Public AI imaginaries: How the debate on artificial intelligence was covered in Danish newspapers and magazines 1956–2021

In recent years, the digital agenda in Denmark and elsewhere has been marked by the launch of a number of public strategies focusing on developing ethical guidelines regarding the implementation of artificial intelligence (AI) in society (Robinson, 2020). In these guidelines, AI – an umbrella term covering a variety of technological solutions and machine learning methods – is characterised by increasing investment and optimism, but at the same time is linked to a number of cases where its use in society has been regarded as problematic (Kerr et al., 2020). Such aspects play a central role in public meaning-making and people's encounters with AI. Public media venues, including so-called legacy media, have played a central role in covering the topic and have thereby fuelled considerable debate on AI in Denmark and globally.

From a historical perspective, printed newspapers and technology magazines have been particularly prominent in this regard, and active for a much longer period than digital media outlets, serving to set a public agenda both before and during the period when AI was coined as a term in 1956 (Markoff, 2015). Thus, newspapers and magazines have played a central role in shaping collective perceptions and public imaginaries of AI. This has been combined with political strategies putting AI on the agenda, and as a result, AI has become a much-discussed topic publicly. However, if we are to understand the importance of AI in society and use it in responsible, human-centric ways, we must understand the background against which the AI debate has taken place.

Being one of the world's most digitised countries (e.g., European Commission, 2020), Denmark presents an interesting case for studying public imaginaries of AI. As a small, innovative, and internationally oriented country, Denmark is known for prioritising new technology and quickly adapting to new technological opportunities (Digital Growth Panel, 2017; Ministry of Industry, Business and Financial Affairs, 2018). With official Danish political strategies putting the digital transformation of the welfare state high on the agenda, Denmark can be regarded as a visionary country that is perhaps ahead of others in relation to public debate in this area. The Danish context thus makes a particularly relevant case for studying how current public AI imaginaries about human–machine relationships are constructed, what they entail, and how they come to matter in society.

This article offers a historical analysis of the Danish print media's coverage of AI, which has been crucial in forming the background for a collective discussion of AI – reflected in what I refer to as the construction of public AI imaginaries about human–machine relationships. By way of a critical discourse analysis, supported by a quantitative content analysis of articles on the subject of AI, I investigate how such imaginaries have been constructed in Danish newspapers and magazines. I further identify how they have been significant in shaping the current strategic plans for AI in Denmark.

Following the launch of several political AI strategies at national (Dutton, 2018) and transnational levels (European Commission, 2018), an emerging body of academic literature is focusing on the intersection of AI imaginaries, discourses, and framings. Some of this literature deals with AI strategies in national settings, such as Australia (James & Whelan, 2021), Germany (Köstler & Ossewaarde, 2022), the UK (Brennen et al., 2020), and China (Zeng, 2021), whereas other studies take a global or transnational approach (e.g., Bareis & Katzenbach, 2021; de Sousa et al., 2019; Jobin et al., 2019; Natale & Ballatore, 2020; Paltieli, 2021; Radu, 2021; Roberge et al., 2020; Sinanan & McNamara, 2021). For instance, Bareis and Katzenbach (2021) conducted an empirical study of how AI imaginaries are being presented in the national AI strategies of China, the US, France, and Germany. Their analysis reveals remarkable differences in the construction of AI imaginaries, although they identify a common rhetorical trait running through the strategies of framing the imagined AI future as an uncertain but inevitable path (Bareis & Katzenbach, 2021). These findings correspond with the ambiguity identified by Roberge and colleagues (2020) in their study on public AI discourse in three cases of public debates on AI imaginaries. As they argue, this ambiguity “comes with an uncertain and contingent future” and goes hand in hand with the “multifaceted hype” surrounding AI imaginaries (Roberge et al., 2020: 4).

