm of science in the digital and connected operating environment of the 4IR.

On the other hand, engineering activities are beyond discovering new knowledge and focus more on the design, manufacturing (or construction), operating, and support of useful systems for human society, thus having much more complex interactions and relations with other systems in the operating environment, including human stakeholders. Given this complexity of engineering, the disruptive digital technologies, and the associated digital, connected, and smart environment in the 4IR, how should we conduct engineering? What are the new engineering paradigms in the 4IR?

No doubt, the data-intensive paradigm of scientific research is a crucial one. It includes both aspects of knowledge discovery from data and knowledge sharing (including data). The current “data science” central theme has focused more on knowledge discovery from data. While “digital engineering” in the US DoD’s vision is more from the perspectives of model-based engineering and sharing of models and data. Interestingly, model-based engineering is an effort, like the development of theoretical and computational paradigms in science, to introduce models in engineering workflows, including traditionally informal activities such as requirement elicitation, requirement representation, system concept modelling, system design, among others. Certainly, modelling needs to be based on scientific knowledge as much as possible. Also, the availability and richness of data in the engineering environment and the fast progress of machine learning make it a powerful way to build models from data, leading to data-intensive engineering.

However, given the complex interactions and relations between engineering, human society, and the operating environment, the data-intensive paradigm (that focuses on scientific discovery) alone is insufficient for tackling the complexity of engineering. The need for engineering paradigm shifts in the 4IR is driven by the engineering environment, the disruptive digital technologies, the associated higher social-economic needs, and the new challenging problems, such as the trust issues of AI systems (as discussed in Section 4). In the landscape of the 4IR, the digital engineering transformation needs new concepts, models, tools, methods, theories, methodologies, technologies, and standards.

As discussed in section 2, the 4IR has three aspects of paradigm shifts: digitalizing everything of interest, provisioning ubiquitous machine intelligence, and big data and machine intelligence-centred business processes. On the other hand, digitalization and ubiquitous machine intelligence also triggered broad concerns and potential issues of trustworthiness, as discussed in section 4. To put the above-discussed pieces together, we could have a relatively clear big picture in the direction of Industry 4.0, as illustrated in Fig. 8, in which we need three interdependent essential building blocks for the digital transformation of engineering and industries:

Digitalization of engineering

OPEN IN VIEWER

Fig. 8

Three essential building blocks for digital engineering towards Industry 4.0: digitalization of engineering, leveraging ubiquitous machine intelligence, and building digital trust & security.

Digitalization is a foundation to realize ubiquitous machine intelligence and needs to support digital trust. For this consideration, to digitalize a thing of interest (which could be an object or a process), we need the following four essential components:

Digital representation of the thing of interest in a standard form with well-defined semantics to make it universally accessible by different types of machines on different platforms.

For the digitalization of engineering, basically, we need (i) to digitalize engineering artifacts, engineering processes, and enterprises; (ii) enable the sharing and interoperability of digitalized artifacts across the engineering lifecycle; (iii) to develop digital model-based engineering. Virtualization is a view from the perspective of operating with digitalized artifacts and processes.

Digitalization needs to be realized with digital trust in design and supported by ubiquitous machine intelligence among digital technologies. Naturally, in this direction, what are enabling technologies for digitally modelling engineering artifacts, processes, and enterprises? Semantic web technology (a branch of AI) (Berners-Lee et al., 2001), RFID, IoT, and blockchain are essential. Also, directly related, intensive research on enterprise modeling and enterprise integration has been conducted since the 1990s and can be used to support model sharing across boundaries (Chen, Doumeingts, & Vernadat, 2008; Fox & Gruninger, 1998; Fox & Huang, 2005; Goranson, 2002; Vernadat, 2020). Enterprise modelling combined with digital identity management could form a basis for digitalizing engineering artifacts. Ontologies play an essential role in the digitalization of engineering with respect to artifacts, entities, organizations, activities and processes, among others (Ahmed, Kim, & Wallace, 2006; Demoly, Kim, & Horváth, 2019; Dimassi et al., 2021; Fox, Barbuceanu, Gruninger, & Lin, 1998; Gruninger & Fox, 1995; H. M. Kim, Fox, & Grüninger, 1999; K.-Y. Kim, Manley, & Yang, 2006; Sirin, Coatanéa, Yannou, & Landel, 2013).

