Chance or Challenge for the Future of Work?
Frames of Artificial Intelligence in the News
Graduate School of Communication
Master’s programme Communication Science: Corporate Communication Master’s Thesis Author Viviane Soldenhoff 12014540 Supervisor Dr. Anne Kroon 28th January 2020
Abstract
The implementation of new technology in the workplace poses an enormous challenge. The media play a crucial role in shaping positive or negative perceptions of this change. This study examined, by conducting a quantitative content analysis, how Artificial Intelligence (AI) in the context of work is portrayed in opinion-leading newspapers in Germany and Switzerland from 2010-2019. First, the issue salience of AI in the reports was mapped. Second, it was explored which frames of AI existed during the years, how their popularity changed, and which frames accounted for portrayed chances and challenges. According to technological framing literature, when a new technology is emerging, journalists focus more on its chances as compared to challenges. Hence, third, the effect of time on chances and challenges was tested. The results showed that AI gained salience on the news agenda after 2015, its coverage soaring ever since. The content analysis revealed that macro-frames, focusing on organisations and society were consistently preferred over micro-frames focusing on consequences for the individual and that the reporting neglected consequences on social aspects of work. Chances and challenges were in balance over the research period, while chances were mostly linked to increased productivity, and challenges to negative societal consequences. It appeared that rather than time passed, the presence of specific frames influenced the valence of the coverage.
Keywords: issue salience, technological framing, issue framing, artificial intelligence,
Chance or Challenge for the Future of Work? Frames of Artificial Intelligence in the News
One of today’s most notable technological developments is the rise and rapid progress of Artificial Intelligence (AI), which includes the possibility to supplement, amplify, or completely substitute human brainpower for practically all mental tasks (Makridakis, 2017). Developments of AI are still in early stages, but researchers believe AI not only to support and complement, but “outperform humans in all tasks in 45 years and of automating all human jobs in 120 years”, even in creative ones such as writing a book (Grace, Salvatier, Dafoe, Zhang, & Evans, 2018, p. 729). Therefore, an incredible transformational impact on society and all spheres of life has been predicted (Makridakis, 2017; OECD, 2017).
Substantial investments in AI and technological transformations include extensive
changes to work as we know it today, such as redeployment of significant parts of the workforce, as well as the need for reskilling, upskilling, and the willingness to job transitions. This change raises information needs – and an individual’s willingness to adapt and navigate through them (Geiß, Jackob, & Quiring, 2013; World Economic Forum, 2018). To constructively facilitate change, individuals must take on a positive mindset towards progress in AI and recognize the continually changing environment, where one has to stay proactive and agile while continuously learning (World Economic Forum, 2018). This, however, might pose a challenge if individuals are confronted with conflicting or controversial information about AI. Accordingly, it is crucial to understand how its impact on work manifests on a personal and societal level and how this is evaluated.
Individuals are exposed to information through the mass media, where assessments of AI seem to present opposing value propositions and far-reaching consequences. Media coverage
also reflects how the public thinks about an issue (Bainbridge, 2002; Cobb & Macoubrie, 2004). Research has shown that opinions are shaped and generate support of new technologies through the framing of technology (Cobb, 2005; Nisbet & Mooney, 2007). Especially people with little contact with the technology to be implemented may rely stronger on second-hand information and can be influenced more through media framing (Rössler, 2001; Treem, Dailey, Pierce, & Leonardi, 2015). How the media present AI might influence people’s openness, future
willingness, and general trajectory on how to adapt to new technology and organisational change. Surprisingly, even though recent official governmental statements and initiatives (Bundesamt für Kommunikation BAKOM, 2019; Bundesministerium Verkehr Innovation Technologie, 2019; Menzel & Winkler, 2018; OECD, 2017) stress AI’s societal transformational power, until today, it is unclear how the news portray information about AI. Though scholars often focus on the potential of information technology in organisations, the question of what it means for employees and how this has changed over time is hardly ever raised (Treem et al., 2015).
This study asks the question of which frames exist of AI in the news, and how their popularity changed. It focuses on the information the media supply about the implementation of the technology in the workplace, and while relying on framing theory, breaks open the concept of “work” in different subframes. By conducting a content analysis of German (D) and Swiss (CH) newspapers to track the issue framing of AI from 2010-2019, an attempt is made to map how the portrayal of AI in the context of work has changed since the surfacing and intensification of the issue over the past decade.
Knowing the content of media portrayal might help understand how individuals and organisations understand AI and help identify stepping stones for the future of work. As AI developments are just gaining in speed, norms and values about AI and its implementation in the
workplace can still be adjusted before they are cemented, if it is known which and how its challenges and benefits are portrayed, and how individuals are told to deal with technological change (Harmon & Mazmanian, 2013).
As for theoretical contributions, this study presents a new approach of arguing based on technological framing literature in combination with issue framing literature, providing a different angle to researching the framing of new technology. Once the framing of AI is known, future research on its effects can be based on it.
Artificial Intelligence as a Concept
Researching emergent technological change, such as developments in AI, is fairly complex. Manifestations of this are the countless efforts to draw up a definition of intelligence, and still, there is no standard established (Legg & Hutter, 2007). An attempt to define AI entails a machine’s “ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience” (Gottfredson, 1997, p. 13). Particular emphasis is laid on adaption to and problem-solving in complex, uncertain environments and to meet goals with limited resources in the most successful way (Legg & Hutter, 2007). As the technology is still in early stages, newspaper articles are expected to mirror this vagueness. Therefore, AI in this paper is understood in the broadest sense of algorithm-based technology, possibly facilitating or
complicating daily work.
In recent years, many Western European states have founded initiatives to anticipate changes in the digital environment and steer their impact on society (Bundesamt für
Kommunikation BAKOM, 2019; Bundesministerium Verkehr Innovation Technologie, 2019; Menzel & Winkler, 2018; OECD, 2017), with most of them sharing the common shortcoming of not splitting the concept of work and job transformation into different subcategories. The little
research on AI in the public eye remains Anglo- and American-centric – the media as a mirror and monitor of public opinion are neglected.
Media Framing Explained
The mass media play a crucial role in shaping our knowledge of the world we live in (Luhmann, 1996). One way of investigating how the media cover issues, such as the portrayal of new technology, is through framing theory. Robert Entman (1993, p. 52) defined framing as “to select some aspects of a perceived reality and make them more salient in a communication text”, including a problem definition, its causal interpretation, moral evaluation, and suggestion for solutions and prediction for possible effects. According to framing theory, journalists are in the role of interpreting actors and the resulting news a product of continuous social-constructivist efforts. By consuming media content, we actively search for information but also passively learn about issues, sparking processes of how we make sense of them (Weick, 1995). Though media framing research has experienced a boom in the past decade (Vliegenthart & van Zoonen, 2011), the literature is vast and differs significantly in terms of operationalisation, methodology and conceptual clarity (de Vreese, 2005).
Nonetheless, one primary differentiation of frames is between generic and issue-specific frames (de Vreese, 2005). Generic frames transcend thematic boundaries and can be applied to a range of different topics. Issue frames are tied to a specific topic, therefore more clearly defined, and add to the relevance of the attributed details in the research process (de Vreese, 2005). This project applies issue specific frames, guided by the logic that they “highlight qualitatively different, yet potentially relevant considerations” (Cornelissen & Werner, 2014, p. 195).
