1 Introduction

In January 2021, a policy brief from the European Commission stirred up the industrial science community by introducing the concept of Industry 5.0 as a paradigm shift from previous growth-driven industrial transformation concepts to the value-driven quest for a sustainable, human-centric and resilient economy (Breque et al. 2021). Since the early 2010s, European manufacturing professionals have been in a tug of war to accomplish the promises of Industry 4.0, formerly branded by the German government and tasked by the European Union to “drive digital manufacturing forward by increasing digitization and the interconnection of products, value chains, and business models” (The European Commission 2017). Xu et al. (2021) assert that Industry 4.0 and Industry 5.0 largely coexist and essentially rely on similar technological baselines that include Additive Manufacturing, Industrial IoT, Digital Twins, and Artificial Intelligence (AI). McCarthy et al. (1955) define AI as a broad area of computer sciences that intends to give machines the ability to perform actions that would be considered intelligent if performed by humans. The purpose of this work is to analyze how AI affects innovation capacity in the context of Industry 5.0. It has important implications for innovation strategy and policy-making which are addressed below.

The classic literature conveys innovation capacity the leading and starting role to generate business growth due to technology development and more effective managerial practices (Bodrožić and Adler 2018). But while AI spreads as a General-Purpose Technology (GPT) with pervasive uses across a wide variety of sectors and technical dynamism (Trajtenberg 2018), there is growing evidence of a retro-effect by which AI technology affects innovation capacity Liu et al. (Liu et al. 2020). Previous research proves the positive effect of Human Intelligence (HI) on Innovation (Squalli and Wilson 2014). There is also evidence of the positive effect of AI on Innovation Capacity (IC) (Aghion et al. 2017). Yet, there is little consensus on the effect of AI on HI, which according to different research streams is found to be synergic (Lichtenthaler 2018), competitive (Shabbir and Anwer 2018), deceptive (Sætra 2020), or substitutive (Avolio et al. 2014). Therefore, our research question can be stated as follows: what factors influence AI effects on Innovation Capacity in the context of Industry 5.0?

According to the European Commission (2022), the increment from Industry 4.0 to 5.0 reflects a change of focus toward the societal challenges of sustainability, human-centricity, and resilience. However abundant engineering literature highlights the singular contribution of AI to this transformation. For example, Bryndin (2020) highlights the need for substantial improvements in cognitive systems. Industry 5.0 requires the adoption of AI agents with cognitive skills (Bryndin 2020) and collabo-rative robotics (Nahavandi 2019). Therefore we formulate hypothesis n°1 (H1): AI Maturity [AM] is associated with Company Age [CA]. Economists have also investigated this field, finding AI transformative beyond the limited scope of engineering systems, with effects on business models and supply chains. Trajtenberg (2018) thus describes AI as the next general-purpose technology (GPT). Therefore we formulate hypothesis n°2 (H2): Manufacturing Strategy [MS] is associated with AI Maturity [AM]. Finally, the innovation management literature questions the interplay of AI and human intelligence (HI) and its effect on innovation capacity. Therefore we formulate hypothesis n°3 (H3): Innovation Capacity [IC] is associated with Manufacturing Strategy [MS].

This scoping review aims to identify the factors influencing the effects of AI on IC in the context of Industry 5.0, qualify these effects and extract related indicators. These findings will provide researchers with a new understanding of the interplay between artificial intelligence and human intelligence. They provide practitioners with decision metrics for a successful transition to Industry 5.0. Although a European label, Industry 5.0 comes as a global paradigm shift with similar concepts being proposed in many regions and countries of the world. The impact of this work is therefore global. Our paper is structured in 7 sections. Following this introduction (Sect. 1), Sect. 2 describes the scope and methodology. In Sect. 3 we review the engineering literature uncovering H1. In Sect. 4 we review the humanities literature uncovering H2. In Sect. 5 we review the management literature uncovering H3. In Sect. 6 we discuss the main findings and propose an exploratory model. In Sect. 7 we conclude.

2 Scope and methodology

The research question opens broad areas of research from different disciplines that allow for many different possible study designs. Therefore, we have decided to conduct a scoping review. This type of knowledge synthesis follows a systematic approach to map evidence on a topic and identify main concepts, theories, sources, and knowledge gaps (Tricco 2018).

2.1 Literature search

Following the concepts of the research question, a systematic literature search was carried out on works published since January 2021 using the following keywords: “Industry 5.0” and “Artificial Intelligence” and “Innovation”. A Scopus search (updated on 02/09/2023) gives 48 results. After a review of all abstracts, 34 were found to be irrelevant, meaning that they did not effectively address the research question or any related research question. 14 are found to be partially relevant, meaning that they address alternative research questions interlinked with our research question. Acknowledging the common technological baseline between Industry 4.0 and Industry 5.0 (Xu et al. 2021), we expand the search by replacing 5.0 with 4.0 and we obtain 479 results. The review of all abstracts gives evidence of no more than 36 literature works discussing key concepts of the research question. Adding both search results gives a total of 50 relevant works. From this set, a bibliographic analysis brings up 283 additional works of interest. The figure below gives an overview of the literature search and selection procedure (Fig. 1).

Fig. 1
figure 1

Literature search and selection procedure

Full size image

In total 333 works match the criteria: “provides insights on at least one factor influencing the effect of AI on innovation capacity in the context of an industrial transformation”.