However, previous findings also suggest that although AI imaginaries often operate on a global or transnational scale, they come to prominence when they are performed and articulated within national and local societal institutions and contribute to actual social practices (e.g., Schiølin, 2020). This includes national news media institutions, as shown in, for instance, Brennen and colleagues's (2020) study on AI imaginaries in British news media and Vicente and Dias-Trindade's (2021) study on how AI is framed in the Portuguese national press. Both studies point to a co-constructive link between national public news media institutions and national governmental strategies in the debate and digital agenda surrounding AI. Likewise, they identify a “persistent ‘hubris’” articulated though rhetorical and discursive patterns in news media that is based on the myth that machines will soon be able to outperform and replace human intelligence (Brennen et al., 2020: 33; see also Natale & Ballatore, 2020).

Notably, despite being a digital global frontrunner, Denmark is rarely studied as an individual case in the literature on the construction of AI imaginaries. Schiølin (2020), though, examines the discourses and imaginaries of the fourth industrial revolution (4IR) related to the development of AI in a specifically Danish context. He uses the concept of “future essentialism” – the “discourses, narratives, or visions that, through different means and practices […] produce and promote an imaginary of a fixed and scripted, indeed inevitable, future” – to analyse the performative role of these imaginaries, in which they ultimately define and govern specific future scenarios (Schiølin, 2020: 543). One example Schiølin cites is how Danish politicians in the Siri Commission (a Danish commission set up for developing recommendations regarding AI and digitisation) reproduce internationally held 4IR imaginaries by simply asking “how to harness and steer this predetermined future” for Denmark, instead of “questioning the premises of 4IR and its desirability, or envisioning other possible destinations for the future” (Schiølin, 2020: 555). Hockenhull and Cohn (2021) approach imaginaries related to Denmark's AI agenda in an ethnographic study of private and public Danish tech events. As found in some of the above-mentioned studies on national AI strategies, Hockenhull and Cohn (2021: 304, 317) emphasise the rhetorical nature of constructing AI imaginaries that are “taken up in practices” as they are socially performed by corporate agents towards a wider public at Danish local tech events and conferences, through “different articulations of hype”. This article complements and further expands such research by casting a wider net across AI as it has been articulated in public debate over time and exploring how this has influenced current AI strategies.

The general discussion of contemporary scholarship on the digital agenda and AI strategies indicates that the national setting serves as an important venue for studying how specific AI imaginaries are constructed, what they entail, and how they contribute to formulating current national AI strategies, as well as how national news media play a role in shaping the public construction of AI imaginaries. In this article, I investigate Danish news media, technology magazines, and official strategies as important places to look for ways in which the concepts of AI, human, and machine are collectively imagined.

Jasanoff (2015a: 11) argues that people's articulation of imaginaries about technological phenomena are sociotechnical by nature; they are constructed through social practices and “characterised by a number of specific collective visions of what is in the interest of society”, including individual and collective perceptions about technology's future impact on humans. More specifically, Jasanoff refers to them as “sociotechnical imaginaries”, defined as,

collectively held, institutionally stabilized, and publicly performed visions of desirable futures, animated by shared understandings of forms of social life and social order attainable through, and supportive of, advances in science and technology.

Central to this definition is that imaginaries can be articulated collectively within smaller groups in society. In addition, several different and conflicting imaginaries can coexist in a way where societal power institutions, such as the media, have the ability to prioritise and legitimise certain imaginaries above others (Jasanoff, 2015a, 2015b).

In line with Jasanoff's (2015a, 2015b) definition of sociotechnical imaginaries, scholars have studied how (big) data imaginaries (e.g., Beer, 2019; Lehtiniemi & Ruckenstein, 2019; Olbrich & Witjes, 2016; Ruppert, 2018; Williamson, 2017) or algorithmic imaginaries (e.g., Bucher, 2017; Lomborg & Kapsch, 2019; Mager, 2015; Seaver, 2017) feed into social change. For instance, Beer (2019) highlights the sociotechnical nature of imaginaries in an analysis of how visions of (big) data futures instigate social practice when they are deployed as “actual data solutions” within the data analytics industry. In a similar vein, Bucher (2017) examines the “productiveness” of public text-based reflections on algorithms and on the ways in which public spheres (i.e., spaces where people meet and discuss algorithms) shape not only the way people think and talk about algorithms, but also the practical and technological aspects of the algorithm itself. This correlates with Jasanoff's (2015b) description of how such articulations are part of an ongoing, dynamic, and socially constructed process, where specific perceptions of technology in future scenarios emerge and become embedded within society; over time, some of these perceptions will stabilise and become more prominent than others.