Attention on AI in the Media Discourse
The media are known to guide public attention and if an issue is to be understood of bringing positive or negative consequences (McCombs & Shaw, 1972). One of the key aspects of framing is “sizing” (Entman, 1991, p. 9), which relates to the attention that is placed on an issue to make it appear more salient. Sizing includes, among the length and placement of a news text, how frequent an issue is picked up by journalists. As the media catalyse and mirror pressing public issues, by assessing media attention on AI, salience can be an indicator of controversy and the level of attention the public and journalists place on AI (Bauer, Kohring, Allansdottir, & Gutteling, 2000). Considering that the quantity of coverage of controversial emerging technology is associated with negative public bias and opposition (Mazur, 1981, 1990), it is essential to know how much emphasis the media discourse places on specific issues, and if it has increased.
Fast and Horvitz (2016) have already ascertained an increase of attention on AI from 2009 until 2016 in the New York Times. Hence, it is expected that this translates onto the
German-speaking media landscape. Attention to AI is expected to increase even further. Drawing from news values theory, characteristics such as the positive or negative valence of a topic, especially controversy and relevance to significant parts of society, are crucial in news story selection by journalists (Schulz, 1982), influencing how frequent an issue is picked up.Those characteristics apply to AI, due to the technologies’ increased impact on our lives and
multifaceted discussion in the news. Thus, the first hypothesis is:
H1: Media Attention on Artificial Intelligence (AI) in the context of work has increased over time (2010-2019)
How the Media Shape the Way We See Technology
In the past, framing theory has often been applied to assess news report’s portrayal of issues in society, also to emerging technologies (Anderson, Allan, Petersen, & Wilkinson, 2005; Marks, Kalaitzandonakes, Wilkins, & Zakharova, 2007; Nisbet, Brossard, & Kroepsch, 2003; Nisbet & Lewenstein, 2002). Technological developments are influenced through the
interpretative and sense-making processes of members in an organisation, which in turn can influence public acceptance and uptake for new technology, and the politics around it (Cave & Dihal, 2019; Fast & Horvitz, 2016). The concept, which attributes communicating actors also an active role by pointing out that a technology is both shaped and reflected by the narratives around it (Cave & Dihal, 2019; Garvey & Maskal, 2019; Harmon & Mazmanian, 2013; Kaplan & Tripsas, 2008), is called technological framing (Cornelissen & Werner, 2014; Orlikowski & Gash, 1994). In the case of framing technology, issue frames are “a collectively constructed set of assumptions, knowledge and expectations regarding a technology and its uses and applications in organisations” (Cornelissen & Werner, 2014, p. 185). As Harmon and Mazmanian (2013) have found, stories about technology are often diametrically opposite, stating differing agency and calling for contradicting action, precipitating the individual into a state of inner conflict between expectations of public discourse and private feasibility of dealing with those expectations. Taking on this perspective helps understand technology as an “object that is multiply constituted, and multiply experienced” (Harmon & Mazmanian, 2013, pp. 1051–1052). Orlikowski and Gash (1994) argue that “where incongruent technological frames exist,
organisations are likely to experience difficulties and conflicts around developing, implementing and using technologies” (Orlikowski & Gash, 1994). However, knowing about and being aware
of paradoxes may help organisational decision-makers work through them (Lüscher & Lewis, 2008), and in turn, influence further development and adaption of the technological change.
Technological frames are often constructed in the workplace, especially if individuals only get in contact with the technology in a work setting, for instance, when co-workers discuss its use. In the past, technological frames were mostly researched in this context. Treem et al. (2015) raised the valid concern that there is a need to analyse different sources where frames could stem from, as the first contact with a technology – or tales about it – is quite possibly outside the organisational black box. Hence, this indicates a reciprocal process, where micro frames of dealing with technology and how people work with it have an impact on the
organizational meso- and societal macro-level (Harmon & Mazmanian, 2013), and vice versa. This logic will be considered when the frames are outlined in the following chapter.
Knowing more about the portrayal of a new technology in early and later stages of its implementation into different work contexts is essential, because even if the level of awareness and information literacy due to lack of experience with the technology is still quite low, people form opinions about the technology, by drawing their information from other sources, like the media, and the key heuristics they provide by the way they frame an issue (Nisbet & Lewenstein, 2002; Scheufele, 2000). Thus, as the media mirror public discourse, identifying risks and
benefits in the AI discourse may help uncover normative value positions which guide, through portrayed frames, how to deal with technology (Harmon & Mazmanian, 2013). This is why in this study, two competing perspectives on issue-specific frames are taken on: a chance and challenge perspective. Before discussing this overall perspective, issue frames of AI in the context of work are introduced, which may highlight AI in the context of work from a chance perspective, challenge perspective, or both.
Frames of AI in the Context of Work
The focus of this study lies on five main frames, which are expected to appear in news reports. To account for the possibilities AI in the context of work can be covered, frames on different levels will be considered: the individual micro-level versus the organisational meso- and societal macro-level. To start with possible appearing micro level frames, Harmon &
Mazmanian (2013) have studied the cultural discourse about smartphones in the work context in American newspapers and highlighted “three areas of conflict, tension and instability relevant to the relationships among values, technologies and everyday life” (2013, p.1052). According to the authors, different frame dualities are tied to the use technology brings to the workplace, and how an individual might experience the communication technology on a personal level. Hence, AI might affect an individual’s autonomy and control over different aspects of work-life and affect the social ties of co-workers and individuals outside an organisation. Also, individuals could improve their productivity and work more efficiently, which consequentially enhances the productivity of the business (meso-/macro-level). The presence of micro frames will shed light on what picture the media paint about which types of behaviour AI might afford the individual employee or the impact it could have on working teams. This knowledge is valuable. Authors have previously analysed new technologies in organisations by what new behaviours they afford, instead of looking at their specific qualities (Treem & Leonardi, 2012). They found that people will change their work routine wilfully if they view AI as being useful and beneficial instead of hindering conducting their tasks (Leonardi, 2011; Schönberger, 1998).
The public scrutiny on AI in a work context often results in diametrically differing commentaries by scholars, corporate and governmental actors, and journalists. Reading those evaluations, micro-level issues and affordances of AI take a back seat. On one side of the scale,
some researchers point out a normal, potentially positive restructuring of society and economic benefits to all (Brynjolffson & McAfee, 2011, Brynjolffson & McAfee, 2014, Vermeulen, Kesselhut, Pyka, & Saviotti, 2018, Smith & Anderson, 2014), possibly facilitating and complementing the work of the individual (McKinsey, 2017). On the other side, there are
doomsday scenarios, with researchers complementing the research agenda with further emphasis on possible threats and challenges to humanity, not only of outperformance (Grace et al., 2018), but of mass redundancies, “intelligence explosion” and loss of control to superintelligent
machines (Horwitz, 2014).
The omnipresence of ethical concerns, job loss scenarios, and pressing career matters highlight the importance of macro level frames. Drawing from technological framing theory as previously outlined, frames of micro-level implementation of technology can reciprocally influence macro-level frames. Hence, it is important to know which frames are present in the media discourse, and how they account for positive or negative evaluations of AI, guiding an explorative research question:
RQ: Which frames of AI appear in the context of work, and how do they account for the chances and challenges present in news reports?