2.2 Literature data

The collected works belong to 3 distinct bodies of literature which we call “classes” in the following lines:

  • The Engineering class (ENG: 102) includes works of science and technology concerned with engines, machines, and structures;

  • The Humanities class (109 works) includes works of economics, sociology, and industrial organization.

  • The Management class (122 works) includes works of business strategy, industrial policy, and innovation management.

The following key concepts appear from a first overview of the literature:

  • Company Age [CA], defined as the level of industrial advancement of a company according to Industry 5.0 maturity framework (from Industry 1.0 to Industry 5.0);

  • AI maturity [AM], defined as a level of technological advancement according to the properties of AI assets (learning ability, embeddedness, anthropomorphism);

  • Manufacturing Strategy [MS], defined as the adopted business strategy according to product/service mix and supply chain structuration;

  • Innovation Capacity [IC], defined as the organizational potential to innovate, determined by the skills and strengths in research and technology.

For the purpose of structuring the review process, we assess the coverage of the above concepts by each literature work. The result is a dataset of 333 items with 4 binary variables: [CA], [AM], [MS], [IC]. We look for a statistical technique enabling to retrieve patterns and trends within this literature. We consider factor analysis or cluster analysis relevant for this investigation. However, the distribution of data is slightly positively skewed for [AM], [MS] and [IC]. Therefore we opt for cluster analysis technique, for which requirements of normality do not apply (Hair 2009).

2.3 Cluster analysis

With this paper, we aim to answer the research question by consigning findings from literature content analysis, not by interpreting bibliometric figures. However, to describe their respective contributions, each class of literature is sub-split into clusters corresponding to influential trends or theories. In the following lines, we call each theory a “cluster”. For conciseness, we have set to 4 the maximum number of clusters per class and consigned outliers into a fifth cluster called “miscellaneous”. The cluster assignment decision has been made based on a complete critical review, controlled by keywords frequency check and consistency check with the journal editorial line. Clusters are mapped to the key concepts above identified by a scoring system. Scores are computed by summing the number of works effectively addressing the considered factor within the cluster. Table 1 provides the distribution of the selected literature by cluster and their scoring by key factor.

Table 1 Literature review statistics
Full size table

The statistics provide evidence of good coverage of the key concepts at play and a balanced distribution of research work across the 3 bodies of literature, which justifies this structuration of the review scope. In particular, it sets clear the respective focus of the engineering, humanities, and management literature on AI maturity [AM], manufacturing strategy [MS], and innovation capacity [IC], respectively, as factors of interest. Works providing support for investigation methodology are marked [MY].

2.4 Clusters interpretation

The Engineering class (ENG: 102) score high in [CA] and [AM], enabling to assess the hypothesis H1: AI Maturity [AM] is associated with Company Age [CA]. This class is split into:

  • Artificial intelligence theories cluster (AIT: 36): consigning works about AI techniques, algorithms, and implementations; helping to characterize AI agents in the context of Industry 5.0;

  • Manufacturing automation cluster (MFG: 9): consigning works about industrial processes and operations, helping to define the place of AI in the manufacturing lifecycle;

  • Smart manufacturing cluster (SMT: 16): consigning works about ICT integration in manufacturing operations, providing insights into the role of AI in manufacturing use cases;

  • Cyber-physical systems cluster (CPS:36): consigning works of system engineering, defining characteristics of Industry 5.0 systems relative to anterior technology, inferring [CA] from [AM].

The bivariate scatter chart below casts the cluster class centroids on a standardized bidimensional chart with [CA] in abscissas and [AM] in ordinates (Fig. 2).

Fig. 2
figure 2

Bivariate scatter chart X = [CA]; Y=[AM]

Full size image

The chart confirms the “Engineering” literature class as a predictor for the association between [CA] and [AM]. The average within-class distance is 0,25 while the average between-class distance is 0,42 showing acceptable class homogeneity.

The Humanities class (109 works) scores high in [CA], [MA] and [MS], enabling to assess the H2: Manufacturing Strategy [MS] is associated with AI Maturity [AM]. This class splits into:

  • Creative destruction theory cluster (CDT: 44) consigning works of economics inspired by the theory of Schumpeter (1942), discussing the AI effects on firm resources;

  • Transaction cost economics cluster (TCE: 44): consigning works of industrial organization inspired by the theory of Williamson (1981), helping to assess AI effects on firm boundaries;

  • Theories of intelligent manufacturing cluster(TIM): consigning works focusing on the effects of digital transformation on industrial strategy, helping to assess AI impact on the marketing mix;

  • Social exchange theories (SET), consigning works of economics and sociology inspired by Macneil (1974), help predict AI effects on manufacturing strategy [MS].

The bivariate scatter chart below casts the cluster class centroids on a standardized bidimensional chart with [AM] in abscissas and [MS] in ordinates (Fig. 3).

Fig. 3
figure 3

Bivariate scatter chart X = [AM]; Y=[MS]

Full size image

The chart confirms the “Humanities” literature class as a predictor for the association of [AI] with [MS]. The average within-class distance is 0,26 while the average between-class distance is 0,47 showing acceptable class homogeneity.