The realization of ubiquitous machine intelligence, the second aspect of paradigm shifts in the 4IR, is on the ground of digitalization. With digitalized engineering artifacts and processes, we can apply ubiquitous machine intelligence in engineering to design, build, and support smart products, provide smart services, build trustworthy supply chains, and conduct engineering intelligently. Consider the following examples. Machine learning has been broadly applied in additive manufacturing (Jin, Zhang, Demir, & Gu, 2020; Meng et al., 2020; Wang, Tan, Tor, & Lim, 2020). In the digital, smart, and connected environment, unprecedented big data provide rich information for using AI in engineering operations. Condition-based maintenance or predictive maintenance are typical scenarios where AI&ML can be leveraged (Black, Richmond, & Kolios, 2021; Carvalho et al., 2019). Engineering involves many sequential decision making; with rich information about engineering systems’ status, reinforcement learning can be used to learn from simulations, experiments, routine operations, and generally experience for optimization, such as in mesh generation (Pan, Huang, Cheng, & Zeng, 2022), in manufacturing (Su, Huang, Adams, Chang, & Beling, 2022), for engineering design (Dworschak, Dietze, Wittmann, Schleich, & Wartzack, 2022). Machine learning can also be applied to even traditionally labour-intensive and time-consuming requirement elicitation process (Cheligeer et al., 2022; Mokammel et al., 2018). Essentially, engineering systems design starts with the environment (Zeng, 2004, 2015, 2020). In the new engineering environment of the 4IR, many engineering systems become cyber-physical-social smart systems, and it is a challenging issue regarding how to design such systems (Horváth, 2022; Tavčar & Horvath, 2018).

Digitalization brings to us not only advantages but also new issues, most significantly, trust and security issues with digital artifacts and those digital technologies. Some examples of those issues include how do we ensure the authenticity and integrity of digital artifacts, what are policies regarding who can access what, when to access and where to access, and how do we ensure traceability, transparency, accountability, and reproducibility, among others. We must develop proper digital mechanisms to address those issues. How could we achieve secure information sharing in digital engineering? As an example, a security policy integrating role-based access control and attribute-based access control (including security classification-based mandatory access control) (Huang, Nicol, Bobba, & Huh, 2012; Jin, Sandhu, & Krishnan, 2012; Servos & Osborn, 2017) can help. Digital identity management and digital trust mechanisms (Huang & Nicol, 2013), particularly blockchain (Liu et al., 2020; Zheng et al., 2020), can support trust management of digital artifacts. Digital archives are critical to ensure data integrity and proof of existence (Vigil, Cabarcas, Buchmann, & Huang, 2013), and blockchain facilitates a new approach to distributed digital archives. Scientific computing integrity (Huang, 2018; Peisert, Cybenko, & Jajodia, 2015) can be further developed to support trusted engineering workflows.

Digital Engineering is an emerging form of engineering in the digital revolution. Many issues and questions there need to be researched. Just name a few, in digitalization, should the standard forms for digital representation and augmentation be supported by a centralized standardization (such as creating international standards) or a distributed evolutionary standardization (such as many ontologies competing to be standards at a fine-grained level and evolving gradually and naturally)? How do we achieve trustworthy AI systems in 4IR? What are digital trust mechanisms for digital engineering? Many trust mechanisms used in cloud service (Huang & Nicol, 2013) are applicable to digitalized products, systems, and services, while still, what new mechanisms should be introduced? Should be the sharing of digital engineering models and data in a centralized way or distributed way? There are many issues and challenges ahead as we conduct engineering in a very different new digital environment (Coatanéa, Nagarajan, Panicker, & Mokhtarian, 2022; Horváth, 2022; Huang et al., 2020).