AI in the Workplace – A Chance or Challenge? The Dependent Variable
In previous research, it was hypothesised that public discourse mirrors public opinion. Emphasis on chances might reflect a positive view of AI, whereas challenges might be a sign of concern (Bauer, Kohring, Allansdottir, & Gutteling, 2000). Journalists, especially those working for opinion leading newspapers, usually include both positive and negative aspects of AI in order to meet journalistic objectivity standards. The weighing of costs and risks takes on a central role in researching the portrayal of new technology, as shared risk and benefit perceptions play a
crucial role in successful technology implementation, especially if a technology is controversial (Edmondson, 2003). Therefore, likely trends of the appearance of chances and challenges will be derived by consulting past research on sentiment about AI and related technology.
Zubiaga, Procter and Maple (2018) conducted mediated sentiment analysis of Twitter Data from 2009-2016. It showed that topics such as analytics, machine learning and Big Data & tech were generally viewed positive over the research period. In the German speaking part of Europe, Tweets about big data & tech were among the most positive world-wide, and security concerns did not take on such an important stand as in the United States. Hence, it is possible that this may translate onto news media, as Twitter in German-speaking Europe is considered to be used to a significant part by opinion leaders, and the news might adapt their newsworthy utterances.
Others have claimed that public perception of AI is generally negative, following the representation in popular Science Fiction scenarios, and leading to negative public perception of AI in the United States (Garvey & Maskal, 2019). These hypotheses, however, could not be supported. Shoham et al. (2018) conducted sentiment analysis on all English news articles from 2013-2018 and found that AI articles became less neutral and more positive since 2016. Fast and Horvitz (2016) found in their large-scale crowd-sourcing analysis of over 3 Million articles of the New York Times that since 2009, the discussion of AI has been consistently more optimistic than pessimistic. However, they also found that articles on AI and work have become less positive and more negative until 2015.
Translating these findings to a chance and challenge perspective, it is possible that after the expected intensification of the AI debate after 2016, articles focused more on challenges as compared to chances than before 2016. Research on nanotechnology has shown that at an early
stage of an issue cycle, benefits outweigh risks and that opinions towards new technology are very positive (Bainbridge, 2002; Cobb & Macoubrie, 2004; Ho, Scheufele, & Corley, 2013). The second hypothesis is:
H2: There will be a significant prediction of time on the score of Chances (vs. Challenges) of the articles.
Method
To systematically analyse the framing of AI in the context of work, a quantitative content analysis was conducted. The method’s advantage is the accessibility of records of frames past the moment of their construction in a communicating context, and the possibility to quantify a large data set, as preliminary explorations showed that very little articles were published before 2016 in either medium. Two quality newspapers of record were analysed: The German Süddeutsche Zeitung (SZ), which takes on a centrist-left liberal view and to add a broader geopolitical scope, the right-liberal Neue Zürcher Zeitung (NZZ) from Switzerland (Institut für Medien- und Kommunikationspolitik, 2012b, 2012a). They were chosen because of their accessibility in LexisNexis and their opinion-leading status, which is why they experience great recognition by journalists (Jandura & Brosius, 2011). Therefore, the frames presented in those papers are likely to penetrate big parts of the German-language media landscape. What is more, the NZZ and SZ are read by many organisational decision-makers such as managers (Neue Zürcher Zeitung, 2019; Süddeutsche Zeitung, 2018) who have been described as key actors in
frame-determination within organisations who tend to share and form their group’s technological frame (Orlikowski & Gash, 1994).
The selected newspapers are published daily and are thus deemed suitable to track the developments in the framing of AI. Considering the continuous transformation of the topic and
the intensification of the AI-debate over the last few years, the researched period starts dating back ten years from the beginning of the field phase, specifically from 1.1.2010 until 28.11.2019. Hence, this study applies a longitudinal design, which enables the exploration of trends over time.
Data and Analysis of Media Attention on AI
To gain an overview over the development of the media attention placed on AI over the years, a census of all articles related to AI and work between 2010 and 2019 was gathered via the LexisUni database applying a keyword search string in German containing the key term
“artificial intelligence” in combination with a set of words related to work, including their masculine/feminine as well as singular/plural version (Appendix A). This resulted in N = 2360 (SZ = 1546, NZZ = 814) newspaper articles. After manually deleting all duplicates, N = 1610 (SZ = 1085; NZZ = 525) remained. Table 1 shows the resulting census sample as well as an overview of the articles that were coded. Most of the articles were published in the business section or sections dealing with science and research.
The assessment of articles gathered, and differing keyword combinations applied showed that LexisNexis added many articles into the framing analysis that did not meet the inclusion criteria, which was determined by the relevance of an article to the topic and was measured with three different items (Appendix G) To ascertain if an article could be coded according to the developed codebook and therefore covering AI in combination with work as defined, it had to be entirely and thoroughly read. For instance, sometimes, AI was mentioned without further
attributes to work as defined, AI was mentioned in a research context, or a customer-centric focus is taken on. Therefore, as due to time constraints and the scope of this paper not the complete census could be read, for the media attention hypothesis (H1), attention on AI was
evaluated solely based on the keyword string, potentially including some articles not meeting the inclusion criteria. On the other hand, for the framing analysis, only articles meeting the inclusion criteria were coded.
Table 1.
Article count of the census and content analysis sample over time
Newspaper 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Total NZZ Attention Hypothesis Framing Analysis 1 0 7 1 3 1 8 1 17 6 20 6 62 15 118 16 136 14 153 16 525 76 SZ Attention Hypothesis Framing Analysis 10 1 7 2 14 4 12 6 23 12 35 17 138 15 209 15 315 15 322 15 1085 102 Total Attention Hypothesis Framing Analysis 11 1 14 3 17 5 20 7 40 18 55 23 200 30 327 31 451 29 475 31 1610 178
Data for the Explorative and Explanatory Framing Analysis
Out of the census gathered to answer the attention hypothesis H1, a representative
subsample of articles was selected and coded. A stratified random sampling method was applied. The strata consist of each year between 2010 and 2019, resulting in 10 strata. A within-strata simple random sample was drawn, aiming at approximately 15 articles per newspaper per year. If an article did not meet the inclusion criteria, another random article was chosen to code, until the target amount was reached. The years 2010-2015 did not meet this target, even though the census of all published articles of those years yielded by keyword search was read.
Coding Instrument
A measuring tool developed by an ongoing research project at the University of Amsterdam about AI in the media was adopted and qualitatively assessed in a pilot study. By checking randomly chosen articles containing the issue throughout different years, the codebook was adjusted and complemented. The codebook aimed to be inclusive and cover as many frames discussed in the context of AI and work as possible, while still maintaining the five issue-specific
frames the ongoing research project had already identified. The final codebook is in Appendix G.
Table 2 gives an overview of the main frames. Operationalisation
To answer the explorative research questions, the data collected by applying the codebook was analysed. All items are categorical and mutually exclusive. The articles were coded according to five deductively adapted, mutually exclusive frame categories. After checking if the main frame was present (i.e., Is the Autonomy & Control theme present in this
article? 0 = no; 1 = yes; 2 = Unclear), the presence of several chance, challenge, and some rare
neutral issue sub-frames was coded. The categories were developed by researchers working on a content analysis project at the University of Amsterdam. They adapted the micro-frame
categories from Harmon and Mazmanian (2013), and the newly developed meso-/macro-categories find support in related literature.
Micro-Frames and Items
Frame 1: Autonomy & Control relates to possible freedom from tasks, such as boring,
monotonous work, or the possibility to focus on more creative aspects of work. Potential challenging aspects could include important decisions being taken over by a machine or less flexible workdays (Harmon & Mazmanian, 2013; Tussyadiah & Miller, 2019) (i.e., Is this article
about enhanced control over the workday (i.e. schedule flexibility) as a result of adoption of AI in the work place? 0 = no; 1 = yes; 2 = Unclear).
