Browsing by Person "Vogelgesang, Jens"

Now showing 1 - 4 of 4

AI in media organisations: Factors influencing the integration of AI in the newsroom
(2024) Grimme, Meike; Vogelgesang, Jens
Exploring the potential of immersive virtual reality for social science research
(2025) Hepperle, Daniel; Vogelgesang, Jens
Immersive Virtual Reality (IVR) holds out the promise of laboratory‐grade experimental control while preserving much of the richness of real world experience, yet several issues remain unresolved. The central theme of this dissertation is spanned around the idea of using IVR as a tool to help researchers conducting empirical studies in the domain of social sciences. To address the question, the thesis incorporates five related studies. Paper 1 introduces the main areas of concern in a typical research process and offers guidance where IVR toolkits might be a valuable addition. Based on those identified areas of concern, the paper suggests solutions such as automation workflows in order to reduce the human‐error (i.e. using predefined scenes that already offer different basic standard methods in order to track all changes in the virtual world). Paper 2 examines seven open‐source IVR toolkits, demonstrating how to standardize modular scene setup, participant sensing, and data export. The analysis clarifies the features currently available in different toolkits and provides a basis for researchers to decide which features and toolkits offer the greatest benefits. We also discuss novel features such as AI‐based analysis which is not present in most toolkits. Based on this we provide guidance for future IVR‐based research software development. Paper 3 offers a PRISMA guided systematic scoping review of 56 publications, mapping the field of studies that compare either IVR with the real world or IVR with 2D screens. In short, the review finds that there are more similarities than differences between IVR and the real world. However, between IVR and 2D screens, more findings show differences between the two environments than similarities. Paper 4 provides an empirical test of transferability: the mere‐exposure effect was successfully replicated in the original study setup (n = 70 m; 49 f) as well as within IVR (n = 39 m; 24 f). Overall, the studies demonstrate the efficacy and practicality of employing IVR to induce effects analogous to those observed in a real‐world context in the case of the mere exposure effect. Finally, Paper 5 introduces asymmetric normalization, a novel manipulation that decouples self‐perception from how others see a participant in social IVR, thereby expanding the experimental design space with the possibility to reduce bias. This may concern various attributes such as size or age, as well as other visual or spatial characteristics. Pilot data from 40 participants shows that this technique reliably alters interpersonal‐distance preferences, opening a new design space for social science research. This dissertation advances research in the social sciences by showcasing the capabilities of IVR toolkits and illustrating how they can be integrated into established research processes. It further demonstrates that a cognitive‐affective mechanism (mere exposure) also is present in IVR. Moreover, it introduces asymmetric normalization as a novel manipulation technique that expands the experimental design space beyond what is feasible in physical laboratories. For research practice, the papers within the dissertation lower the barriers to entry for non‐technical scholars, provide a decision matrix for selecting and extending IVR toolkits. Together, they shift IVR from a technological novelty to a mature, shareable, and cost‐effective platform for conducting experiments in the social science domain.
The role of public opinion on ethical AI principles and its implication for a common good-oriented implementation.
(2024) Kieslich, Kimon; Vogelgesang, Jens
Artificial Intelligence (AI) has a tremendous impact on society. While artificial intelligence encompasses a variety of different systems, algorithmic decision-making (ADM) systems in particular are being used to augment or even replace human decision-making. Because ADM systems are susceptible to ethical ramifications, such as fairness issues, opacity, and lack of accountability, how to manage the implementation of ADM is a societal challenge. This is particularly relevant when ADM is used in high impact situations that can potentially affect every member of society, such as in the public sector. As a way to address the harms of ADM, ethical guidelines have been proposed by companies and policy makers. However, scholars argue that these guidelines lack reinforcement mechanisms and that additional incentives are needed for decision makers to actually invest in ethical systems. In my dissertation, I focus on one potential factor that could contribute to the development of ADM in the public interest or for the common good, respectively -- public opinion. Critical public discussions about whether and how society wants ADM to make decisions can put pressure on decision makers to actually develop and implement ADM systems that adhere to ethical standards. The public does this by articulating (political) demands and by legitimizing or critiquing current practices. My research is situated within normative theories of the political public sphere, which propose different approaches for public discourse in the formation of democratic will, as well as the Society in the Loop framework, which emphasizes the need to include citizens' perceptions in decision-making about ethical trade-offs in the design and implementation of AI systems. This cumulative dissertation consists of four peer-reviewed papers: Paper 1 critically discusses at a theoretical level the role of public opinion as an influential factor in technology adoption. It argues that given the serious implications that the emergence of AI could have on society, public opinion can be a crucial incentive for both technological and political decision-makers to invest in AI for the common good. However, the paper also acknowledges and discusses the limits of public influence, and outlines potential avenues for greater inclusion of the public voice. Paper 2 presents data from a large-scale survey on public opinion on AI in Germany. In particular, it examines 1) what citizens have in mind when thinking about AI, 2) what role ethical AI issues play in this regard, 3) which demographic and AI-related factors contribute to a higher salience of (ethical) AI issues, and 4) what consequences the salience of ethical AI issues has in terms of AI avoidance and engagement in public discussions