Furthermore, Bucher (2017) points out how people tend to construct their own imaginaries in situations where technology is considered non-transparent or too complex to comprehend. As such, the concept of imaginaries and their sociotechnical nature can be used as a framework for studying advanced technological phenomena and their social impact, even when they are difficult to gain access to (Bucher, 2017). This resonates with Lomborg and Kapsch's (2019) way of approaching the same phenomena. Rather than trying to “open the black box itself”, they use the concept of decoding to understand how Danish media users imagine and respond to algorithms in their daily lives (Lomborg & Kapsch, 2019: 2). Thus, literature on sociotechinal imaginaries invites us to study how they are publicly articulated, as this is essential for understanding “technology in the broad sense of shaping our society” (Lomborg & Kapsch, 2019: 4).

Because of the rapid and diversified nature of how advanced technology is currently being developed, we cannot only study the sociotechnical imaginaries in isolated fragments of the advanced technological systems of today, such as big data and algorithms. Instead, we must direct our attention towards the broader range of technological phenomena and solutions encompassed by “AI” in order to better understand how imaginaries about today's systems are constructed and how they might feed into actual social practice, and ultimately, lead to change. This includes taking into account the different technological approaches to AI that have emerged as different sets of collectively held visions and imaginaries of AI.

As part of the definition of sociotechnical imaginaries, Jasanoff (2015a) describes how societal power institutions, such as media institutions, influence the positioning of particular imaginaries over others and thereby influence which future visions of technology come to the public's attention. As I regard media as a central empirical source for studying imaginaries, for this study I collected a range of articles from Danish newspapers and technology magazines as well as official Danish strategies on developing and implementing AI in society. The data corpus covers the persistence and temporal development of media coverage about AI and the human–machine relationship, which are crucial to establishing sociotechnical imaginaries (Jasanoff, 2015b). I take Danish newspaper and magazine articles to represent general societal communication, influencing the way in which humans perceive and understand themselves as part of the world they live in and the way in which they perceive technology (Jensen, 2013). Furthermore, official reports published by the Danish government give insight into the particular AI imaginaries that are being stabilised and legitimised in Denmark and elsewhere (Jobin et al., 2019).

To build the corpus, I carried out a systematic selection of all newspaper and magazine articles on humans, machines, and AI from January 1956 to January 2021. I searched for and extracted these articles from the Mediestream and Infomedia archives (Mediestream, n.d.; Infomedia, 2021), which offer digital access to news published in Denmark. Mediestream was used to search for articles from 1956 to 1990, and Infomedia for articles from 1990 to 2021. While it is possible that the sampling was affected by Mediestream's and Infomedia's different archiving structures, I allowed an overlap of ten articles to ensure that both archives provided the same results from 1990 onwards.

To consider historical changes in the discourse associated with the debate on AI and the human–machine relationship, the archives were searched for longer articles (700+ words) written in a Danish context. More specifically, the archives were searched for articles containing the Danish words for the following: “human”, AND “electro brain”, “robot”, “machine”, “data machine”, or “electric calculator”, AND “artificial intelligence” (including Danish variants “kunstig intelligens” and “artificiel intelligens”), “simulated intelligence”, “machine learning” (including Danish variants “maskinlæring” and “maskinel læring”), “expert system”, or “expert support system”. The first group of keywords in the search string included several words representing the historical changes in how a computer has been referred to in Danish. In a similar vein, the second group of keywords represented the historical changes in the Danish language for what is today typically known and referred to as artificial intelligence. The keywords were identified via an initial chain search in literature on the field, including that written by the pioneer in Danish computer science research, Peter Naur (1954).