Frame 2: Togetherness & Community relates to the social and personal effects of AI in
the workplace. AI could either enhance social interaction through accessibility of one another or friends and family via new innovative tools or reduce personal contact precisely because of these tools, which could impair the development of social and soft skills at the workplace (Harmon &
Mazmanian, 2013) (i.e., Is the article about a reduced sense of community between employees as
a result of adoption of AI in the workplace? 0 = no; 1 = yes; 2 = Unclear).
Frame 3: Productivity & Effectiveness focuses on affordances due to new software or
devices, and the change of the way of work. In a supporting role, AI can help employees and teams perform better or more (Harmon & Mazmanian, 2013; Makridakis, 2017; Syam & Sharma, 2018; Tussyadiah & Miller, 2019) (i.e., Is the article about the influence of AI on how
employees conduct their work (i.e., their work process)? 0 = no; 1 = yes; 2 = Unclear).
Adding to the three micro frames adapted from Harmon and Mazmanian (2013) by researchers of the University of Amsterdam, they have thus outlined two additional frames: Organisation & Society and Career & Development.
Frame 4: Career & Development focuses on human resources related issues that could
spur from the implementation of technology, such as the need for reskilling or upskilling, increased or reduced salaries due to AI implementation (Makridakis, 2017) (i.e., Is this article
about the shift of current jobs to creative ones due to the adoption of AI in the workplace? 0 = no; 1 = yes; 2 = Unclear).
Meso-/Macro-Frame and Items
Frame 5: Organisation & Society focuses on all the implications AI implementation in
the workplace has on the organisational and societal level. This can range from practical gains such as safer workplaces due to surveillance, increased productivity of the business, mass redundancies or psychological implications for big groups of employees (Tussyadiah & Miller, 2019) (i.e., Is the article about the relationship between AI technology and a decline in equal
Pre-Test and Adjustments
Additional items were added if they were missing to code relevant information, and it appeared that some few items needed clarification and some opposites of the same item needed to be included (i.e., Is this article about the increase in salaries for employees thanks to the
adoption of AI in the workplace? / addition: Is it about the decrease of salaries?). Items not
coded in the pre-test were kept in the codebook, as the possibility remained that they could still appear in other articles because some frames were expected to appear little, while their absence indicates what information individuals could miss about AI.
Table 2.
Overview and operationalization examples of frames
Examples
Duality Frame/Factor Chance Challenge Scale
Mic
ro
Autonomy & Control*
Freedom from tedious / monotonous work / schedule flexibility monotony / AI dictating tasks 0 = not present 1 = present 2 = unclear Togetherness & Community* Accessibility of co-workers / connectivity reduced face-to-face contact / superficial relationships / loss of soft skills
Productivity & Effectiveness*
Increased efficiency thanks to new devices and platforms (quality or quantity)
Inefficiency through reshaping / shifting ways of work / increased work load through new tools
Career & Development** Increased salaries / development opportunities
Need for training or skills (negatively framed) / shift of jobs Me so /Ma cr o Organisation & Society**
Fairer recruitment / safer work spaces
Loss of jobs / psychological effects within society
Note: *based on Harmon and Mazmanian (2013), adapted by UvA research project (2019), ** UvA
The complete coding was conducted solely by the author, a native German speaker. Because of the scope of this project and the lengthiness of the coding procedure, intercoder reliability was not assessed, and only intracoder reliability was calculated. A pre-test with 27 articles (15% of the total sample N = 178) yielded either excellent intracoder reliability results of
Krippendorff’s α 0.8 to 1 or were a constant value. Therefore, there was no variation (see Appendix B). Changing the items yielding less than Krippendorff’s α 0.9 was considered and
again overruled, as the change in coding was assigned to chance and the complexity and length of the codebook, because the item description appeared clear to the coder.
Explanatory Analysis Dependent variable
The key dependent variable Score Chances (vs. Challenges) variable (0 = no Chances, 100 =
100% Chances) is constructed by weighing the chances relative to the challenges presented
within each media text. It was constructed by applying the following formula:
Score Chances (vs. Challenges) (%) =
∑ ( 𝐶ℎ𝑎𝑛𝑐𝑒𝑠
𝐶ℎ𝑎𝑛𝑐𝑒𝑠 + 𝐶ℎ𝑎𝑙𝑙𝑒𝑛𝑔𝑒𝑠)
𝑎∈𝑎𝑟𝑡𝑖𝑐𝑙𝑒𝑠
𝑥 100
It was applied to each item in the codebook that could clearly be identified as covering only a chance or a challenge aspect (i.e., Is the article about enhanced independence of
employees from the office (i.e., location, coworkers, supervisors, etc.) as a result of adoption of AI in the workplace? 0 = No; 1 = Yes; 2 = Unclear / Is the article about reduced independence of employees from the office (i.e., location, coworkers, supervisors, etc.) as a result of adoption of AI in the workplace? 0 = No; 1 = Yes; 2 = Unclear; Appendix C). In total, 20 items were assigned
out of the equation. By calculating this relative score, dummy coded items could be transformed into a scale variable, and the multiplicity of the framing of AI in the work context could be incorporated into the analysis.
Independent variable
This study’s main predictor variable is the temporal variable. It is constructed to track a possible change in frames in the phase after the expected increase of issue coverage. A dummy variable was created distinguishing between the period before (0) and after (1) the increase of coverage. To create the dummy, the years 2010-2015 were assigned to 0 and 2016-2019 to 1.
Analysis
To analyse the impact of time on the score of chances (vs. challenges), a multiple linear regression was conducted. Several control variables were included in the model.
Covariates. First, it was controlled for the two newspapers (0 = Süddeutsche Zeitung, 1 =
Neue Zürcher Zeitung) to account for variation of the medium’s political orientation. Also, all five frames (0 = not present, 1 = present) were included as controls: Autonomy & Control, Togetherness & Community, Productivity & Effectiveness, Career & Development and Organisation & Society.
Robustness Check. To account for possible autocorrelation over time, a robustness check
was conducted by evaluating a second model with included lagged values of the dependent variable as explanatory values (Wilkins, 2018).
Results Dynamics of Media Attention on AI
The first Hypothesis H1expected media attention on AI in the context of work to have increased over time. This could indeed be supported by examining the absolute frequencies.
Figure 1 and the numbers from Table 1 indicate that AI in the context of work has only recently
gained salience on the public agenda, and though slightly increasing, barely receiving
recognition before 2015. The coverage of AI increased steeply after 2015. Though the NZZ (M = 52.5, SD = 60.48) published about half the number of articles compared to the SZ (M = 108.6,
SD = 129.09), the media coverage generally proceeded fairly parallel to each other.
Figure 1. Dynamics of attention on AI in the context of work by newspaper between 2010 and 2019.
As part of the explorative analysis of the frames and to test whether the emphasis on frames of the two newspapers do not differ and could be considered as a unity, five chi-square difference tests were conducted in order to determine if there was a significant correlation between the newspapers and the coverage of all five frames.The difference tests demonstrated that the effect of the newspaper on all five frames was not statistically significant.1
Having determined an increase in coverage of AI over the past ten years, I now come to discuss the issue framing of AI over time.