about AI. The paper's main contribution is to provide an empirical database on the engagement of German citizens with AI, thus helping to assess citizen influence on technological and political decision makers. Paper 3 empircally examines citizens' perceptions of the ethical trade-offs that must be made in the design process of AI systems. It uses a use case of AI in the public sector which, from the normative standpoint of AI development for the common good, requires citizen participation. Paper 3 provides insights into 1) citizens' ethical preferences for the design of AI systems, 2) shows that there are different publics with different preferences, and 3) describes how these publics are constituted in terms of demographic as well as AI-related factors. The paper's main contributions are to propose a measurement for evaluating ethical AI principles and to describe different preference patterns of the German public. Paper 4 delves deeper into the topic of the consequences of (non-)compliance with ethical design of AI systems. Again, the paper presents a use case of AI in the public sector and discusses the role of trust in AI as influential factor leading to the legitimization of AI technology. The main contribution of the empirical paper is to elaborate on the role of trust in AI, as it is treated as a major factor in empirical research and policy discussion in light of a widespread implementation of AI. In summary, my dissertation contributes to the literature on public opinion research, the Society in the Loop framework, and the efforts of the FAccT (Fairness, Accountability and Transparency) community, and specifically discusses the role of the public as a potential critical voice in the design and implementation process of AI systems. On a methodological level, I propose a measure for exploring the preference for trade-offs in ADM systems. On an empirical level, I provide a rich empirical (baseline) data on citizens' perceptions of ADM on which future studies can build. On a theoretical level, I discuss my findings in terms of normative theories about the role of the public and the Society in the Loop framework. On a practical level, I address the interplay between public opinion and the economic, media, educational, and political & legal sectors, and I elaborate on future steps that can be taken to strengthen the common good orientation in the development and implementation of AI systems.
Subtitling vs. dubbing and original version
the effects of different translation methods on consumer behavior towards product placement on audiovisual content
(2024) Noschang, Luis Octávio; Vogelgesang, Jens
The main objective of this dissertation was to investigate whether different versions of a program concerning audiovisual translation formats as well as multitasking would influence product placement effectiveness on audiences. Also, if there would be any relation between levels of multitasking and the version of the program being watched. The first problem to be addressed was determining whether the above-mentioned questions have not been the subject of research in the past. A systematic search on the three most prominent research databases available (Scopus, Web of Science, and EBSCO) followed by a manual analysis of the publications was conducted and research items of these publications were further investigated to identify effects caused by placements as well as variables or drivers of product placement effectiveness. Some 320 research items (hypotheses, research questions, empirical generalizations, and results) featured effects caused by placements. The total of effects identified was 11. Brand attitude and brand recall were the two most prominent effects identified. In the case of variables, or drivers of effectiveness, 417 research items included 57 different variables that could be divided into 2 categories: 33 variables derived from characteristics of the audience and 24 variables pertaining to the content, medium, or placement. None of the publications featured audiovisual translation formats as variables, confirming the research gap in the field. For multitasking, 2 publications were identified that covered the subject, nevertheless, in a different form than the intended of this dissertation. The empirical investigation was performed through an experiment where participants could freely choose between four options of pre-defined episodes of two different sitcom series as well as the version between original, dubbed, and subtitled, the last only as a choice for those that could not understand the spoken dialogues. Some 2302 participants were recruited and answered a questionnaire containing questions about aided recall of brands placed in the episode, and multitasking behavior while watching the program. Results on brand recall of placements remained constant between groups watching different versions of the programs. Punctual percentual differences could be identified, nevertheless, reversed results appeared for different products, and Chi-square tests revealed no statistically significant differences. The presence of subtitles did not significantly alter brand recall. Brand recall also remained stable among viewers of programs in original, dubbed, and subtitled versions. Results compared brand recall of placements between viewers who declared that they engaged in other activities while watching the program and viewers who did not multitask. The levels of brand recall did not differ significantly between these two groups. The time spent not looking at the screen was the next aspect evaluated in the experiment. Here, no significant difference in the levels of brand recall was identified between the viewers who did not look at the screen for less than 1 minute and the viewers who looked away from the screen for between 1 and 5 minutes during programs of around 20 minutes in length. The different versions of the program were tested with regard to their influence on multitasking behavior. The presence of subtitles decreased viewers’ levels of multitasking when compared with those of viewers who watched the program in its original or dubbed versions with no subtitles added to enhance comprehension of the spoken dialogues. The levels of multitasking remained constant between viewers of the dubbed and original versions of the program. To increase the validity of the results, a method of observing the respondents was also implemented. A total of 274 respondents were recorded in videos that were later analyzed manually. This method was introduced to more precisely determine the time they spent not looking at the screen as well as to detect false positives on the respondents’ answers.