Thuy Dung Le

Thuy Dung Le

Auto-generated profile, based on publicly available information

Thuy Dung Le

Department of Economic Sciences and Media, Technische Universität Ilmenau, Ilmenau, Germany

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 December 2024 in Current Sexual Health Reports
Verified
Journal Article
Open access
PDF file
NDNicola Doering
Nicola Doering
TLThuy Dung Le
Thuy Dung Le
LVLaura Vowels
Laura Vowels
MVMatthew J. Vowels
Matthew J. Vowels
TMTiffany L. Marcantonio
Tiffany L. Marcantonio
Purpose of Review: Millions of people now use generative artificial intelligence (GenAI) tools in their daily lives for a variety of purposes, including sexual ones. This narrative literature review provides the first scoping overview of current research on generative AI use in the context of sexual health and behaviors. Recent Findings: The review includes 88 peer-reviewed English language publications from 2020 to 2024 that report on 106 studies and address four main areas of AI use in sexual health and behaviors among the general population: (1) People use AI tools such as ChatGPT to obtain sexual information and education. We identified k = 14 publications that evaluated the quality of AI-generated sexual health information. They found high accuracy and completeness. (2) People use AI tools such as ChatGPT and dedicated counseling/therapy chatbots to solve their sexual and relationship problems. We identified k = 16 publications providing empirical results on therapists’ and clients’ perspectives and AI tools’ therapeutic capabilities with mixed but overall promising results. (3) People use AI tools such as companion and adult chatbots (e.g., Replika) to experience sexual and romantic intimacy. We identified k = 22 publications in this area that confirm sexual and romantic gratifications of AI conversational agents, but also point to risks such as emotional dependence. (4) People use image- and video-generating AI tools to produce pornography with different sexual and non-sexual motivations. We found k = 36 studies on AI pornography that primarily address the production, uses, and consequences of – as well as the countermeasures against – non-consensual deepfake pornography. This sort of content predominantly victimizes women and girls whose faces are swapped into pornographic material and circulated without their consent. Research on ethical AI pornography is largely missing. Summary: Generative AI tools present new risks and opportunities for human sexuality and sexual health. More research is needed to better understand the intersection of GenAI and sexuality in order to a) help people navigate their sexual GenAI experiences, b) guide sex educators, counselors, and therapists on how to address and incorporate AI tools into their professional work, c) advise AI developers on how to design tools that avoid harm, d) enlighten policymakers on how to regulate AI for the sake of sexual health, and e) inform journalists and knowledge workers on how to report about AI and sexuality in an evidence-based manner.

3 January 2026 in Current Sexual Health Reports
Verified
Journal Article
Open access
NDNicola Doering
Nicola Doering
TLThuy Dung Le
Thuy Dung Le
LVLaura Vowels
Laura Vowels
MVMatthew J. Vowels
Matthew J. Vowels
TMTiffany L. Marcantonio
Tiffany L. Marcantonio

4 March 2026 in Sexuality & Culture
Verified
Journal Article
Open access
PDF file
TLThuy Dung Le
Thuy Dung Le
NDNicola Doering
Nicola Doering
Non-consensual deepfake pornography is an emerging form of image-based sexual abuse (IBSA). It involves the non-consensual creation, distribution, or threat to distribute nude or sexual imagery generated using artificial intelligence (AI-IBSA). Although the legal, socio-cultural, and technological aspects of this phenomenon have received scholarly attention, the lived experiences of those involved remain underexamined. This study addresses this gap by analyzing Reddit discussions to examine the perspectives of perpetrators, victims, bystanders, and the community responses, drawing on routine activity theory (RAT). A deductive analysis of 134 Reddit initial posts identified who initiated the discussions (RQ1) and the relational, gender, and age dynamics between parties (RQ2). Seven typical discussion threads (269 comments) were inductively analyzed to explore the experiences of perpetrators (RQ3), victims (RQ4), and bystanders (RQ5), alongside community responses. Findings indicate that most threads were initiated by bystanders (45.5%; RQ1). Victims (85.8%) and bystanders (46.3%) were predominantly women, while perpetrators were mostly men (81.3%), often targeting individuals within their social networks (RQ2). Experiences ranged from perpetrators’ guilt to victims’ and bystanders’ distress. Across groups, participants reported emotional, psychological, relational, and interpersonal harms (RQ3–5). Through the lens of RAT, the study highlights the convergence of perpetrators’ motives and technical capabilities, victims’ vulnerability factors, and bystanders’ lack of capable guardianship. Community responses on Reddit offered support and moral reflection but also revealed patterns of harm minimization and technological fascination. These findings underscore the need for multi-level interventions, including mental health support, legal resources, digital literacy, and public education on consent and tech-facilitated abuse.