The Framing of AI: Descriptive Findings
This study asked the question of which frames about AI in the context of work exist in German-language newspapers over time. The table in Appendix D gives an overview of the five
0 50 100 150 200 250 300 350 400 450 500 NU M BER OF A RT ICL ES YEAR
main frames, their frequency, absolute and relative share. The complete table including all items can be found in Appendix E. The results must be reviewed critically, as from 2010-2014 only few articles per year could be analysed. From 2010-2019, though showing great differences between all five main frames, they were present multiple times, and over the years, each of the subframes was present at least once, with one exception: AI was never expected to reduce connectedness with family and friends.
Figure 2. Frequencies of the main five frames over the years present per article. Note: i.e., in 2015, the Autonomy &
Control frame was present in 12 articles.
Another aim of this research was to evaluate how the popularity of frames changed over time. Figure 3 shows the absolute presence of frames within each article in each year. As the graphs show, in earlier years with little news coverage, the frames do not show a clear trend, similarly to chances and challenges in Figure 4. In 2014, a pattern started to develop, leading to a clear favouring of the macro-frame Organisation & Society, followed by the two steadily
increasing micro-frames, Autonomy & Control and Productivity & Effectiveness. The Career & Development frame rose steeply and peaked in 2016 but dropped again. Though the micro-frame Togetherness & Community was present since 2013, it did not rise to the popularity of the other
1 3 5 7 18 23 30 31 29 31 1 2 3 6 11 12 15 18 17 15 0 0 0 2 2 2 4 0 3 2 0 3 2 6 8 9 13 13 16 16 0 0 0 1 6 7 15 9 10 9 1 0 3 1 11 17 19 22 21 21 0 5 10 15 20 25 30 35 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 NUM B E R OF A R T IC L E S
n articles per year Autonomy & Control Togetherness & Community Productivity & Effectiveness
Career & Development Organization & Society
frames and remained at a very low level. Overall, these findings indicate that more emphasis is placed on the macro level as well as the impact of AI on how work is conducted, how this increases productivity, and that social aspects and career matters do not receive comparable attention.
Figure 3. The absolute presence of each frame in each article per year. Note: i.e., in 2011, the Productivity &
Effectiveness Frame was present in 100% of the analysed articles.
Another explorative aim was to discover if AI is covered more from a chance or challenge perspective, and which frames account for it. The cumulative relative presence of chance, challenge and neutral frames shows that chance (2014-2019: M = 47.33, SD = 17.83) and challenge (2014-2019: M = 44.31, SD = 21.99) frames are almost present equally, with challenge frames showing greater variance (Figure 4). Though there is no clear trend, chance frames were slightly increasing in recent years, whereas challenge frames were decreasing. Challenge items were most present in 2015 and 2017, tilting in 2016 and 2018. The great variation from 2010
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% YEAR
Autonomy & Control Productivity & Effectiveness Togetherness & Community Career & Development Organization & Society
until 2013 is most probably due to the little article count in those years, which is why they are not considered here.
Figure 4. Cumulative relative percentage of chance items (n = 20) versus challenge items (n = 23) (DV), and rest (n
= 3) per year.
Consulting the numbers of the absolute presence of the frames in the number of articles per year, opportunities, advantages and challenges were most popular among the Productivity & Effectiveness, Autonomy & Control and Organisation & Society frames. Frames most discussed over the years included enhanced freedom, autonomy and control through AI in general (present in 32% of all articles), enhanced control over knowledge (34.4%) and focus on tasks (17.4%), enhanced efficiency and effectiveness to perform tasks (40.4%), enhanced productivity of the business (25.3%) and employees (25.3%).
Risks, disadvantages and challenges were most prominent among the Autonomy & Control, Career & Development and Organisation & Society frames. This included reduced
0 60 42.7 66.7 45.3 41.4 47.5 44.6 53.5 51.7 100 20.1 42.8 21.6 45.4 52.2 43.7 47.8 36.6 40.2 0 19.9 14.5 11.7 9.3 6.4 8.8 7.6 9.9 8.1 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Chances Challenges Rest
freedom, autonomy and control (28.1%), the shift of jobs (14.6%) and the consequential
reskilling needs (20.2%), ethical concerns (29.8%), predicted job loss (28.1%), reduced control over work tasks (20.2%) and reduced control over knowledge (18.4%).
A Change of Chances? Explanatory Analysis
The second hypothesis H2 suggested that after 2016, articles focused more on challenges as compared to chances than before 2016. The regression model with the score of total chances versus challenges per article as the dependent variable and time as the predictor, while
controlling for type of newspaper as well as the five frames Autonomy & Control, Togetherness & Community, Productivity & Effectiveness, Career & Development and Organisation &
Society was significant (F(7,170)=7.744, p <.000). The regression model can therefore be used to predict the presence of the discussion of chances and challenges of AI in German-language newspapers, though the strength of the prediction is moderate. The explanatory variables predict only 24.2 percent of the score of chances (vs. challenges) present in an article about AI in the context of work (R2 =.242). Table 3 summarizes the results. The main predictor of interest, time,
does not have a significant impact on the score of chances (vs. challenges) (B= .443, SE= 1.26,
p= 73). Moving to the control variables, as the model shows, the presence of the Productivity &
Effectiveness frame (B= 30.28, SE= 5.32, p< 00) will increase the chances present per article by 30.2%. The presence of the Organisation & Society frame (B= -10.77, SE= 5.59, p= .06) did only show a marginally significant effect, and according to the model, its presence in an article would decrease the score of chances (vs. challenges) per article by 10.8%. In sum, these findings indicate that time does not predict the score of chances (vs. challenges), but that specific frames do. H2 must be rejected.
Table 3.
Regression model to predict the score of chances (vs. challenges) before and after 2016 (N=178)
Score Chances/Challenges
B SE
Constant 41.862*** 10.377
Time .357 1.254
Newspaper type (NZZ vs. SZ) 2.237 5.249
Autonomy & Control -.780 5.301
Togetherness & Community 13.543 9.195
Productivity & Effectiveness 30.279*** 5.465
Career & Development -7.272 5.465
Organisation & Society -10.773† 5.593
Note. The data predict changes in Chances present (%). Cell entries are unstandardized coefficients and standard errors.
†p < .10. *p < .05. **p < .01. ***p < .001.
The robustness check with lag values of the score of chances (vs. challenges) showed that autocorrelation over time could be neglected. The regression yielded almost identical results to the reported regression model, therefore the presence of chances in articles was also not
explained by the presence of chances in earlier published articles.
Discussion
The aim of this study was threefold, and this discussion will be structured accordingly. First, the results will be recapitulated. Then, the development of media attention on AI in the context of work will be discussed, followed by a descriptive part covering how the media debate took shape in German-language high-quality newspapers over the past decade. Lastly, the
variation of framing from a chance and challenge perspective will be accounted for by discussing the explanatory analysis, followed by the study’s limitations and suggestions for further research.
Having appeared little before, AI emerged on the news agenda in 2015 and has since received rising attention every year. Though the steep increase of coverage of AI happened after
2015, it was already increasing since 2009, supporting previous findings (Fast & Horvitz, 2016), and indicates that information needs are growing. That the issue steadily gained salience suggests that the debate is still nascent and will receive even more attention in the future.