The sampling of official strategies was carried out by searching for publications on the website of each Danish ministry. Due to the narrow search options on these web-sites compared with the two media archives, the websites were searched for strategies containing one of the Danish words for AI, as listed above, within the same time period as the gathered articles. The sampled strategies from public agencies address current societal matters directly, as they concern the contemporary implementation of AI in Danish society.

The search resulted in a total of 253 articles (217 newspaper articles and 36 magazine articles) and four public strategies. General patterns in the articles were identified using quantitative content analysis, which then provided the background for a deeper investigation into the qualitative nuances of public discussions about AI, humans, and machines in the articles and strategies. An initial pilot test (coding 4% of the articles found) was conducted in order to create a coding manual for the categorisation of the patterns in the remaining articles. This resulted in eight consistent categories: media outlet, date, sentiment, perspective, context, actor, human, and technology (since these categories form the background for the qualitative analysis, the overall findings from the quantitative content analysis are not reported further). I then coded the full corpus of newspaper and magazine articles according to these categories. The results showed that the coverage of AI and the human–machine relationship increased significantly over the period analysed across several Danish media outlets, although this is far more pronounced in the daily and weekly newspapers than in the technology magazines included in the search (see Table 1).

The articles spanned from the 1980s to the 2020s. Thus, we can see a time lag from the time when AI and IA and the focus on the human–machine relationship developed in the academic field until it reached the public arena in the print media of the 1980s (see Table 2).

From the 1980s onwards, the overall number of articles increases each decade (except for the 2020s, where only two years were included, resulting in 12 articles), indicating a growing interest in AI, which follows the generally increasing research into and public interest in AI, for example, as seen in recent publications of official strategies for developing AI in Denmark and elsewhere (Jobin et al., 2019; Robinson, 2020). Based on the general patterns across the articles sampled for the qualitative analysis, Fairclough's (2013, 2015) critical discourse analysis approach was used to address the discursive aspects of the AI imaginaries articulated in the articles. Therefore, using Fairclough's (2013) three-dimensional framework, the critical discourse analysis addressed 1) the text dimension, 2) the discursive practice dimension of the articles, and 3) the social practice dimension of the strategies. This made it possible to examine the chains of equivalence connected to AI as a nodal point for analysing linguistic and discursive patterns, and later to gain an understanding of the significance of such AI imaginaries for larger societal structures. Any headlines and qhotations from newspaper and magazine articles used in the analysis and in the following discussion are my own English translations of the original Danish versions.

Using the text dimension as the first lens of analysis made it possible to address the articles at word level and identify how particular ways of communicating about AI and the human–machine relationship are initiated. For instance, in the headline “Increasingly clever computers on the way to conquering the world” (Alt om Data, 2017a), the chain of equivalence “increasingly clever” that is connected to the computer as a nodal point was interpreted as a personification in the way the computer is being associated with a certain type of knowledge and reason usually used to describe human qualities.

The second analytical lens, the discursive practice dimension, allowed the examination of the composition of words, their conveyance of attitudes, and the use of intertextuality. Using this lens, the sentence above was interpreted as an expression of how the computer is being imagined as a powerful and particularly independent entity that will be able to supersede humans. The description of its path towards “conquering the world” was additionally interpreted as a way of painting a dystopian and uncertain picture of how technology might gain power over humans, which was supported by other statements from the article on how robots “are ready to make our decisions, take our jobs, and maybe soon our entire livelihood” (Alt om Data, 2017a).

Lastly, the social practice dimension made it possible to unmask the ways in which the use of language becomes a social practice as it is linked to a specific societal and historical context and used by the Danish authorities to prioritise certain AI imaginaries above others. For instance, the same type of uncertainty and futuristic image painted of AI and its ability to overrule humans, identified in the above example, was likewise identified in the Danish AI strategies: “It is impossible to say for certain when – or whether at all – we will reach a point where computer technology can reason at a human level” (Danish Government, 2019: 6). However, it was followed by the idea that AI should not replace, but assist humans: “Artificial intelligence should help us analyse, understand and make better decisions. […] The technology cannot, and should not, replace people or make decisions for us” (Danish Government, 2019: 7). Using this third analytical lens, it was thereby possible to examine how the latter idea, which resonates with IA, was the one being legitimised by the Danish authorities. Thus, the following analysis focuses not only on the relatively vague notions of what AI is, but also on how the human–machine relationship itself is framed from the two parallel overarching sociotechnical visions for the practical development of advanced technology: AI and IA.

