Fakt oder Fake? Erkennung und Wahrnehmung von Deepfakes

Abstract

The rapid development of deepfake technology poses a significant challenge in the digital media landscape. The deepfakes generated by advanced machine learning techniques, which have a high level of authenticity, jeopardise the credibility and integrity of information. This technology is increasingly being used in harmful contexts such as the spread of disinformation or political manipulation. In view of these risks, it is essential to develop reliable methods for detecting deepfakes and to promote media literacy among users in order to minimise the spread and influence of fake content. This thesis is divided into several chapters. The first part discusses the theoretical foundations, while the second part presents an empirical study on the recognition and perception of deepfakes. Following the introduction, the second chapter provides a detailed discussion of the history of deepfakes, the basic definitions and the development phases of deepfakes. The areas of application of this technology are also highlighted. The third chapter is dedicated to the detection of deepfakes. Relevant parameters and methods are first analysed. Chapter four analyses the effects of deepfakes on users and their perception in social media. The empirical part of the thesis (chapters five and six) comprises two online experiments. The first experiment serves to evaluate the technical performance of deepfake detection tools, while the second experiment examines the human ability to recognise deepfakes. The methodology combines a literature review with quantitative online experiments to provide a comprehensive picture of the current challenges and potential solutions. The results of the empirical study show that both technical and human detection methods for recognising deepfakes are associated with significant challenges. An evaluation of existing deepfake detection tools has shown that while some tools are highly accurate, there are still a significant number of false positives and false negatives. The ROC curve analysis and the precision recall curves illustrate the performance and shortcomings of the tools analysed. Based on the results of the online survey of users, it can be deduced that the ability to recognise deepfakes is subject to strong fluctuations and is influenced by a variety of factors. These include media literacy and awareness of deepfakes. It is also clear that deepfakes can significantly affect trust in digital content and social media. Overall, the study emphasises the relevance of increased research and development at both a technological and educational policy level in order to increase society's resilience to the dangers of digital manipulation.

Date of Award	2024
Original language	German (Austria)
Supervisor	Andreas Auinger (Supervisor)