We used bibliometrics to assess the present condition of the field of AI-I as well as its potential and emerging trends. Bibliometrics is a valuable tool as it scrutinizes research progress, including important journals, authors, papers, countries, and other relevant factors (Hood & Wilson, 2001). Thus, through effective processes of sorting, mapping, and visual analysis, bibliometrics highlights the overall structure of particular themes or fields. The present study employed the five-step methodology proposed by Fahimnia et al. (2015) for comprehensive data collection and analysis within a specific field: mentioning the database used, screening the initial results, refining the obtained results, developing particular data statistics, and examining the dataset with some bibliometric techniques. Fig. 2 presents our methodological framework for conducting searches, screening relevant information, and refining our strategies.

In line with a previous approach for identifying relevant keywords (Cheng, 2023), we have reviewed the recurrent interdisciplinary aspects related to AI-I. Subsequently, we sought input from researchers in both the fields of science and technology and social science to identify the most relevant components in the relationship with which AI-I is scored. Finally, we used the search string to encompass not only the keywords “artificial intelligence” and “innovation” but also at least one of the components identified by the researchers, such as social, technological, sustainable/sustainability, personal, cultural, moral, and ethical. All these components will be further developed in our qualitative systematic review, given that these recurring components contribute to shaping different types of knowledge regarding AI-I.

The Web of Science (WoS) database was utilized to comprehensively search for research articles, review articles, and proceeding papers containing the following terms: “artificial intelligence” AND “innovation” AND (social OR technological OR sustainab* OR person* OR cultural OR moral OR ethical). Given the explicit focus on the economic side of previous studies centered on the AI-I field (Mariani et al., 2023; Truong & Papagiannidis, 2022), we mainly focused on the non-economic aspects of this field to highlight less popular research areas associated with AI-I. We have included all entries published between January 2000 and October 2023. The WoS database was chosen because of its comprehensiveness and frequent use in bibliometric analyses and systematic literature reviews (Benavides-Velasco et al., 2013; Khan & Wood, 2015; Köseoglu et al., 2019; Uyar et al., 2020).

Previous studies have primarily focused on screening processes within well-established research areas such as business, economics, and management (Mariani et al., 2023). In contrast, the present study expands the scope by incorporating less prominent areas within the social sciences and related fields. First, our analysis focused exclusively on articles, reviews, and procedural papers published between January 1, 2000, and October 31, 2023. Subsequently, we carefully examined all research areas on the WoS database, including all specific subfields of the social sciences and humanities. The most popular subdomains that match our search strategy are as follows: Business Economics (565 entries), Science Technology Other Topics (235 entries), Social Sciences Other Topics (128 entries), Educational Research (107 entries), Public Administration (88 entries), and others. In addition, this broader inclusion encompasses less popular entries related to AI-I research, such as information science (95 entries), sociology (26 entries), communication (21 entries), psychology (43 entries), and international relations (20 entries). Fig. 2 presents the complete range of research areas included in the analysis.

Following topic refinement, we identified 1225 publications worth investigating. As mentioned in the introduction, including less popular areas of AI-I research in our qualitative analysis for the systematic review is advantageous. This approach allows us to highlight the various levels of analysis related to AI-I. Upon gathering the 1259 publications, all three authors examined their abstracts and titles to identify the elements that were not relevant to the topic of our inquiry. The subsequent grouping of these publications is based on two criteria: 1) the nature of the innovation discussed concerning AI, and 2) the research area of which the publication is a part. After screening for potential abstract unavailability or content irrelevance, we eliminated 24 entries, resulting in a total of 1225 publications included in our analysis.1

Fig. 1 shows the distribution of the 1225 publications discussing innovation and AI in various related fields in the social sciences. The first publication related to the AI-I field in the analyzed period is that of Schulz-Schaeffer (2002), which was published in a German sociological journal and has the title “Innovation by means of concept transfer: Recourse to established knowledge in the production of technological innovations exemplified by multiagent research,” as well as the keyword “artificial intelligence” in the Keywords Plus category. The upward trend of publications in the AI-I field is evident. However, the number of publications per year declined to less than ten by 2014. A significant “leap” was observed in 2018, with 60 publications, compared to only 19 in the preceding year. This was followed by 92 publications in 2019, 175 in 2020, 291 in 2022, and 350 potentially related publications by the end of 2023.

