Deepfake Survey

I have prepared a survey as part of my postdoctoral research project at the Taiwan Foundation for Democracy. I would be very grateful for your participation. All participant information will remain strictly confidential and will not be disclosed under any circumstances. The results will not be used in any academic publication without prior ethical approval and will be used solely for the Taiwan Foundation for Democracy’s final report.

LINK: https://docs.google.com/forms/d/e/1FAIpQLSfrpkGSoYSb9lc68nKh5iHEJ8bkHabyKPQXvKvwegMhAtZRZQ/viewform?usp=dialog

Evaluating web-based deepfake detection tools: Risks, limits, and practical solutions for information security

ABSTRACT

The rapid advancement of artificial intelligence (AI) has facilitated the rise of deepfakes—synthetic audio, image, text, and video content generated by deep learning algorithms that can convincingly fabricate events, statements, or identities. While deepfake technology has positive applications in fields such as education and entertainment, it has predominantly been exploited for malicious purposes, including sexual exploitation, political manipulation, fraud, and disinformation campaigns. In response, web-based deepfake detection tools have emerged as accessible alternatives to algorithmic models, enabling non-experts such as researchers, journalists, and policymakers to verify suspicious content. However, systematic evaluations of these tools remain scarce in the academic literature. To address this gap, the present study examines 37 web-based deepfake detection tools. Employing descriptive and statistical analyses—including accuracy analysis, weighted score, and margin of error calculations—the study evaluates tool performance and categorizes them into high, moderate, and low accuracy groups. Results indicate that photo detection tools are the most reliable, with several achieving weighted scores above 80 percent, while video detection tools demonstrate only moderate effectiveness. Audio detection remains underdeveloped, showing significant weaknesses in both accuracy and tool availability. Hive Moderation stands out as the most versatile platform, though none of the tools offer comprehensive, multi-modal, or real-time detection. The findings highlight critical gaps in current detection capabilities, particularly in audio and in large, long-form media. This underscores the importance of scalable, multi-modal, real-time solutions to safeguard democratic institutions, organizations, companies, individual security, and public trust.

LINK: https://www.tandfonline.com/doi/full/10.1080/07366981.2025.2564783

Deepfake Interest in South Korea: A Temporal Analysis of Google Trends from 2017 to 2024

Deepfake technology, which utilizes artificial intelligence to generate hyper-realistically manipulated videos, images, texts, and audio, has garnered significant public and academic interest. The proliferation of deepfakes, especially in non-consensual pornography, financial fraud and political misinformation, has sparked ethical, moral, legal, and security debates worldwide. While existing research predominantly focuses on deepfake detection, legal frameworks, and their potential impact on the democratic process, few studies have examined public interest in deepfakes and the factors influencing search behavior. This study addresses this gap by analyzing public interest in deepfakes in South Korea, using Google Trends data from January 2017 to August 2024. This timeframe is particularly significant as it encompasses the initial emergence of deepfake technology in 2017 and its increasing use in fraudulent and non-consensual content in South Korea. The country represents a unique case due to its global leadership in deepfake-related searches, widespread consumption of non-consensual sexual deepfakes, and frequent occurrence of deepfake fraud. This study employs dictionary-based text analysis to categorize search queries into three main themes: sexual content, techniques for creating deepfakes, and methods for accessing deepfake materials. The findings indicate that 77.81% of searches are related to non-consensual sexual content, primarily targeting female celebrities. Contrary to global trends, political deepfakes did not significantly influence search patterns in South Korea. These insights highlight the urgent need for stronger regulatory frameworks and technological interventions to mitigate the harms associated with deepfakes.

LINK: https://dergipark.org.tr/tr/pub/ikad/issue/90816/1570974

Bridging Disciplinary Gaps and Methodological Challenges in Understanding Deepfake Discourse: A Study of Turkish Reddit Posts Related to Deepfake

Abstract

This study explores the evolving landscape of deepfake discussions, focusing particularly on Turkish posts on Reddit. While deepfake technology gained prominence in 2017, the academic community's interest surged, with studies primarily concentrating on deepfake detection. However, there is a dearth of research on public perceptions and online discourse surrounding deepfake. Through text analysis, this study explores Turkish Reddit posts, uncovering that a substantial portion (69.4%) focuses on deepfakes with sexual content, particularly featuring celebrities (60.2%). Politics also emerges as a prominent theme, constituting 22% of the content, with an additional 8.6% dedicated to technical aspects of deepfake creation. Over time, posts on sexual deepfake content, particularly those featuring celebrities, dominate, but during the May 2023 presidential and general elections, political deepfakes gained traction. Despite shedding light on content trends, the study faces two significant limitations. The first limitation is methodological. Studies relying solely on text analysis struggle to offer comprehensive insights into people's perceptions of deepfake. Therefore, a more profound understanding can be attained through qualitative face-to-face interviews, allowing for a deeper exploration of individuals' perspectives. The second limitation is disciplinary. Research encompassing diverse disciplines such as psychology, sociology, and media studies would yield more nuanced results in comprehending both the broader framework of public and societal perceptions on the topic and their deeper intricacies.

Key words: Deepfake, Türkiye, Reddit, Interdisciplinary Approach, Dictionary-based Text Analysis, Mixed Methods

Link: https://ojs.library.carleton.ca/index.php/J-ICSLAC/article/view/4839/3551

A Comprehensive Analysis of Public Discourse and Content Trends in Turkish Reddit Posts Related to Deepfake

ABSTRACT 

This study investigates the content and changes in deepfakes-related discussions on 5,220 Turkish Reddit posts from October 2019 to August 2023. Although the academic community has shown an increasing interest in deepfakes since 2017, focusing on detection methods and the technology itself, scant attention has been paid to public perceptions and online debate. The analysis reveals that 69.4% of the examined posts feature deepfake content with sexual themes, with celebrity women being the primary targets in 60.2% of cases. In contrast, 22% of the content is about politics and political figures, while 8.6% provides technical guidance on creating deepfakes. The study also observes content changes over time, noticing a rise in sexually explicit deepfake posts, particularly involving celebrities. However, in May 2023, coinciding with the presidential and general elections in Türkiye, discussions about politics and political figures have significantly increased. This study sheds light on the changing landscape of discussions, emphasizing the predominant presence of sexual content and the increasing prevalence of political content, particularly during election seasons. 

LINK: https://iga.pknu.ac.kr:446/html/jga/online-jga.php?mode=view&idx=28