The analysis indicates how language use over time establishes competing notions of the human–machine relationship in the context of AI development. These notions can be seen as sociotechnical imaginaries – as AI imaginaries about the human–machine relationship. At the same time, these imaginaries have left the term AI itself relatively vague even after decades of public debate. Thus, the analysis paints a picture of a public debate on AI in Denmark where it is still relatively unclear what is meant by AI. There seem to be two aspects to this. Firstly, it seems that the fictional universe, drawing on traditional notions of AI, can help us understand advanced and abstract technological developments and applications. At the same time, it has helped to generate urgent public debate about the human component in AI development and use. Secondly, however, the vagueness of what is meant by AI and IA in the debate could almost be seen as an unconstructive attempt to cover up a loaded term like “AI” by connecting it through language use with human-centered chains of equivalence that popularly go under the name “ethical AI”.

Findings from previous literature on the “persistent ‘hubris’” of AI (Brennen et al., 2020: 33) are not uniformly confirmed by the analysis, which, in contrast, reveals a more multivocal public debate on AI. The ambiguity of AI, however, is confirmed in the analysis, in line with the previous study by Roberge and colleagues (2020) on public AI discourse, which looked at three cases of how the public perceived AI imaginaries. The different positionings expressed indicate a malleable AI imaginary, reflecting the nuances brought out by collective perceptions of AI, which in turn feed into the public strategies formulated as a result of the AI debate. In addition, they show how a particular imaginary, rooted in the public debate and based on IA, is officially legitimised in a Danish context, where it impacts the future public planning projects and social practices presented in the strategy documents. Although the analysis cannot anticipate how the discursive frameworks in the articles themselves will lead to change, they do reveal underlying discourses in the public debate on AI that compete to set the direction for the further development of AI in Denmark. Despite the fact that the strategy documents tend to not refer to IA explicitly, the original ideas associated with the IA vision seem to feature consistently in the strategies. As shown through the specific language used, these ideas have contributed to setting a “human-centred direction” for AI focusing on combining the strengths of humans with those of machines in a “symbiotic partnership” (Licklider, 1960).

Although newspapers and magazines are not the only co-creators of future systems based on AI, they do fuel public debate on the issue, thereby informing the process by which people can form opinions about AI in society. In addition, as in this case, such articles can highlight the human element in AI's development, including ethical aspects of responsibility and fairness. The analysis of linguistic and discursive patterns in the articles and strategies thus illustrates how language can be a powerful tool, particularly in the context of societal institutions, such as the media or political institutions. It can thereby become embedded within social practices, ultimately resulting in social change.

The nuances brought out in the construction of AI imaginaries illustrate how the ways in which we talk about technology – and the ways in which AI is portrayed through specific institutional use of language – shape how future strategic visions of AI are created. However, this is not only the case for the construction of AI imaginaries; in the construction of, for example, algorithmic imaginaries, Lomborg and Kapsch (2019) note how opinions are formed, both individually and collectively, and how algorithmic systems are decoded to shape future visions of such systems. This would seem to indicate that the way in which we communicate about AI, humans, and machines influences how we envision ourselves as humans in relation to AI, and how we use language in planning our future societal practices in relation to AI. The discursive ambiguity identified in the analysis, which is in line with previous literature (e.g., Roberge et al., 2020), suggests that different AI imaginaries are thriving and stimulating public debate on the topic. Yet, the continuous reference to fictive examples might arguably also prove unconstructive in furthering public understanding and debate, in that it risks causing uncertainty rather than describing accurately how human-centered AI might develop and operate in empirical reality.