Fig. 1.

Number of WoS publications on AI and innovation

Note. For 2023, we estimated that if the number of publications is maintained, a further upward trend is expected compared to 2022.

(0.11MB).

Fig. 2.

The research protocol applied in this study.

(0.55MB).

In order to perform the bibliometric techniques and meaningfully visualize the results obtained, we conducted a co-word analysis using VOSviewer (Waltman et al., 2010), a well-known tool in this field. Co-word analysis is one of the most popular bibliometric techniques, as proposed by Callon et al. (1986). This analysis examines the intensity with which two nodes, such as keywords, authors, journals, or countries, are used together, thereby reflecting the underlying structure of a particular knowledge field and the outline of some thematic clusters formed by these nodes. Thus, a high co-occurrence between two or more nodes highlights their strong relationship (An & Wu, 2011; Hâncean et al., 2021; Hu & Zhang, 2015; Ravikumar et al., 2015). Co-word analysis is already frequently used in a variety of bibliometric analyses, such as those related to intellectual capital (Faraji et al., 2022), corporate entrepreneurship (Funko et al., 2023), creativity (Zhang et al., 2015), auditing (Uyar et al., 2020), and others.

One notable observation derived from the WoS database is that author-selected words (author keywords) are accompanied by the words automatically rendered by the WoS (Keywords Plus) algorithm. Therefore, co-word analysis is useful due to its multifaceted capacity to identify thematic connections (Khasseh et al., 2017) and domains and trending topics (Dai et al., 2020; Faraji et al., 2022).

To explore the latent content of these publications, we used bibliometric social network analysis (SNA). In the context of SNA, nodes are represented by individuals (in our case, prominent researchers) or keywords, while ties represent the relationships between these nodes (Köseoglu et al., 2019; Yang et al., 2012). In our analysis, SNA demonstrates its utility by elucidating the conceptual map for a particular field or topic (Uyar et al., 2020). By applying SNA in our research on AI-I, we precisely investigate the main thematic clusters for this new and emerging field.

The 1225 articles included in our analysis provide a total of 5543 keywords. However, we included only keywords with at least ten occurrences, reducing the number to 162 that met the threshold. Fig. 3 illustrates the grouping of these words into relevant clusters. The most important words from AI-I research are presented in Table 1, with their relevance determined by both occurrences (oc) and total link strength (ls). Thus, the most important words, in addition to AI (oc = 538; ls = 2063) and innovation (oc = 358; ls = 1831), are technology (oc = 163; ls = 866), big data (oc = 125; ls = 742), management (oc = 99; ls = 627), future (oc = 96; ls = 577), performance (oc = 98; ls = 575), and impact (oc = 92; ls = 551).

Fig. 3.

Keywords co-occurrences (min. 10 keywords threshold).

(1.02MB).

Fig. 3 depicts AI as part of the green cluster, while innovation is part of the purple cluster. In general, the red cluster evaluates AI-I in relation to its related technological and sustainable implications, as well as other intra-cluster words such as sustainability, technologies, big data, and smart cities. Additionally, it examines the effects of AI operating independently, focusing on keywords such as industry 4.0, digital innovation, and smart. Similar to the green cluster, the red cluster contains 33 keywords, and those with the highest level in terms of oc and ls centrality are words from the business and management sphere, such as technology, adoption, and social media. The cluster also contains keywords such as machine learning, trust, and satisfaction. The third cluster (n = 27), denoted by the color blue, mainly includes themes related to consequences and implications regarding management, as well as the potential acceptability regarding the AI-I field through words such as impact, information, management, strategies, and perspective, among others. The fourth cluster (n = 22 items) is also popular and mainly discusses the nature of innovation regarding AI. Thus, noticeable words in this cluster are technological innovation, open innovation, emerging technologies, and creativity. Cluster 5 (n = 21), denoted by the color purple, contains terms related to policymaking and ethical implications through keywords such as education, challenges, and governance. It also contains terms related to the preservation of the environment: responsible innovation, health, and regulation. Finally, cluster 6, denoted by the turquoise color, contains sixteen keywords related to the digital side of AI conducive to innovations: digital transformation, digitalization, entrepreneurship, and value creation.