The Management literature class (MGT: 122): scoring high in [CA] [MS] and [IC], enables to assess H3: Innovation Capacity [IC] is associated with Manufacturing Strategy [MS]. It splits into:

  • Resource-based view cluster (RBV): consigning works of management science influenced by the theory of Barney (1991), helping to assess the effect of AI on internal innovation;

  • Open innovation theory cluster (OIT): consigning works of innovation management influenced by Chesbrough (2014), helping to assess AI effect on collaborative innovation strategies;

  • Innovation policy literature cluster (IPL): consigning works of political science addressing innovation policy decisions, helping to assess the effect of innovation policy on AI adoption;

  • Innovation management theory cluster (IMT): consigning works of management addressing effects of innovation on firm performance, helping to predict the AI effect on innovation capacity.

The bivariate scatter chart below casts the cluster class centroids on a standardized bidimensional chart with [MS] in abscissas and [IC] in ordinates (Fig. 4).

Fig. 4
figure 4

Bivariate scatter chart X = [MS]; Y=[IC]

Full size image

The chart confirms the “Management” literature class as the best predictor for manufacturing strategy effect on innovation capacity. The average within-class distance is 0,26 while the average between-class distance is 0,55 showing good class homogeneity.

3 Review part I: review of the engineering literature

3.1 Characterization of AI agents

The AI engineering literature defines several different types of AI agents that have different abilities and consecutively different applications in the context of Industry 5.0. Symbolic AI handles discrete symbols, such as rule-based or logic-based systems (Bunte et al. 2019). Machine learning (ML) gives computers the ability to learn without being explicitly programmed (Samuel 1959). Deep learning (DL) further extends the ability of machines to perform human tasks that can be learned through observation and experience (Wason 2018). Bryndin (2020) highlights the need for substantial improvements in cognitive systems to achieve the Industry 5.0 promise for human centricity. A debate, however, remains on whether AI should aim at thinking humanly, acting humanly, thinking rationally, or acting rationally (Russell and Norvig 2009). Unlike virtual agents, embodied AI gives machines the ability to perform tasks in the physical space, similar to humans (Ziemke 2003). Burden (2008) notes that embodied AI can, however, be involved in virtual worlds alike. An important obstacle to AI adoption is the limited human trust in AI (Siau 2018). Anthropomorphism potentially helps develop trust in user relationships (Złotowski 2014). However, a nonlinear relationship between a robot’s degree of anthropomorphism and likeability was formalized with the uncanny valley theory (Mori 1970). Additionally, Terada et al. (2007) found that humans tend to attribute intentions even to nonhumanoid robots. Trajtenberg (Trajtenberg 2018) considers AI as the next general-purpose technology (GPT) characterized by pervasive use across a wide variety of sectors and technical dynamism. Agrawal et al. (2019), however, reject the claim for ML to form a true “Artificial General Intelligence” (AGI). Building on this literature we can characterize AI agents in the context of Industry 5.0 as follows (Fig. 5).

Fig. 5
figure 5

Characterization of AI agents

Full size image

The AI engineering literature defines different types of AI agents with different levels of learning ability, embodiment, and anthropomorphism (Russell and Norvig 2009). AI type [AT] is therefore a possible indicator of AI Maturity [AM].

3.2 AI place in manufacturing lifecycle

Industrial AI (IAI) can be broadly defined as any application of AI relating to the physical operations of an enterprise (Lee et al. 2018). The manufacturing literature provides evidence of IAI implementation at all steps of the lifecycle. Before commissioning, it is found to support generative design (Autodesk 2015) and design to production (Balu et al. 2016). There are many examples of use for production scheduling(Norrie et al. 2001) and job shop scheduling (Calis Uslu and Bulkan 2013). After commissioning, IAI supports product inspection (Park et al. 2019), yield management (Shin and Park 2010), operations sequencing (Gacek 2012), supply chain risk management (Baryannis et al. 2018), automated fault prevention (Milazzo et al. 2020) and predictive maintenance (Korvesis 2017). Gupta et al. (2019) address the use of AI in waste management and recycling. Yin & Yuanyuan (2022) demonstrate the contribution of AI technology to the quality, efficiency, and benefit of green innovation in manufacturing. Building on this literature we can describe the place of AI in the manufacturing lifecycle as follows (Fig. 6).

Fig. 6
figure 6

AI in manufacturing lifecycle

Full size image

Manufacturing automation research specifies AI place [AP] in the manufacturing lifecycle as an observable indicator of AI Maturity [AM] (Solanki et al. 2020). A broader application of AI along the manufacturing lifecycle reveals a greater level of AI Maturity [AM].

3.3 AI role in manufacturing use cases

Many Industry 4.0 studies provide evidence of the operational benefits of AI for process monitoring (Lihu and Wei 2017). Other studies have demonstrated AI-based process monitoring (Li et al. 2017) or optimization (Xu and Lu 2019). Other use cases leverage AI for process control and automation (Wang 2019). In Industry 5.0, AI agents can achieve more sophisticated tasks requiring social behavior (Seeber et al. 2020), collaborative learning capacity (Bauer et al. 2018) or group cognition and negotiation abilities (Koch and Oulasvirta 2018). These abilities are best demonstrated in collaborative robotic applications (El Zaataria et al. 2019). Dong (2023) demonstrate that AI contributes to the propagation of digital green innovation in the building supply chain. Bécue et al. (2020) also demonstrate the possible contribution of AI to manufacturing system resilience towards cyber threats. Based on this literature we can describe the possible roles of AI in Industry 5.0 as follows (Fig. 7).