Earlier in section 2, we discussed the journey of four industrial revolutions in human history. Potentially, what is next? Given the broad and profound impacts of digital engineering transformation on human society, a farther vision will help us to develop long-lasting engineering paradigm(s) and to better design and develop digital engineering systems.

Figure 9 illustrates the journey of industrial revolutions from passively following nature to actively exploiting nature and possibly achieving harmony between human society and nature. Let us start the discussion with the physical limits of computing. Computational power is the core of ubiquitous machine intelligence, the defining power we gain from the 4IR. However, with Moore’s law reaching its limit, the chips for CPUs have hit the ceilings of size and performance, particularly power consumption. The High-Performance Computing (HPC) community has been striking hard to achieve an exascale computing system with power consumption within 20 MW (Lucas et al., 2014). In the most recent Top500 List issued in June 2022, the current fastest supercomputer frontier is the first exascale machine with 1.1 ExaFlop/s at 21 MW. What we hope to beat Moore’s law are new computational technologies emerging in the horizon, such as new semiconductor materials, Quantum Computing, and DNA computing.

OPEN IN VIEWER

Fig. 9

The journey of human society through the industrial revolutions.

Data centres are the most significant contributors to today’s computational power, and data centres also consume a tremendous amount of energy and have significant environmental footprints. For example, in the US, data centres consumed 1.8% of electricity in 2014 and were responsible for 0.5% of greenhouse emissions in 2018 (Siddik, Shehabi, & Marston, 2021). It has become more significant regarding how to improve data centres’ sustainability. Microsoft’s Project Natick conducted a two years experiment and found that subsea data centres are feasible, reliable, more energy-efficient, and environmentally sound (Microsoft Research, 2020).

In a global context, since the first industrial revolution, economic growth has been associated with high environmental costs, including environmental pollution, overexploitation of natural resources, deterioration of ecosystems, loss of biodiversity, and global climate change. The problem is challenging because entities individually lack the motivation to behave environmental-friendly in the market economic mechanism due to environmental externalities. “Sustainable development is development that meets the needs of the present without compromising the ability of future generations to meet their own needs.” (UN WCED, 1987) It has become a consensus of the international community (IPCC, 2022; United Nations, 1992). According to the IPCC AR6 report (IPCC, 2022), global warming is speeding up - The increase in global surface temperature in 2011–2020 is 1.09 [0.95 to 1.20] °C higher than in 1850–1900, compared to the increase of 0.19 [0.16 to 0.22] °C in 2003–2012. The trend of increase will reach or exceed 1.5°C in the near term, even for the very low greenhouse gas emissions scenario. Climate change has led to some irreversible impacts on the earth and is approaching a tipping point. This cross-century challenge has become urgent and needs the whole world to act collectively and immediately before it is too late. This global campaign appears geared towards the next industrial revolution for renewable natural resources, particularly renewable energy, reusable materials, and eco-environmental sound economies to achieve harmony between human activities and nature.

In this direction, science, such as disciplines in biological and environmental sciences, energy science, and material science, among others, will be a scientific foundation to reveal facts and provide knowledge about sustainability. Engineering will bring scientific knowledge into the real world by developing technologies and designing innovative solutions for sustainability. Sustainability is a highly complex problem because it is across disciplinary domains, across industry sectors, across regional economies, across cultures, across human groups, across human generations, and across human society and natural systems. The solutions for sustainability need to be coordinated, comprehensive, and systematic. Systems science and engineering will play a unique role in applying systems thinking and designing systematic mechanisms and solutions to address the complex problem of sustainability.

With sustainability in mind, in today’s efforts for digital engineering transformation, we should take into account the need for sustainability in digital engineering systems design and development. For example, product provenance can help reuse and recycle after retirement. Particularly, it is critical to consider the sustainability of new digital technologies (Colorado, Velásquez, & Monteiro, 2020; Kellens et al., 2017; Paris, Mokhtarian, Coatanéa, Museau, & Ituarte, 2016). The consideration of sustainability in digitalization will give the emerging digital engineering paradigm(s) a long vision and better address future needs.