Fig. 4 presents a visual representation of the “time-zone” view related to the AI-I field, illustrating the evolution of knowledge within specific temporal frameworks. However, given the significantly low frequency of publications within our reference interval, specifically the first 100 publications out of 1225 published between 2000 and 2016, we were compelled to reduce the threshold from 10 to 5 occurrences of a keyword. In addition, given the scarcity of relevant words prior to 2018, our time-zone analysis in Fig. 4 included 2016 – 2023 as a reference interval. The “oldest” word in the overlay visualization includes technology diffusion, with 2016 being its average publication year. In 2017, we noticed knowledge trends like neural networks, sociology, and uncertainty, while in 2018, we noticed the emergence of keywords such as e-learning and information technology. In 2019, many new relevant keywords emerged, introducing different fields in which AI-I can be effective: triple-helix, robotics, and university. There were also some structural causes and processes regarding the AI-I performances: inequality, competencies, and dynamics. The year 2020 drew public attention to important keywords such as model, internet of things, and networks, alongside the potential ethical, moral, environmental, and legal implications through keywords such as ethics, responsible innovation, and diffusion. In the year 2021, new technological opportunities emerged in the AI-I field, such as technological innovation, the future, and sustainability. These opportunities were seen in diverse contexts where AI adoption could enhance innovative processes, such as industry 4.0, fintech, blockchain, and digitalization. There were also adjacent frameworks that explain the contexts in which users adopt different AI implementations leading to innovation, such as the technology acceptance model. By around 2022, there was evidence of isolated nodes that possessed the capacity to bring about potential social, technological, and organizational implications, taking more into account the climate crisis, such as transformation, corporate social responsibility, resource-based view, firms, business-model innovation, perceived risk, and so on. Thus, there is a growing interest in ecological and sustainable concerns related to the AI-I field, although these concerns may differ according to certain organizational characteristics, such as company size, category, and age (Cripps et al., 2020; Mariani et al., 2023).

Fig. 4.

Keywords heatmap (5 keywords threshold).

(0.75MB).

A co-citation analysis was utilized in this study, which involves a situation when a third document cites two or more documents together. Co-citation analysis retains its technical flexibility specific to bibliometrics since its networks might include references from authors, journals, and articles (Cobo et al., 2011). Table 2 and Fig. 5 present the most co-cited journals. For Fig. 5, we included a minimum number of 30 citations for a source, thereby keeping the 353 most co-cited sources out of a total of 29,765. Thus, in terms of total ls, we note that the most co-cited sources regarding AI-I are Technological Forecasting and Social Change (ls = 77,522), Journal of Business Research (ls = 71,062), Sustainability (ls = 52,864), and Journal of Cleaner Production (ls = 47,618). The Journal of Innovation and Knowledge is also included in this ranking, having 73 citations and a link strength of 3686. In terms of author co-citations, we observe that the most central authors in the co-citation network (Fig. 6) are Erik Brynjolfsson (ls = 1869), David Teece (ls = 1857), and Anjan Chatterjee (ls = 1694).

Fig. 5.

Co-citation of sources map (30 citations threshold).

(0.39MB).

Fig. 6.

Co-citation of authors.

(0.64MB).

In terms of citations, the most cited studies regarding the multiple implications of AI-I are as follows: “Brave new world: service robots in the frontline” (Wirtz et al., 2018), “Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy” (Dwiveldi et al., 2021), and “Applied artificial intelligence and trust—The case of autonomous vehicles and medical assistance devices” (Hengstler et al., 2016). Tables 3 and 4 and Fig. 7 present a visual representation of the most important studies, journals, and authors in the field. In terms of citations, the most relevant journals are Sustainability (nc = 1359), Technological Forecasting and Social Change (nc = 1295), International Journal of Information Management (nc = 927), Journal of Business Research (nc = 667), and Journal of Cleaner Production (nc = 353). Concerning documents related to the diverse aspects of AI-I, the most prominent journals are Sustainability (nd = 120), followed by Technological Forecasting and Social Change (nd = 35), IEEE Transactions and Engineering Management (nd = 25), and Journal of Business Research (nd = 16).