The framing analysis showed that the societal frame was consistently favoured by journalists. The most controversially discussed frame was Autonomy & Control, with chances and challenges outweighing each other. The Productivity & Effectiveness frame appeared almost only in a positive context, though the texts portrayed much uncertainty about how AI influences the way we work. The Organisation & Society frame appeared most in combination with
challenges. The Togetherness & Community frame was barely present over the years. How AI implementation in the workplace might affect our social interaction is, therefore, a question journalists do not answer in their news reports. Similar to what other authors found (Weaver, Lively, & Bimber, 2009), the increase in coverage of AI over the years resulted in frame diversification. Generally, journalists presented challenges and chances to equal parts in their reports.
The results of the explanatory analysis revealed that the coverage’s content of chances and challenges was not significantly influenced by time passed during which AI gained salience. However, the valence of the coverage was related to discussed frames in the news text –
specifically to the Productivity & Effectiveness frame, and the Organization & Society frame.
Discourse Developments
The concept of AI as defined was tightly knit to the notions of robots, robotization, and digitalization, and those terms were often mentioned synonymously. However, this should not be ascribed to journalists being negligent of the differences, but low awareness and knowledge of the public. Complexity reduction might serve as a tool to facilitate information on a
difficult-to-understand technology and its effects. The critique that AI is often dropped as a buzzword without a proper explanation of what it is, and which effects it could have corresponds to this statement (Jung, 2017). Though thoughts and visions of AI date back to the 1950s (Buchanan, 2005), it is not until after 2015 that journalists properly recognize its disruptive potential, and first scenarios are outlined on the impact on work when the notion of deep learning catapulted the issue on the news agenda. Accordingly, in 2015 challenge frames increased compared to the year before and after. As a result, the question was raised what AI-implementation could mean for the economy, and reskilling needs were uttered.
Another important year standing out in the data was 2017. Not only did many articles mention a complete change of work how we know it today, but challenge frames increased in 2017, shortly before the World Economic Forum (WEF) published its Global Risk Perception Report (World Economic Forum, 2018), underlining the need for reskilling. However, the “shift of jobs” was not affected and decreased in later years, but came back in 2019. Though those challenges are acknowledged, AI implementation is perceived to lead also to “increased productivity of the business”, with this frame peaking in 2017.
This organisation-focus does not surprise. Adopting the new practice presents an opportunity rather than a threat because it allows an organisation to achieve a performance advantage relative to its competitors since few organisations have yet adopted the practice (Kennedy & Fiss, 2009). Switzerland and Germany tend towards high salience of economic benefits for technology (Bauer et al., 2000), and the NZZ and SZ as liberal newspapers are prone to this stance, also explaining the salient chance frames from increased Productivity &
and losses can be expected, because by then it will be clearer, how and where AI is introduced, and what this would mean for societal groups.
What is more, AI technology that was celebrated in earlier years, such as IBM Watson, which was used as a helping tool for patient diagnosis in hospitals, has been found to make mistakes in 2018. This could have led to the perception of “reduced efficiency” and “ethical concerns” specifically, and generally sparked suspiciousness towards the power of AI in recent years, even though the discussion recently made a turn to a more chance-oriented perspective.
Though AI can help employees make use of and work with data, it is also feared that control over work tasks will be lost, leading to less autonomy and freedom, such as being dictated by a machine when, what and how to work – and not knowing how this was decided, because of the perception of the technology as a black box. Whereas in the first two years of the field phase, AI was believed to lead to increased independence from office, this frame was hardly present anymore in recent years, further implying a certain hesitation towards the technology. Those frames could influence public acceptance and hinder uptake for new technology, if they lead to stress and feelings of helplessness if technology is framed as being an inevitable social process which in turn is discussed within organisations, and therefore participates in framing the technology (Harkness et al., 2005; Schönberger, 1998).
The study does not support previous findings of mainly positive media coverage or even hyping of emerging technologies (Bubela & Caulfield, 2004; Metag & Marcinkowski, 2014; Shoham et al., 2018). The stable presence of challenges to almost equal parts to challenges over the years highlights that though the economic promises of AI are alluring, journalists are still concerned about the consequences of its dissemination, and critically reflect it. Hence, they stick to their objectivity standards of always presenting two sides. However, those critical perspectives
on technology are tied to ambiguity and uncertainty (Metag & Marcinkowski, 2014) and focused mainly on the societal level, leaving aside the effects on social ties in the workplace. Considering that technological frames are constructed through interaction in the workplace, it is notable that from 2010-2014, there was no emphasis on the Togetherness & Community frame, and in the following years, no clear projections concerning that frame were made. Going back to
Orlikowski and Gash (1994), the problem that individuals could be hesitant towards AI
implementation in the workplace remains, as journalists fail to acknowledge this important part of the work-life.
Effect of Time
Lastly, coming to discuss the explanatory findings, the analysis revealed that time passed could not explain if AI was discussed more from a chance or challenge perspective. This is surprising, as in previous literature, many authors found differences in the valence of the
coverage of technology over time (Bainbridge, 2002; Cobb & Macoubrie, 2004; Ho et al., 2013). Consulting the descriptives, no clear pattern on chances and challenges could be derived in earlier years. As the debate about AI is still nascent, it is possible that the issue cycle has just begun, which might be a reason for the absence of significant effects of time. By continuing the monitoring, this could be verified. Drawing from the previously outlined descriptives, the cover of the Productivity & Effectiveness, as well as the Organization & Society frame, conversely to the other frames, showed a clear pattern of emphasis on chances and challenges over time, which can thus explain the presence of chances versus challenges.
Limitations and Future Research
It is debatable to what extent the findings can be generalised because issue specific framing analysis is generally hard to apply to other contexts (de Vreese, 2012). Though the
codebook was originally developed for the New York Times, the good fit for German-language opinion-leading newspapers suggests that this would translate into similar media in other
geopolitical regions. Albeit almost all possible ways AI in the context of work could be coded, a few items appeared to be missing because they did not appear in the texts in the pre-study. Future studies should also include the hopeful scenario belonging to the Organisation & Society Frame that “more jobs will be created” and “jobs only temporarily lost”, both frequently appearing after 2016, and the unfavourable scenario “substitution by machine”.
The results of this study are tied to opinion-leading national newspapers. Analysing other types of newspapers, such as tabloids, could have yielded different results. Tabloids, focusing on soft news, could be focusing more on personal and private matters, whereas quality newspapers stress more public and societal consequences. Also, tabloids tend to report more negative
consequences (Reinemann, Stanyer, Scherr, & Legnante, 2012). One of the most pressing issues about the implementation of AI in the workplace often addressed is the loss of jobs, especially by older and low skilled manual laborers, as well as ethical concerns. Proposed solutions are
upskilling as the need for high-skilled and creative personnel increases, as well as a basic income or policy regulations, but not what kind. It seems this considerable part of the workforce is not specifically addressed and their information needs not fulfilled. It is also the demographic of tabloid media, which also generally receives great recognition in Germany and Switzerland. Therefore, further analysis of articles about AI in the context of work published in tabloids and their effect on public perception should appear on future research agendas.
Additionally, numerous factors that influence technology coverage such as journalistic cultures, role conceptions (van Dalen, de Vreese, & Albæk, 2012), organisational and production structures and relation to sources (Hughes, Kitzinger, & Murdock, 2006), geography (Bauer et
al., 2000; Marks, Kalaitzandonakes, Wilkins, & Zakharova, 2007), or the notion of agenda-setting power to establish a prominent frame (Benford & Snow, 2000; Hänggli, 2011) could be included in further research, as frames are not limited to the vacuum of sole media content.