7 July 2026 in DOCA - Database of Variables for Content Analysis
Verified
Journal Article
Open access
PDF file
NDNicola Doering
Nicola Doering
TLThuy Dung Le
Thuy Dung Le
ASAnastasiia Shevtsova
Anastasiia Shevtsova
News and other non-fiction media – including educational material, advertising, science communication, and social media infotainment – increasingly use informational images generated with artificial intelligence (AI) tools. AI-generated images are visual outputs produced by computational models, most commonly diffusion models or generative adversarial networks, that synthesize novel imagery from learned patterns in large training datasets instead of capturing any real-world scene or object through a camera (Sanseviero et al., 2024). Unlike traditional digital editing or illustration, AI image generators can produce photorealistic images or stylistically coherent visuals (e.g., illustrations, diagrams) from simple text prompts. While AI-generated images have the potential for meaningful and helpful visualizations, they are also criticized – for example, for their lack of accuracy and diversity. These concerns are particularly salient in informational contexts, where images are expected to represent reality and where misleading or skewed visuals can shape audience beliefs, reinforce stereotypes, and erode trust in media. Drawing on theories about AI image generation, AI content moderation, AI generation errors, and AI biases (mainly representational and presentational biases), this entry introduces a codebook with a set of four key concepts and related variables and operational definitions designed to allow coders to consistently identify potentially problematic characteristics of AI-generated informational images. Field of Application/Theoretical Foundation In the field of visual communication (Thelander, 2018), images generated with artificial intelligence (AI) tools have gained increasing relevance for researchers and practitioners. AI image generators are now easily accessible through public text-to-image tools (TTI or T2I generators), leading to the growing use of AI-generated images across a broad range of entertainment (e.g., AI-generated pornography: Lapointe et al., 2025) and informational contexts (e.g., AI-supported health communication: Döring & Shevtsova, 2026). This widespread availability and adoption have made the characteristics and effects of AI-generated images an important topic of communication research. When used for informational visual communication, AI-generated images offer genuine opportunities: they may help explain complex structures or processes, attract attention, foster reflection or improve the persuasiveness of messages (e.g., Amri & Hisan, 2023; Reed et al., 2023). At the same time, there is growing concern that the broader use of AI image generators introduces new risks, including the dissemination of large amounts of low-quality AI images (so-called AI slop) that elicits fatigue and distrust in media, the deliberate generation of AI disinformation, and the unintentional creation and dissemination of problematic AI images by well-meaning communicators due to the technological limitations of AI models (e.g., so-called AI hallucinations or confabulations; Buzzaccarini et al., 2024). Developers of AI image generators are aware of the last-mentioned risk. They take different measures to avoid problematic AI-generated outputs and, hence, aim to design responsible AI (Gunasekara et al., 2025): they curate and filter training data, train their models to produce prompt-aligned and ethics-conforming images, and implement content moderation to block outputs that could violate user policies or laws (Döring & Shevtsova, 2026). However, it remains an empirical question how successful different AI companies are in providing text-to-image generators that produce trustworthy, accurate, and unbiased images when prompted to visualize different information topics or messages. Based on theories about AI image generation, AI content moderation, AI generation errors, and AI biases, four main characteristics of AI imagery are theoretically discussed and measured in content analyses (e.g., Ahmad et al., 2026; Döring et al., 2025; Döring & Shevtsova, 2026; Mehrabi et al., 2022): • AI image over-blocking: Blocking means the refusal of an AI image generator to create the requested image, returning a textual refusal instead. Over-blocking occurs when an AI image generator incorrectly refuses to generate an image in response to a legitimate prompt. Over-blocking can be an issue in many relevant information contexts where AI moderation systems do not adequately differentiate between harmful content (e.g., a sexualized image of a child) and harmless or helpful content (e.g., an image for a campaign against child marriage), and hence censor relevant visualizations, for example in fields such as sexual education, reproductive health, or the prevention of gendered violence (e.g., Vargas Penagos, 2024). • AI image flaws: AI image generators can produce flawed images that do not accurately represent the requested content. These flaws fall into two categories: visual errors (e.g., distorted body parts, non-existent objects, or illogical spatial relationships) and – for images with textual elements such as diagrams – textual errors (e.g., factually incorrect or misspelled image labels). Both are examples of what researchers call AI hallucinations or confabulations – a well-documented problem where AI models produce content that may look convincing at first glance but is factually wrong (e.g., Amri & Hisan, 2023). This happens because AI generates what is statistically probable, not what is accurate. Users who share AI-generated images without fact-checking all visual and textual details therefore risk spreading false or misleading visuals. • AI image representational biases: The systematic over- or underrepresentation of certain social groups in AI-generated images depicting people is referred to as representational bias. It concerns whether the full diversity of humans is adequately represented. Because AI models...