Fig. 7
figure 7

IAI roles in manufacturing use cases

Full size image

The smart manufacturing literature highlights the AI role [AR] as an observable indicator of AI Maturity [AM] (Bauer et al. 2018).

3.4 Company age and AI maturity

The system engineering literature provides a classification of manufacturing systems according to AI maturity levels. A system is a complex set of interacting elements with properties richer than the sum of its parts (Bertalanffy 1969). Yilma et al. (2020) define elementary systems as cyber (C), physical (P), or social (S). Their combination forms physical-social systems (PSS), cyber-social systems (CSS), cyber-physical systems (CPS), and cyber-physical social systems (CPSS). PSS are compositions of physical and social elements, where the social part is materialized through the physical part (e.g., a worker). CSSs are systems comprising a cyber and a social component where the social element is expressed through cyberspace (e.g., social networks). CPS integrate computing elements with physical components and processes (e.g., industrial robot). Finally, CPSS comprises cyber, physical, and social components. CPSS spaces are metasystems where humans and CPS cohabit (e.g., smart factories). A CPSS object is enhanced with social capabilities (e.g., a collaborative robot). Based on this literature, we classify the different levels of industrial AI maturity as follows (Fig. 8).

Fig. 8
figure 8

AI maturity in system engineering

Full size image

The systems characterizing Industry 5.0 belong to the class defined by the system engineering literature as cyber-physical social systems (Yilma et al. 2020). They correspond to the highest level of AI development (Radanliev et al. 2021). Therefore, we verify H1): AI Maturity [AM] is positively associated with Company Age [CA].

4 Review part II: review of the humanities literature

4.1 AI effect on firm resources

Creative destruction is the process of industrial mutation that incessantly revolutionizes the economic structure from within, incessantly destroying the old one and incessantly creating a new one (Schumpeter 1942). However, while innovation has been known to affect lower-qualification jobs, the advent of AI now increasingly exposes the management level (Frank 2019). Pessimistic views generally claim (i) that AI kills jobs and (ii) that AI has a limited impact on society. Mokyr (2017) notes that these views cannot both be right and proves that they can very well be both wrong. Acemoglu and Restrepo (2018) expect the mismatch between skills and technology to widen by the combined effects of “displacement”, whereby machines take over human tasks, and “expansion”, whereby growth leads to the creation of new tasks for humans. Furman (2018) forecasts that low wages and skills are more exposed to job losses and that this effect is amplified by inequitable educational systems. Agrawal et al. (2019) reply that in certain domains, downskilling is equally likely and would potentially reduce inequalities. Existing theories mostly diverge between supporters of “replacement” or “augmentation” (Noponen 2019). Theories of “replacement” postulate that AI will gradually replace human management (Avolio 2000) and that the “embodied leader” will be blown by digital platforms (Auvinen et al. 2019). Algorithmic management is expected to become the ultimate bastion for managers (Lee et al. 2015) and is accused of raising exploitative information asymmetries (Rosenblat, 2015). However, several authors highlight the possible symbiosis (Jarrahi 2018) or augmentation (Autor et al. 2020) enabled by the distinct abilities of AI and HI. According to Frey and Osborne (2013), tasks demanding social intelligence would be less affected. This, however, contradicts the proliferation of chatbots to support customer relations (Foye 2017). Based on this literature we can represent the possible effects of AI on firm resources as follows (Fig. 9).

Fig. 9
figure 9

IAI effect on firm resources

Full size image

Creative destruction theories highlight the relevance of company skills [CS] as an observable indicator of company age [CA] (Acemoglu and Restrepo 2018). Furthermore, they provide resource strategy [RS] as an observable indicator of a manufacturing strategy [MS] (Noponen 2019).

4.2 AI effect on firm boundaries

Transaction cost economics (TCE) postulates that the success of organizations essentially depends on their ability to make rational decisions between market or hierarchy (Williamson 1981). Asset specificity, transaction costs, uncertainty, and governance costs determine the optimal position of the firm boundary. TCE-based studies have found that information technology (IT) reduces negotiation time, (Malone et al. 1987) search costs (Bakos 1998), coordination complexity (Hart 1991) and formalism (Clemons 1993). The ICT economy has simultaneously triggered a reduction in the site specificity of industrial assets, a reduction in transaction costs, and, less expectedly, an increase in internal management costs due to the complexity of offerings (Kasper and Streit 2000). AI studies show that customer interaction, sales platforms, and marketing are among the most common uses of AI, which contributes to further reducing transaction costs (Davenport et al. 2020). However, there is no evidence in the literature that this applies to industrial AI. In a recent study, AI was found to trigger organizational complexity, increased market asymmetry, uncertainty, and decisional inertia, potentially driving a preference for hierarchy and consecutively a concentration of the industry (Mateos-Garcia 2018). Based on this literature we can predict the following effects of AI on firm boundaries (Fig. 10).

Fig. 10
figure 10

IAI effect on firm boundary

Full size image

Whether IAI plays for market or hierarchy, we retain network strategy [NS] as an observable indicator of a manufacturing strategy [MS] from a TCE perspective (Mateos-Garcia 2018).