Conclusion
Today, still much uncertainty of affected demographics and economic sectors remains. This unpredictability of the effects of AI implementation in the workplace could lead to
employees feeling left behind (Betschon, 2015). Convincing them to change their work routines will be an enormous challenge, especially if social interactions could potentially be affected negatively.At this point, further unfolding of the issue has to be awaited. The salience of AI in the news is likely to increase further. Once the technology is further implemented in the workplace, it will be more clear what consequences it brings, also on a micro-level, possibly changing the established stable discourse.
Bibliography
Bainbridge, W. S. (2002). Public attitudes toward nanotechnology. Journal of Nanoparticle
Research. https://doi.org/10.1023/A:1022805516652
Bauer, M. W., Kohring, M., Allansdottir, A., & Gutteling, J. (2000). The dramatisation of biotechnology in elite mass media. Biotechnology, 35–52.
Benford, R. D., & Snow, D. A. (2000). Framing processes and social movements: An overview and assessment. Annual Review of Sociology, 26(1), 611–639.
https://doi.org/10.1146/annurev.soc.26.1.611
Betschon, S. (2015). Von Ameisen und Übermenschen. Retrieved January 10, 2020, from Neue Zürcher Zeitung website: https://advance.lexis.com/document?crid=9c60d872-566d-420a-
aa3c-b951fe48966f&pddocfullpath=%2Fshared%2Fdocument%2Fnews%2Furn%3AcontentItem
%3A5HC7-KGC1-JCN5-V3V1-00000-00&pdcontentcomponentid=12122&pdmfid=1516831&pdisurlapi=true
Bubela, T. M., & Caulfield, T. A. (2004, April 27). Do the print media “hype” genetic research? A comparison of newspaper stories and peer-reviewed research papers. CMAJ, Vol. 170, pp. 1399–1407. https://doi.org/10.1503/cmaj.1030762
Buchanan, B. G. (2005). A (very) brief history of Artificial Intelligence. AI Magazine, 26(4), 53– 53. https://doi.org/10.1609/AIMAG.V26I4.1848
Bundesamt für Kommunikation BAKOM. (2019). Digitale Schweiz. Retrieved November 13, 2019, from https://www.bakom.admin.ch/bakom/de/home/digital-und-internet/strategie-digitale-schweiz.html
und künstliche Intelligenz. Retrieved November 19, 2019, from
https://www.bmvit.gv.at/themen/innovation/forschungspolitik/robotikrat.html
Cave, S., & Dihal, K. (2019). Hopes and fears for intelligent machines in fiction and reality.
Nature Machine Intelligence, 1(2), 74–78. https://doi.org/10.1038/s42256-019-0020-9
Cobb, M. D. (2005). Framing effects on public opinion about nanotechnology. Science
Communication, 27(2), 221–239. https://doi.org/10.1177/1075547005281473
Cobb, M. D., & Macoubrie, J. (2004). Public perceptions about nanotechnology: Risks, benefits and trust. Journal of Nanoparticle Research, 6(4), 395–405. https://doi.org/10.1007/s11051-004-3394-4
Cornelissen, J. P., & Werner, M. D. (2014). Putting framing in perspective: A review of framing and frame analysis across the management and organizational literature. The Academy of
Management Annals, 8(1), 181–235. https://doi.org/10.1080/19416520.2014.875669
de Vreese, C. H. (2012). New avenues for framing research. American Behavioral Scientist,
56(3), 365–375. https://doi.org/10.1177/0002764211426331
Edmondson, A. C. (2003). Framing for learning: Lessons in successful technology implementation. California Management Review, Vol. 45, pp. 34–54.
https://doi.org/10.2307/41166164
Entman, R. M. (1991). Framing U.S. coverage of international news: Contrasts in narratives of the KAL and Iran Air incidents. Journal of Communication, 41(4), 6–27.
https://doi.org/10.1111/j.1460-2466.1991.tb02328.x
Fast, E., & Horvitz, E. (2016). Long-Term trends in the public perception of artificial
intelligence. 31st AAAI Conference on Artificial Intelligence, AAAI 2017, (January 1986), 963–969. Retrieved from http://arxiv.org/abs/1609.04904
Garvey, C., & Maskal, C. (2019). Sentiment analysis of the news media on artificial intelligence does not support claims of negative bias against artificial intelligence. OMICS: A Journal of
Integrative Biology, 7, omi.2019.0078. https://doi.org/10.1089/omi.2019.0078
Geiß, S., Jackob, N., & Quiring, O. (2013). The impact of communicating digital technologies: How information and communication technology journalists conceptualize their influence on the audience and the industry. New Media and Society, 15(7), 1058–1076.
https://doi.org/10.1177/1461444812465597
Gottfredson, L. S. (1997). Mainstream science on intelligence: An editorial with 52 signatories,
history, and bibliography. 13–23.
Grace, K., Salvatier, J., Dafoe, A., Zhang, B., & Evans, O. (2018). Viewpoint: When will ai exceed human performance? Evidence from ai experts. Journal of Artificial Intelligence
Research, Vol. 62, pp. 729–754. https://doi.org/10.1613/jair.1.11222
Hänggli, R. (2011). Key Factors in Frame Building. In Political Communication in Direct
Democratic Campaigns (pp. 125–142). https://doi.org/10.1057/9780230343214_9
Harkness, A. M. B., Long, B. C., Bermbach, N., Patterson, K., Jordan, S., & Kahn, H. (2005). Talking about work stress: Discourse analysis and implications for stress interventions.
Work and Stress, 19(2), 121–136. https://doi.org/10.1080/02678370500160068
Harmon, E., & Mazmanian, M. (2013). Stories of the smartphone in everyday discourse: Conflict, tension & instability. Conference on Human Factors in Computing Systems -
Proceedings, 1051–1060. https://doi.org/10.1145/2470654.2466134
Ho, S. S., Scheufele, D. A., & Corley, E. A. (2013). Factors influencing public risk-benefit considerations of nanotechnology: Assessing the effects of mass media, interpersonal communication, and elaborative processing. Public Understanding of Science, 22(5), 606–
623. https://doi.org/10.1177/0963662511417936
Hughes, E., Kitzinger, J., & Murdock, G. (2006). The media and risk. In Risk in social science (pp. 250–270).
Institut für Medien- und Kommunikationspolitik. (2012a). Neue Zürcher Zeitung. Retrieved December 17, 2019, from https://www.mediadb.eu/forum/zeitungsportraets/nzz.html Institut für Medien- und Kommunikationspolitik. (2012b). Süddeutsche Zeitung. Retrieved
December 9, 2019, from https://www.mediadb.eu/forum/zeitungsportraets/sueddeutsche-zeitung.html
Jandura, O., & Brosius, H.-B. (2011). Wer liest sie (noch)? Das Publikum der Qualitätszeitungen. In Krise der Leuchttürme öffentlicher Kommunikation (pp. 195–206).
https://doi.org/10.1007/978-3-531-93084-8_12
Jung, H.-M. (2017). Die Maschine und ich; Künstliche Intelligenz ist das Thema der Stunde auf Technik-Konferenzen. Aber was kann sie wirklich, wie sehr muss man sich vor ihr fürchten, und kann Europa bestehen gegen die Konzerne aus den USA und Asien? Retrieved January 10, 2020, from Süddeutsche Zeitung website:
https://advance.lexis.com/document?crid=e6126bd8-5d80-4025-826d-727f2fa4e4b3&pddocfullpath=%2Fshared%2Fdocument%2Fnews%2Furn%3AcontentItem
%3A5R06-M6S1-JDHD-W42Y-00000-00&pdcontentcomponentid=5829&pdmfid=1516831&pdisurlapi=true
Kaplan, S., & Tripsas, M. (2008). Thinking about technology: Applying a cognitive lens to technical change. Research Policy, 37(5), 790–805.
https://doi.org/10.1016/j.respol.2008.02.002
adoption and implementation decisions among U.S. hospitals. Academy of Management
Journal, 52(5), 897–918. https://doi.org/10.5465/AMJ.2009.44633062
Legg, S., & Hutter, M. (2007). Universal intelligence: A definition of machine intelligence.