9 June 2026 in DOCA - Database of Variables for Content Analysis
Verified
Journal Article
Open access
PDF file
NDNicola Doering
Nicola Doering
TLThuy Dung Le
Thuy Dung Le
In the age of AI, the production of pornography has changed in two key ways: (a) purely synthetic images can be generated without any resemblance to real persons, and (b) deepfake pornography of any individual can be created (Döring et al., 2025b). Deepfake pornography refers to sexually explicit images or videos in which a person‘s face or likeness is superimposed onto pre-existing pornographic content, using AI-based face-swapping or image synthesis technology (Döring et al., 2025a). In most cases, this occurs without the consent of the depicted person and without the consent of the pornography performers whose material is appropriated. While the term deepfake pornography is widely used in public and academic discourse, more precise terms that foreground the nonconsensual nature of this practice are AI-generated nonconsensual intimate images (AI-NCII) and AI-generated image-based sexual abuse (AI-IBSA). A growing number of countries have enacted (e.g., Italy, the UK) or plan to enact (e.g., Germany) laws criminalizing the production and distribution of such nonconsensual intimate media content that depicts both private individuals — including minors — and public figures such as politicians, journalists, and performing artists. The harm of nonconsensual deepfake pornography is threefold: (1) it violates the right to privacy and personal integrity by undermining data protection, reputation, and psychological autonomy; (2) its strategic use to shame and defame women journalists, activists, and politicians threatens freedom of expression and gender equality in public and political participation; (3) it impairs the economic and moral rights of pornography performers whose work and bodies are exploited without authorization or compensation (Yavuz, 2025). On social media, considerable debate surrounds this phenomenon, ranging from potential perpetrators exchanging tips on how to generate such material, to bystanders seeking information on where to find or how to combat it, and victims searching for support (Döring et al., 2025a; Le & Döring, 2026). This entry provides a set of variables for identifying different engagement roles — perpetrator, victim, and bystander — as they appear in online discussion posts. Field of Application/Theoretical Foundation Deepfake pornography, also referred to as AI-generated image-based sexual abuse (AI-IBSA) or AI-generated non-consensual intimate images (AI-NCII), is a form of image-based digital violence that extends sexual and gender-based violence into the digital sphere (Döring et al., 2025a). Accordingly, theoretical perspectives on engagement in nonconsensual deepfake pornography require a gender-based approach. Gender role theory (Eagly, 1987), sexual script theory (Simon & Gagnon, 1984), social dominance theory (Pratto & Stewart, 2011), and objectification theory (Fredrickson & Roberts, 1997) help explain why men are more likely to engage in the creation and consumption of AI-IBSA, while targeted women are more likely to experience social, professional, economic, and psychological harm. Women who are targeted with deepfake pornography face a disproportionate burden compared to men because their sexuality is socially policed through sexual double standards and slut-shaming norms, meaning that a fabricated sexual image activates pre-existing mechanisms of reputational destruction, ostracism, and silencing that have no equivalent for male targets. Several empirical studies have explored engagement with deepfake pornography. Based on large-scale online surveys, Umbach et al. (2026) found that 3.2% of adults across three countries had created, distributed, or threatened to distribute AI-IBSA, while 18.0% reported viewing such content. Umbach et al. (2024) identified a 2.2% victimization rate of AI-NCII across ten countries. Across studies, men and younger individuals consistently show higher rates of perpetration, whereas women disproportionately report emotional, psychological, and professional harms. To better understand these dynamics, recent studies apply the participant role approach (Salmivalli, 1999) and routine activity theory (Cohen & Felson, 1979), distinguishing between perpetrators, victims, and bystanders. Perpetrators create or distribute deepfake pornography, victims are depicted without consent, and bystanders encounter such material without directly participating in the abuse (Le & Döring, 2026). Mirroring above-cited survey results, social media analyses show that perpetrators posting about the issue are primarily identified as men, while victims and bystanders are predominantly women, often within close social networks of perpetrators. Experiences associated with these roles range from perpetrators publicly discussing the technicalities and the joys of creating such material to bystanders seeking intervention information and victims searching for support (Döring et al., 2025a; Le & Döring, 2026). Studies underscore the multiple harms of nonconsensual deepfake pornography and the need for role-specific interventions, including stronger platform governance, digital ethics education, bystander intervention programs, trauma-informed support services, and legal protections addressing AI-IBSA. References/Combination with other Methods of Data Collection Social media content analyses (e.g., of Reddit, X/Twitter, and Facebook), as well as analyses of other materials such as court documents (Witt et al., 2024), provide relevant insights into people’s experiences with deepfake pornography across different participant roles. Such content analysis studies can be conducted as qualitative analyses (e.g., Popova, 2020), manual quantitative analyses (e.g., Döring et al., 2025a), or computational analyses (e.g., Han et al., 2024). Some studies combine qualitative and quantitative content analysis methods to capture both large-scale patterns and nuanced experiences...