Fig. 7.

Authors’ citations heat map (10 citations threshold).

(0.3MB).

Table 5 and Fig. 8 present the most important countries in AI-I research. Table 5 demonstrates significant differences between countries regarding classification by citations and classification by documents; however, the United Kingdom ranks first in terms of citations (nc = 6007). The next countries in terms of citations are the United States (nc = 3795), Germany (nc = 2091), India (nc = 1811), and Australia (nc = 1729). In terms of documents, the most visible countries are China (nd = 192), the United States (nd = 189), the United Kingdom (nd = 161), Spain (nd = 95), and Italy (nd = 94). While previous studies (Faraji et al., 2023) identified significant discrepancies between research in the United States and other countries regarding the study of intellectual capital, our observation indicates that these disparities are less pronounced in the study of AI-I. This finding suggests a significant degree of academic competition within the field of knowledge pertaining to this subject matter.

Fig. 8.

Main countries in the AI-I field, weighted by citations (minimum 10 documents and 10 citations threshold)

Note. If we count for the United Kingdom, it reaches the first place in terms of citations (with 6007) and 161 documents.

(0.6MB).

Table 6 and Fig. 9 present the most prominent organizations in this knowledge field; however, there are some notable differences in terms of citations and documents. The most important universities in terms of citations are Loughborough University, UK (nc = 1450); Swansea University, UK (nc = 885); the National University of Singapore (nc = 815); the University of the West of England, UK (nc = 812); and the Indian Institute of Technology (nc = 796). In terms of documents published in this knowledge field, the most important universities are Bucharest University of Economic Studies, Romania (nd = 14); Oxford University, UK (nd = 12); University of Queensland, Australia (nd = 12); Tsinghua University of China (nd= 11); and University of Nicosia, Cyprus (nd = 10).

Fig. 9.

Main organizations in the AI-I field, weighted by citations (minimum 5 documents and 5 citations threshold).

(0.49MB).

The social dimension is popular in the study of AI-I as it examines the heterogeneous human responses that occur with the planning or implementation of these innovations. Thus, through this dimension of knowledge, the antithesis between certainties and uncertainties reflects exactly the aspects through which individuals go through the processes related to AI-related knowledge and, implicitly, the potential innovations in the field of AI. In this sense, Hengstler et al. (2016) highlight the recurring skepticism in relation to AI-I automation while also offering certain ways in which individuals can overcome potential AI-related uncertainties. For example, trust can be gained through increased knowledge of the assembly itself and by assimilating how firms communicate about these innovations. Another relevant article in this field is Tussyadiah's (2020) review on automation in tourism. The author emphasizes the significance of understanding AI automation as a social phenomenon. Thus, the obvious priorities in terms of the social dimension relate to designing a useful AI, along with highlighting the main risks and benefits associated with tourism automation. However, there are some areas in which the social dimension associated with AI-I is rather paved with uncertainties, particularly when it comes to the use of AI in performing criminal acts. King et al. (2020) show that AI in crime is such a new subfield that its social consequences are difficult to predict in the near future.

Other sociologically oriented studies examine AI-I as an important precursor to realizing a social transformation. Boyd and Holton (2018) argue that innovations that facilitate the advancement of AI contribute to knowledge processes maintained by symbolic and discursive forms of power. Consequently, the adjacent historical dynamics are centered on the epistemological and empirical problems through which individuals come into contact with the specific uncertainties related to AI-I.

While the social dimension is popular in research areas such as social sciences, social issues, sociology, and public administration, the sustainable dimension is frequently addressed in fields such as business economics, social issues, and geography. Each area allocates different knowledge processes that involve certainties and uncertainties associated with AI-I.

In business-related contributions, Di Vaio et al. (2020) investigate the primary models through which AI can be trained to encourage the principles of sustainable development. Thus, the authors observe that the systematic implementation of the UN's Sustainable Development Goals is contingent upon the particular uncertainties associated with individuals’ mentalities while also influencing production and consumption standards. Denicolai et al. (2021) find that the relationship between sustainability, internationalization, and digitalization among small and medium-sized enterprises is not linear but depends on the company's expansion outside the nation; as the company internationalizes, digitalization and sustainability start to compete. According to Chauhan et al. (2022), digital technologies benefit from the principles of the circular economy; this allows the Internet of Things and AI to work together to solve potential uncertainties, such as policy-related issues, information vulnerabilities, and predictability deficiencies.