4.3 AI effect on the marketing mix

Computerization is among the acknowledged factors of the dematerialization of the industry and the emergence of a service economy (Roy 2000). To explain this transformation, Theories of Intelligent Manufacturing (TIM) have come to the concept of “Service-Oriented Manufacturing (SOM)” (Gao et al. 2011), where services and physical products are integrated into Product Service Systems (PSS) to sustain a competitive advantage in a context of overproduction (Tukker 2015.). TIM predicts the polarization of corporate strategies towards either “cost leadership” or “differentiation”. The integration of “supplementary activities” (Porter 1980) leads to “demand-sided servitization” (Gao et al. 2011). Cost leadership is achieved by externalization of “supplementary activities”, which leads to “supply-sided servitization”. An illustration of this polarization can be found in Shih’s “smiling curve” (Shin et al. 2012). Moving up both sides of the value curve towards technological excellence and industry insights can be seen as an effective differentiation strategy. Frank et al. (2019) suggest that digitization further accelerates servitization. They propose a framework model that connects Servitization and Industry 4.0 from a business model innovation perspective. Based on this literature we can describe AI’s effect on the marketing mix as follows (Fig. 11).

Fig. 11
figure 11

IAI impact on marketing mix

Full size image

TIM literature reveals Company Product [CP] as an observable indicator of Company Age [CA] and Product/Service Strategy [PS] as an observable indicator of Manufacturing Strategy [MS] (Frank et al. 2019).

4.4 AI effect on manufacturing strategy

According to Chen et al. (2017), services are inherently relational. Therefore, the servitization of the industry should be examined from a relational perspective. In the mid-seventies, Ian R. Macneil (1974) inspired the Social Exchange Theory (SET) movement with the finding that modern contract law theory suffered transactional bias. Consistently, modern SET literature highlights the prevalence of relational motivations in the adoption of e-commerce and customer-to-customer (C2C) service platforms (Shiau and Luo 2012) (Lu et al. 2010). Because AI supports enhanced interactions between humans and machines, this benefit expands to manufacturing. The social manufacturing literature finds digitization to help manage complex manufacturing communities where the vertical buyer-seller relationship is replaced by a horizontal relationship between active “prosumers” (Ding et al. 2016). Deep learning has been proven to support supply-demand matchmaking in the social manufacturing context (Leng and Jiang 2016). Other studies claim that industrial digitization makes product personalization compatible with large-scale production leading to the concept of “mass customization” (Jiang et al. 2016). Frank et al. (2019) note the double role of data in Industry 4.0 as (i) an enabler for customer value creation and (ii) a channel for internal process improvement. Based on this literature we can forecast an evolution towards the following manufacturing strategies (Fig. 12).

Fig. 12
figure 12

IAI effect on business strategy

Full size image

The SET literature suggests an evolution of manufacturing strategies towards AI-powered social manufacturing and mass customization approaches (Jiang et al. 2016). Therefore, we verify H2): Manufacturing Strategy [MS] is associated with AI Maturity [AM].

5 Review part III: review of the management literature

5.1 AI as a strategic key capability

The resource-based view (RBV) postulates that the acquisition of a sustainable competitive advantage relies on resources gathering singular attributes: (1) value, (2) rareness, (3) imperfect imitability, and (4) non-substitutability (Barney 1991). Teece et al. (1997) claim that substantial performance gains can only be drawn from investment in a company-wide “capability” encompassing tangible, intangible, and human resources. Many RBV studies prove that IT is a reliable source of sustainable competitive advantage (Alalie et al. 2018). However, Mikalef et al. (2019) find employee resistance to be the main barrier to AI adoption. Daugherty and Wilson (2018) propose exploiting the synergy between AI and HI. A systematic analysis of interdependencies between them would enhance a firm’s ability to profit from AI (Davenport and Ronanki 2018). Lichtenthaler (2019) shows that the integration of AI applications with specific human expertise is likely to complicate imitation by competitors (Lichtenthaler 2019). Based on this literature, we set the following properties for AI to form a strategic key capability in the context of Industry 5.0 (Fig. 13).

Fig. 13
figure 13

IAI as a strategic key capability

Full size image

The RBV literature highlights internal innovation [II] primarily as an observable indicator of innovation capacity [IC] and suggests AI innovation as a strategic key capability (Lichtenthaler 2019).

5.2 AI as a driver for open innovation

Open innovation (OI) is defined as a distributed innovation process based on purposively managed knowledge flows across organizational boundaries (Chesbrough, 2014). Goldfarb and Trefler (2018) finds that the development of AI contributes to the Internet to the reduction of boundaries of knowledge within and between companies, consequently supporting faster diffusion of knowledge and new knowledge generation by exchange and integration. Liu et al. (2020) demonstrate the spillover effect whereby intelligent technology drives interdivisional innovation, resulting in profit growth and renewed demand for AI integration. Vlačić(2019) proves that absorptive capacity is an effective means of obtaining and sustaining a competitive advantage. Liu et al. (2020) further prove that AI improves the absorptive capacity of the firm, preventing errors in technology choices and innovation investment decisions. A noticeable initiative towards OI in the field of IAI is the Common European Data Space for Manufacturing (European Commission 2020). However, OI raises issues about intellectual property (Mention and Al-Sharieh 2013). It makes confidentiality keeping of trade secrets increasingly difficult, which eventually conflicts with internal innovation mechanics. Based on this literature we can expect AI to affect the following drivers for open innovation (Fig. 14).