From the analysis, we can conclude that as the strategies have so far stabilised a human-centred imaginary around developing and using AI in practice in a Danish context, they appear to reflect typical principles of IA rather than those of AI. Thus, with its emphasis on collaboration and optimisation, the notion of IA would seem to both reflect more accurately the visions legitimised in the strategies and highlight discussions about ethical principles more than the mechanically oriented AI perspective does, even though AI is the term used in the articles and strategies. The notion of AI is, therefore, often misleading; indeed, as Guszcza and colleagues (2017) note, the systems we have today – and want to develop in the future – are a long way from any system based on the original vision of AI. This is reflected in the analysis in the present study through the linguistic conflict arising from the use of language, where “AI” is used in references to the human-centric principles of IA while simultaneously being associated with futuristic perceptions of traditional AI principles. The actual imaginary resulting from this can be rendered as an IA imaginary – an ethical direction for AI.

As two different sociotechnical imaginaries, the notions of AI and IA can in themselves inform public debate. Understanding the link between language use and social change related to AI is essential if we are to understand “technology in the broad sense of shaping our society” (Lomborg & Kapsch, 2019: 4). Here, IA might offer a specific term that enables what Kerr and colleagues (2020: 10) refer to as an “ethics discourse”, that is, another way of imagining technology. In employing this kind of discourse, the strategies analysed in this study not only provide “assurance” about the development of ethical AI in society (Kerr et al., 2020: 10), but also function as an effective tool in imagining and planning the actual development of future human-centred solutions. As Jarrahi (2018: 579) claims, “what is lacking in this old discourse, as well as with the recent attention paid to AI, is a discussion of how the unique strengths of humans and AI can act synergistically”. Instead of the preoccupation with machines replacing all facets of human intelligence, imaginaries of the human–machine relationship based on IA can serve to set a more ethical direction for the future (Jarrahi, 2018).

The overall quantitative finding that AI has become an increasingly discussed topic in Danish printed news media served as a point of departure for the qualitative, critical discourse analysis in this article. The coverage of AI has been crucial in forming the background for a collective discussion of AI which is reflected in public AI imaginaries about the human–machine relationship. The analysis has provided insights into how several coexisting and conflicting discursive positionings are being constructed through a use of language that expresses positive, negative, and neutral AI imaginaries about the human–machine relationship. Whereas the positive imaginary focused on how AI can be used in various contexts for the benefit of human development, the negative imaginary was characterised by a dystopian view on the social ramifications of AI. The neutral imaginary portrayed AI development as uncertain and marked by an underlying notion of passivity towards the lack of human involvement.

Additionally, the analysis demonstrated that a futuristic discourse is still shaping public perceptions and constructions of AI imaginaries, although the vision for social change brought about by AI is mostly described in ways similar to the vision of IA. Although the strategies continue to use the AI term, the analysis further demonstrates how it is principles of IA that are being reflected and stabilised in the Danish strategies for AI in practice. Thus, despite these IA principles being legitimised with a preliminary endpoint in the official strategies based on the public debate in the articles, the analysis identified a lack of clarity about what is meant by AI, in both the articles and official strategies.

Ultimately, although AI in the public debate in Denmark is not characterised by persistent hubris, it is rather ambiguous in terms of what we mean when we talk about AI, how it may be used in the future, and its possible effect on society. This provides insight into how language can be a powerful tool in the construction of imaginaries that, in the long run, have the ability to shape practical solutions. It also illustrates the unconstructive nature of linguistic ambiguity in generating persistent uncertainty around AI, rather than committing to the construction of an AI imaginary that, at its core, serves human interests. Here, it seems that the notion of IA might be a more appropriate term for the human-centric visions expressed and legitimised in the strategies, serving as a linguistic tool to implement an ethics discourse to ensure fair and responsible human–machine configurations in society.

However, by applying an alternative notion for AI, the limitation is that we risk nudging future AI strategies and related social practices to only focus on what is considered to be IA, instead of focusing on making the advanced AI applications that have already been developed more human centric and true to the original visions of IA. Thus, to pave the way for new and more constructive and realistic ways of imagining human–machine relationships, we need more concrete examples of how AI is used in everyday contexts in partnership with humans. Future research looking at empirical cases of human–machine collaboration can help demonstrate and describe how this partnership works in practice and how it is applied to serve human interests.