From an urban geography perspective, Cugurullo et al. (2021) demonstrate how the transition to autonomous vehicles also changes the design of the road infrastructure. At the same time, this transition gives rise to certain urban politics that go beyond the urban homogeneity specific to the former socialist regions (Chelcea et al., 2021). In addition, the experimental principles related to global cities constitute important sites of uncertainty (Macrorie et al., 2021) such that processes related to responsible urban innovation continue to represent a challenge in the context of the most heterogeneous social infrastructures.

Among the non-economic dimensions addressed in this systematic review, the technological dimension is the most frequently addressed, being found in most research areas within the social sciences and humanities. Thus, within the science and technology domain, several estimates are made regarding automation in the energy sector in the next five to ten years. Tabor et al. (2018) estimate the emergence of certain tools, such as virtual screening or automation-based synthesis planning, aim to reduce emissions from the energy sector. Lin et al. (2022) propose a technological model centered on machine learning to create high-performance classification systems within supply chain management.

In the educational field, Tamayo et al. (2020) discuss a case study of EconBot focused on making distance learning accessible to students enrolled in the microeconomics class. Krouska et al. (2019) discuss the emergence of social networking-based learning, a technological innovation centered on e-learning where students’ social networks play a key role. The authors highlight the potential uncertainties associated with social networking-based learning, as it currently lacks alternatives tailored to the needs of each student. Thus, the emergence of the technological dimension in the educational sector legitimizes the need for a convergence between creative design and a focus on technological automation but also grants an increased role for social media affordances in the educational process (Al Hashimi et al., 2019; Hosszu & Rughiniș, 2020). In this technological context, Garcia-Peñalvo (2023) examines the implications and consequences of ChatGPT adoption in the educational sector, highlighting the nature of the tsunami effect that can easily lead to disruptive innovations, particularly in the absence of coherent regulations that explicitly address the certainties and uncertainties related to language models in education.

In the geography area, Acemoglu and Restrepo (2020) examined the uncertainties surrounding AI-I, finding that the human workforce is in a systematic regression due to the automation processes that have taken place in a manner that prevents the design of new tasks where human employees can prove productive. Pink et al. (2018) argue that the convergence of mobile phones and cars, both of which have become highly digitized in recent years, can create favorable conditions for autonomous driving while facilitating the interaction between completely different technologies. Armour and Sako (2020) highlight the sociological implications of the technological dimension surrounding AI innovations, demonstrating that AI can still not replace human activity in the legal sector. Given the social and legal implications of improperly automated decisions, it is noted that the legal sector is still one where AI evaluations must be treated with caution.

Technological innovations are also examined in the communication area, investigating how certain ads can generate human reactions when AI techniques are involved. Campbell et al. (2022) examine the emergence of synthetic ads, which refer to content created based on data modification through AI. For instance, in the context of an increasingly accelerated presence of deepfakes, it can be seen how future uncertainties generated by AI-based content will significantly affect consumer activity. In addition, García-Orosa (2021) identifies the recurrent technological innovations in the sphere of e-democracy, identifying a trajectory that starts from 1990, when political actors began their activity online, to 2016, when the Cambridge Analytica scandal highlighted the rise of AI in the production of fake digital contents intended to increase disinformation in a political context.

The personal dimension is a relevant form of knowledge regarding AI-I since it sheds light on both the heterogeneous nature of innovations and the personalization component that some innovations make available via AI. However, only a few studies discuss this dimension, and they are found mainly in research areas such as psychology and social issues.

For example, Samuel et al. (2021) discuss the dilemma surrounding automated personalization in the advertising sector. They found that while organizations offering personalized assets may appeal to consumers in many ways, these advanced personalizations are made by obtaining detailed information about their activity, which creates feelings of fear and destruction. Therefore, fully understanding the fears evoked by consumers could enable a deeper understanding of the main barriers to AI-based personalization.