Fig. 14
figure 14

IAI and open innovation

Full size image

OI literature holds open innovation [OI] (also called openness) as a relevant indicator of innovation capacity [IC] (Goldfarb and Trefler 2018).

5.3 AI as a matter of innovation policy

AI adoption by manufacturing might be subject to conflicting policies. Agrawal et al. (2019) believe that research support policies will accelerate AI progress. Tucker (2018) highlights the moderating influence of privacy protection policy on the rate and impact of AI innovation. Agrawal et al. (2019) call “race to the bottom” the temptation for governments to set privacy obligations to a strict minimum. Patent protection laws affect the direction and speed of innovation (Moser 2005). Tort laws are also likely to have an impact on IAI adoption. Galasso and Luo (2018) state that finding the right balance on liability will be important for the industrial diffusion of Agrawal et al. (2019) suggest policies to care for synchronizing business and educational cycles. China now fosters AI education at schools (Xinhua 2018), and analysts fear the deepening of an “AI divide” that would reinforce world inequalities. Sachs (2018) predicts that AI will rise at the expense of the labor share of GDP. Korinek (2018) thus recommend taxation of capital. Furman (2018) fears that a universal income would not be spent on education. The European Commission regularly accuses AI of reinforcing the hegemony of large monopolistic firms. Bajari et al. (2018) counter this argument by showing that data have a decreasing return to scale. Based on this literature we acknowledge different policies moderating AI’s effect on innovation (Fig. 15).

Fig. 15
figure 15

IAI as a matter of innovation policy

Full size image

The innovation policy literature suggests that company size [CZ] is an observable indicator of company age [CA] (Bajari et al. 2018). Furthermore, it provides innovation policies [IP] as an observable indicator of innovation capacity [IC].

5.4 AI Effect on innovation capacity

O’Leary (2013) reports that AI accelerates knowledge creation with a combined impact on data collection, structuration, analysis, exploration, and integration. Aghion et al. (2017) assert that AI has a positive effect on innovation. Zhou (2012) suggests that AI integration in manufacturing provides not only a common enabling technology for mechanical product innovations but also one for manufacturing technology innovations. Liu et al. (2020) add that AI-enabled practices such as remote manufacturing, computer-integrated business systems, and data platforms relieve companies from geographical constraints, enabling a more efficient division of labor, information sharing, and resource integration with a direct positive effect on technology development. Du et al. (Du et al. 2018) claim that technologies such as deep learning can help industrial robots form independent judgments and take corresponding actions, relieving employees from low value-added activities and helping them make better-informed decisions. Liu et al. (2020) demonstrate that AI promotes technological innovation by (i) quickening knowledge creation, (ii) accelerating the spillover of knowledge and technology, (iii) improving the capability of learning and absorption, and (iv) boosting investment in R&D and talent. Agrawal et al. (2018) highlight the contribution of AI to the process of knowledge creation, retrieval, and combination. Theben et al. (2021) suggest open-source artificial intelligence (AI) to boost innovation capacity in both the public and private sectors. However, Sætra (2020) warns about the parasitic nature of AI interaction with humans and fears that “Sharing minds with the mindless” might diminish humans’ whole learning ability. Based on this literature, we can draw the following possible effects of AI on Innovation Capacity (Fig. 16).

Fig. 16
figure 16

IAI effect on innovation capacity

Full size image

The innovation management literature suggests that AI-enabled practices such as remote manufacturing positively affect innovation capacity (Liu et al. 2020). Therefore, we verify H3: Innovation Capacity [IC] is associated with Manufacturing Strategy [MS].

6 Results and discussion

6.1 Summary of the findings

The review of the literature enables to extract 4 factors influencing the effect of AI on Innovation Capacity in the context of Industry 5.0. We define these factors as follows:

Company Age [CA]: a factor representing the industrial development step of a company according to the Industry 5.0 framework, and suggest;

AI Maturity [AM], a factor representing the level of artificial intelligence involved in company operations;

Manufacturing Strategy [MS], a factor characterizing the Business Strategy of a manufacturing firm.

Innovation Capacity [IC], a factor assessing the overall ability of a company to innovate.

The review of the engineering literature verifies H1: [AM] is associated with [CA]. This association is driven by Artificial Intelligence (AI).

The review of the humanities literature verifies H2: [MS] is associated with [AM]. This association is driven by Social Intelligence (SI).

The review of the management literature verifies H3: [IC] is associated with [MS]. This association is driven by Human Intelligence (HI).

We therefore establish the following theoretical framework (Fig. 17).

Fig. 17
figure 17

Theoretical framework

Full size image

A Notable outcome of this review is the finding that three types of intelligence are at play in this framework. The abilities of AI increasingly expand from analytical to social and creative activities, thereby affecting not only industrial processes but also manufacturing strategy and innovation capacity through a subtle interplay with social and human intelligence. The strength and directionality of these effects should be assessed by quantitative techniques which fall beyond the scope of this study. Yet for the purpose of such quantitative assessment, we extract from the literature a set of 12 indicators enabling to measure the 4 main factors. The figure below provides a summary of the variables extracted (Fig. 18).