Minds and Machines.
Leonardi, P. M. (2011). When flexible routines meet flexible technologies: Affordance, constraint, and the imbrication of human and material agencies. MIS Quarterly:
Management Information Systems, 35(1), 147–167. https://doi.org/10.2307/23043493
Makridakis, S. (2017, June 1). The forthcoming Artificial Intelligence (AI) revolution: Its impact on society and firms. Futures, Vol. 90, pp. 46–60.
https://doi.org/10.1016/j.futures.2017.03.006
Marks, L. A., Kalaitzandonakes, N., Wilkins, L., & Zakharova, L. (2007). Mass media framing of biotechnology news. Public Understanding of Science, 16(2), 183–203.
https://doi.org/10.1177/0963662506065054
Mazur, A. (1981). Media Coverage and public opinion on scientific controversies. Journal of
Communication, 31(2), 106–115. https://doi.org/10.1111/j.1460-2466.1981.tb01234.x
Mazur, A. (1990). Nuclear power, chemical hazards, and the quantity of reporting. Minerva,
28(3), 294–323. https://doi.org/10.1007/BF01096293
McCombs, M. E., & Shaw, D. L. (1972). The agenda-setting function of mass media. Public
Opinion Quarterly, 36(2), 176–187.
Menzel, C., & Winkler, C. (2018). Zur Diskussion der Effekte Künstlicher Intelligenz in der
wirtschaftswissen-schaftlichen Literatur Zur Diskussion der Effekte Künstlicher Intelligenz in der wirtschaftswissenschaftlichen Literatur.
comparative analysis of the media coverage of nanotechnology in Austria, Switzerland and Germany. Journalism: Theory, Practice & Criticism, 15(4), 463–481.
https://doi.org/10.1177/1464884913491045
Neue Zürcher Zeitung. (2019). Preise. NZZ Media Solutions. Retrieved December 17, 2019, from https://www.nzzmediasolutions.ch/zuerich/
Nisbet, M. C., & Mooney, C. (2007). Science and society: Framing science. Science, Vol. 316, p. 56. https://doi.org/10.1126/science.1142030
OECD. (2017). OECD Digital Economy Outlook 2017. In OECD Digital Economy Outlook
2017. https://doi.org/10.1787/9789264276284-en
Orlikowski, W. J., & Gash, D. C. (1994). Technological Frames: Making Sense of Information Technology in Organizations. ACM Transactions on Information Systems, 12(2), 174–207. https://doi.org/10.1145/196734.196745
Reinemann, C., Stanyer, J., Scherr, S., & Legnante, G. (2012, February 11). Hard and soft news: A review of concepts, operationalizations and key findings. Journalism, Vol. 13, pp. 221– 239. https://doi.org/10.1177/1464884911427803
Rössler, P. (2001). Between online heaven and cyberhell: The framing of “the internet” by traditional media coverage in Germany. New Media and Society, 3(1), 49–66.
https://doi.org/10.1177/14614440122225985
Schönberger, K. (1998). The Making of the Internet. In Online-Kommunikation (pp. 65–84). https://doi.org/10.1007/978-3-322-87318-7_4
Schulz, W. F. (1982). News Structure and People’s Awareness of Political Events. Gazette
(Leiden, Netherlands), 30(3), 139–153. https://doi.org/10.1177/001654928203000301
(2018). The AI index 2018 annual report. Retrieved from http://cdn.aiindex.org/2018/AI Index 2018 Annual Report.pdf
Süddeutsche Zeitung. (2018). Süddeutsche Zeitung: Die Qualitätszeitung Nr. 1. Retrieved December 17, 2019, from https://sz-media.sueddeutsche.de/de/service/downloads/preisliste-mediadaten-sz_de.pdf
Syam, N., & Sharma, A. (2018). Waiting for a sales renaissance in the fourth industrial revolution: Machine learning and artificial intelligence in sales research and practice.
Industrial Marketing Management, 69(January), 135–146.
https://doi.org/10.1016/j.indmarman.2017.12.019
Treem, J. W., Dailey, S. L., Pierce, C. S., & Leonardi, P. M. (2015). Bringing technological frames to work: How previous experience with social media shapes the technology’s meaning in an organization. Journal of Communication, 65(2), 396–422.
https://doi.org/10.1111/jcom.12149
Treem, J. W., & Leonardi, P. M. (2012). 7 social media use in organizations. 143–189.
Tussyadiah, I., & Miller, G. (2019). Perceived Impacts of artificial intelligence and responses to positive behaviour change intervention. In Information and Communication Technologies in
Tourism 2019 (pp. 359–370). https://doi.org/10.1007/978-3-030-05940-8_28
van Dalen, A., de Vreese, C. H., & Albæk, E. (2012). Different roles, different content? A four-country comparison of the role conceptions and reporting style of political journalists.
Journalism: Theory, Practice & Criticism, 13(7), 903–922.
https://doi.org/10.1177/1464884911431538
Weaver, D. A., Lively, E., & Bimber, B. (2009). Searching for a Frame. Science Communication,
Wilkins, A. S. (2018, April 1). To lag or not to lag?:Re-evaluating the use of lagged dependent variables in regression analysis. Political Science Research and Methods, Vol. 6, pp. 393– 411. https://doi.org/10.1017/psrm.2017.4
World Economic Forum. (2018). Towards a reskilling revolution. (January), 42. Retrieved from www.weforum.org
Zubiaga, A., Procter, R., & Maple, C. (2018). A longitudinal analysis of the public perception of the opportunities and challenges of the Internet of Things. PLoS ONE, 13(12).
Footnotes
1Autonomy & Control, chi2 (1, N =178) = 1.018, p = .313, Togetherness & Community,
chi2 (1, N =178) = .587, p = .444, Productivity & Effectiveness, chi2 (1, N=178) = 1.272, p = .259, Career & Development, chi2 (1, N =178) = .046, p = .830, Organisation & Society, chi2 (1, N =178) = 1.259, p = .262.
Appendix A
Exact keyword search (run seperately for each newspaper)
("Künstliche Intelligenz" ) AND (Arbeit OR Arbeitsplatz OR Firma OR Organisation OR
Unternehmen OR Angestellter OR Mitarbeiter OR Geschäft OR Beruf OR Arbeitnehmer OR Arbeitnehmerin OR Chef OR Job OR Vorgesetzter OR Geschäftsführer OR
Geschäftsführerin OR skill OR Weiterbildung OR Angestellte OR Mitarbeiterin OR
Industrie OR Wirtschaft OR Arbeitswelt OR Betrieb) Publication (Süddeutsche Zeitung OR Neue Zürcher Zeitung)
Roughly translated:
(“artificial intelligence”) AND (work OR firm OR organization OR company OR employee OR coworker OR business OR profession OR job OR boss OR skill OR training OR industry OR economy OR business world) Publication (Süddeutsche Zeitung OR Neue Zürcher Zeitung)