Furthermore, when social issues are involved, different personal valences of AI-I are discussed. For example, Steen (2021) offers some reflections focused on the latest advances in the field of AI, emphasizing the adoption of slow innovation and aiming to create a framework of knowledge that emphasizes the main uncertainties surrounding these innovations. Thus, elements such as uneasy questions and vulnerable experiences must be considered to encourage responsible innovations that contribute to a more human-centric version of AI.

This dimension is debated in multiple research areas involving AI-I, such as science and technology, sociology, educational research, public administration, and philosophy. In science and technology, Winfield and Jirotka (2018) investigate the need for ethical governance for AI systems. Their aim is to establish public trust in AI and promote social solidarity regarding potential innovations based on automation. By shifting the interest from an information-centric approach to a data-centric one, Floridi and Taddeo (2016) highlight data ethics as a relevant epistemological branch, whereby the interest in the ethical dimension of AI-I becomes holistic and inclusive. Floridi (2018) distinguishes between hard ethics, which creates the technological premises for the implementation of legislation in the field of AI, and soft ethics, which takes place after legislation is established and legal compliance is achieved.

In the educational sciences, Grunhut et al. (2022) demonstrate the ethical implications of doctors adopting AI in the medical education curriculum. Moreover, conservative principles prevent the incorporation of AI innovations in the educational sector due to uncertainties that revolve around the lack of technical expertise among physicians and the lack of interest of medical students in learning about AI.

Kravchenko and Kyzymenko (2019) introduce a philosophical approach centered on ethical issues related to the Fourth Industrial Revolution. Thus, various theories from the sphere of the socio-humanitarian paradigm overlap with the empirical concerns surrounding automation, encouraging the adoption of “responsible innovations” as a relevant form of knowledge involving AI-I to address recurring social problems such as technological unemployment.

The cultural dimension is the least prevalent among AI-I from a quantitative perspective; however, we encounter cultural aspects of knowledge in various research areas, such as business economics, information science/library science, public administration, and sociology. Within business economics, several studies (Fountaine et al., 2019) argue that the technological factor is not the biggest obstacle to an organization adopting AI innovations but rather the cultural factor. Specifically, when adopting automation-centric innovations, it is important to consider the cultural barriers of traditional mindsets, along with the constant concern of job loss.

From another point of view, Feijóo et al. (2020) believe that adopting AI on a large scale can be achieved through international policy coordination that considers cultural variations in certainties and uncertainties, which broadly revolve around principles related to efficiency, surveillance, and privacy protection. From the sphere of public administration, Lazzeretti (2022) notes that the cultural aspect is at the confluence of digital transformation and social change, indicating that both the certainties and uncertainties in the sphere of AI-I relate to the preservation of cultural elements. When innovations lack an emotional component, people can no longer fulfill their expectations of cultural proximity (Tubadji et al., 2021). From a sociological standpoint, Brooks et al. (2020) also demonstrate how cultural pressure prevents the rapid adoption of automation and AI transformations, and such resistance systematically challenges different business models. Therefore, the discourse around the cultural dimension confirms the relevance of establishing connections with other components, such as social, technological, and sustainable aspects. Furthermore, we observe that each dimension contributes to shaping particular types of knowledge, with their own certainties and uncertainties regarding the adoption of AI-I. Thus, it is imperative to examine these dimensions collectively.

Our analysis has certain limitations, specifically in regard to the selection of relevant spheres in the study of AI-I. While attempts are made to mitigate this bias by investigating recurrent factors visible in comparable studies (Cheng et al., 2023; Mariani et al., 2023), every search strategy has an evident impact on the main findings obtained. This is the first study to explicitly discuss the non-economic aspects related to AI-I in the social sciences. It places emphasis on the knowledge components centered on certainties and uncertainties, thus highlighting the complex nature of the factors involved in human decisions concerning AI.

This study has specific implications for researchers and practitioners in the field of AI. At the epistemological level, we highlighted the heterogeneous nature of the debate surrounding innovations, demonstrating how each non-economic factor comes with its own certainties and uncertainties in the context of AI automation. Additionally, these findings offer implications for policymakers. They demonstrate that AI should adopt principles related to responsible and slow innovation, taking into account both cultural and ethical factors. By doing so, these innovations can benefit from a level of greater social acceptance.