Fig. 18
figure 18

Variables extraction

Full size image

6.1.1 Company variables (CY)

From the review of all literature bodies, we conclude that certain properties like product (Frank et al. 2019), size (Bajari et al. 2018), and skills (Acemoglu and Restrepo 2018) are indicators of [CA]. For example, TIM predict that AI adoption will drive further service integration in manufacturing products. CDT suggest automation by AI will make smaller-scale production units fit for business. RBV predicts a battle for strategic AI skills and expertise required to build a sustainable competitive advantage. The CPS literature defines observable attributes of manufacturing system characteristics of each industrial age from Industry 1.0 to Industry 5.0. Therefore, [CA] can be measured by observable variables such as company product [CP] size [CZ] and skills [CS].

6.1.2 Artificial intelligence variables (AI)

From the review of the engineering literature body, we conclude that the type, the place, and the role of AI in use, are indicators of AI maturity in the transition to Industry 5.0. For example, AIT shows that self-learning agents are capable of interacting more naturally with human workers (Bryndin 2020). MFG shows that the most advanced companies employ AI at all steps of the life cycle with direct positive effects on the environment (Yin & Yuanyuan, 2022). SMT demonstrates that advanced industries assign AI a greater diversity of roles, with growing considerations for collaboration and resilience functions (Solanki et al. 2020). CPS literature suggests AI maturity as the main predictor of industrial transformation in the frame of Industry 5.0 (Yilma et al. 2020). Therefore, [AM] can be measured by a construct of observable variables such as AI type [AT] AI place [AP] and AI role [AR].

6.1.3 Business strategy variables (BS)

From the review of the humanities literature, we conclude that the resource, network, and product strategies of Industry 5.0 companies will differ markedly from those of companies with lower maturity in AI. For example, CDT shows that AI unlike previous general-purpose technologies does not only affect manual jobs but also intellectual ones (Noponen 2019). TCE highlights there can be a greater asset-specificity in industrial AI, that revives vertical integration against horizontal networking strategies (Mateos-Garcia 2018). TIM shows that AI is likely to accelerate the servitization of the economy and the polarization of product strategies (Tukker 2015.). SET predicts the development of new manufacturing strategies such as social manufacturing and mass customization (Jiang et al. 2016). Therefore [MS] can be measured by a construct of observable variables such as Resource Strategy [RS], Network Strategy [NS], and Product Strategy [PS].

6.1.4 Innovation maturity variables (IM)

The management literature provides evidence of distinctive attributes of an Industry 5.0 innovation capacity boosted by AI (Liu et al. 2020). For example, RBV forecasts a battle for AI skills and technology considered as strategic key capabilities for the realization of internal innovation strategies (Lichtenthaler 2019). OIT suggests a faster diffusion of knowledge across company boundaries through the advent of AI, fostering an increased preference for open innovation strategies (Goldfarb and Trefler 2018). IPL highlights the moderating effect of coercive policies on AI adoption (Agrawal et al. 2019). IMT predicts contradictory effects of AI on innovation capacity through a complex interplay with HI (Sætra 2020). Therefore, [IC] can be measured by a construct of observable variables such as Internal Innovation [II], Open Innovation [OI], and Innovation Policy [IP].

6.2 Comparison with other studies

An important contribution of our research is the in-depth analysis of the specific contribution of AI within the many different technologies usually praised for their role in the advent of Industry 5.0. According to Fazal et al. (2022) advanced implementation of AI is one of the major technologies of Industry 5.0 among others like collaborative systems, digital twins, exoskeletons, smart materials, the internet of things, cloud computing, and 4D printing. Unlike this and many contemporary research works, our study does not attribute the Industry 5.0 transformation to multiple technological enablers but to a single general-purpose technology which is AI (Trajtenberg 2018). We do not aim to diminish the role of other technologies. However, we find the incidence of AI singular in this context and a critical review of the literature comes to support this hypothesis.

Another singularity of this work is to address that singular object of AI through a multidisciplinary approach that brings together works of engineering, humanities, and management. The monodisciplinary approaches from Yilma et al. (2020), Davenport et al. (2020) and Liu et al. (2020) are noteworthy. However, they cannot grasp the interrelation between the technological (AI), socio-economic (BS), and managerial (IM) dimensions of the research questions. Our research confronts these disciplines which concurrently predict: (i) with Yin & Yuanyuan (2022) that AI enables green innovation; (ii) with Autor et al. (2020) that AI triggers the fall of labor and the rise of superstar firms; (iii) with Theben et al. (2021) that it drives open source innovation for public and private goods.

Finally, we provide original research by comparison with other review works that have focused primarily on bibliographic analysis, while our study aims at extracting concepts and variables from a qualitative analysis of the literature content. Among the 333 works reviewed, 15 are literature reviews. Within them, only 2 address the 4 key concepts of our research framework ([CA], [AM], [MS], [IC]). Cannavacciuolo et al. (2023) provide a systematic review of the technological innovation-enabling industry 4.0 paradigm. Oztemel and Gursev (2020) provide a Literature review of Industry 4.0 and related technologies. These works identify the interplay between technological and managerial factors at play in the industrial transformation. However, they do not specifically address the interplay between AI and HI and their effect on innovation capacity. This scoping review is therefore novel in scope and approach to the phenomenon under study.

6.3 Research gaps and debates

While providing substantial support to the study hypotheses, our work also brings up unfilled gaps and unsolved debates of research.

From the engineering literature, we find the resilience challenge (5 works) much less comprehensively covered than other Industry 5.0 challenges such as human centricity (22 works) and sustainability (8 works). The engineering literature in general underestimates psychosocial brakes to AI adoption such as the mistrust in AI righteously introduced by Siau (2018). The debate about whether anthropomorphism affects AI likeability positively (Złotowski 2014), negatively (Mori 1970) or negligibly (Terada et al. 2007) remains open. There is no consensus on the technical feasibility of AGI (Agrawal et al. 2019).

From humanities, we find the debate between Nopon’s theory of replacement (2019) and Autor’s theory of augmentation (2020) open. While Davenport et al. 2020 predicts AI to drive a preference for the market, Mateos-Garcia (2018) suggests industrial AI might trigger increased market asymmetry, thereby reviving a preference for hierarchy. With the advent of AI, we find the divide widening between strategies of cost leadership and strategies of differentiation, leaving entrepreneurs in a state of uncertainty (Tukker 2015). Finally, Ding et al. (2016) praise social manufacturing while Jiang et al. (2016) predict the advent of mass customization.

In management sciences, there is little consensus on the question of whether AI meets the requirements to form a strategic key capability (Barney 1991). The divide persists between OI approaches (Chesbrough 2014) and the contradictory claim for intellectual property (Mention and Al-Sharieh 2013). The conflict between coercive, liberal, and incentive policies brought to light by Agrawal et al. (2019) is unsolved as some countries race for AI regulation while others tactically race to the bottom. Finally, there is still debate whether AI affects IC positively (Aghion et al. 2017), negatively (Sætra 2020) or qualitatively (Liu et al. 2020).

6.4 Avenues for future research

From this review, we extract factors and indicators that the current literature provides to explain the effect of AI on innovation capacity in the context of Industry 5.0. Future work could aim 1) at solving some of the above-highlighted gaps and contradictions of contemporary research using data collection from direct sources, such as interviews of engineers, managers, and researchers involved in Industry 5.0 initiatives. From such sources, an explanatory model featuring variables and effects at play in this transformation could be elaborated. Finally, this model could be quantitatively tested against empirical data from a representative population of companies. We suggest Bayesian network (Heckerman 1997) testing would fit this purpose as i) it can handle situations where some data entries are missing, ii)-it has both a causal and probabilistic semantics, iii) it allows to represent expert knowledge in a probabilistic manner, iv) it can equally infer or predict the state of a non-observable variable This would help predict the impact of AI on innovation capacity [IC] based on a set of observable company/business characteristics, or reversely, infer company age [CA] from observable evidence of innovation and manufacturing strategy. Such a model could help reduce the risk and uncertainty in decision making with regards to AI-driven transformation programs towards Industry 5.0.

7 Conclusion

This scoping review gives consistent findings from existing literature to answer the research question: What factors influence the effect of AI on innovation capacity in the context of Industry 5.0? The reviewed engineering literature converges to support H1: “AI Maturity [AM] is positively associated with Company Age [CA]”. Although Industry 5.0 is defined by social goals, there is substantial evidence in the works of Złotowski (2014), Dong (2023) and Bécue et al. (2020) that these objectives, namely human centricity, sustainability and resilience may not be achieved without significant advances in AI.

The humanities literature largely supports H2: “Manufacturing Strategy [MS] is associated with AI Maturity [AM]”. AI advances enable novel business models specific to Industry 5.0. Frank (2019) highlights the shrinking effect on management lines, Mateos-Garcia (2018) the possible re-verticalization of supply chains, Frank et al. (2019) the acceleration effect on servitization. There is however a debate whether social manufacturing (Leng and Jiang 2016) or mass customization (Jiang et al. 2016) will prevail.

The management literature largely supports H3: “Innovation Capacity [IC] is associated with Manufacturing Strategy [MS]”. Vertical integration strategies like mass customization will drive a race for AI talents through a dominant internal innovation strategies (Lichtenthaler 2019). Conversely, horizontal networking strategies like social manufacturing will foster AI-powered open innovation tactics (Liu et al. 2020). Coercive policies will discourage AI-driven internal innovation while incentive policies will encourage open innovation (Agrawal et al. 2019).

The results of our scoping review suggest that the effect of AI on innovation capacity in the context of Industry 5.0 will be synergic (Aghion et al. 2017), deceptive (Sætra 2020) or substitutive (Avolio et al. 2014), depending on the alignment of company profile, AI maturity, manufacturing strategy and innovation policy. This finding is significant for the research, as it bridges a gap between the disciplines of engineering, humanities and management on this matter. It is significant for practice, as it provides decision metrics for entrepreneurs aiming to take the journey to Industry 5.0. The target population is 2 463 229 industrial companies currently registered by Eurostat. We suggest however that the Industry 5.0 paradigm does not rest within the boundaries of Europe so that a wider worldwide community of entrepreneurs and innovation leaders benefit from these findings.