Threats to conservation from artificial‐intelligence‐generated wildlife images and videos

Generative artificial intelligence (AI) can be used to create hyperrealistic images and videos of wild animals (e.g., OpenArt, DeepAI), and their inappropriate use and dissemination can distort people's perceptions of wildlife. The recognition of species diversity by society, especially among children, is largely influenced by the limited scope of media coverage of, for the most part, a narrow range of iconic and charismatic species (Ballouard et al., 2011; Valente et al., 2020). This situation has assumed heightened significance due to the potential for AI images and videos to alter the information presented by the media. Three distinctive characteristics of contemporary society can exacerbate this issue: the pervasive use of social media, anthropomorphizing of animal behavior, and disconnection from nature. First, the power that social media has to influence people's perceptions of wild animal can be problematic (Lenzi et al., 2020; Riddle & Mackay, 2020). Some social media users have already posted AI-generated videos and pictures of wild animals and claimed the content is real (Appendix S1). Posting of videos and images of animals that appear realistic can result in a significant increase in the number of followers, impact, and feedback, which is an attractive prospect for many individuals. Consequently, it is reasonable to anticipate an increase in the prevalence of AI-generated content on social media. On social media, AI-generated, realistic yet misleading portrayals of wildlife are reshaping public attitudes toward biodiversity. Generative AI allows for effortless creation of entirely fictitious species, behaviors, and ecological interactions, further distancing the public from real-world conservation challenges. Negative framing, highlighting human–wildlife conflicts, can lead to negative perceptions and comments, whereas positive framing can foster a desire for conservation (Ballejo et al., 2021). The capacity for social media to exert influence is likely to increase significantly if one considers the potential impact of the combination of inaccurate framing and biased AI-generated photographs and videos. Such a combination could be an effective instrument for the dissemination of misinformation, especially because AI-generated wildlife content is increasingly indistinguishable from real footage, making its detection and regulation challenging. Second, some people anthropomorphize wild animals (Horowitz & Bekoff, 2007). Although anthropomorphism can be useful in educational and conservation contexts, in that it fosters empathy and interest in animals, the increase of anthropomorphizing in AI content can result in misconceptions about animals' genuine behaviors and ecological features, which are notably distinct from those of humans (Mota-Rojas et al., 2021). Third, the growing disconnect between society and the natural world (Barrable & Booth, 2022) can be attributed to the increasing reliance on technology and the subsequent lack of outdoor experiences and the concentration of human populations in urban areas that are geographically distant from natural environments (Soga & Gaston, 2023). This disconnection from nature implies a general lack of awareness of the natural world, which could make people more likely to believe AI-generated images and videos are real. This dissociation between people and nature means people may be easily misled by artificial representations of wildlife. The inability to discern genuine natural patterns makes individuals more vulnerable to distorted AI narratives, thereby exacerbating their distance from the natural world. The proliferation of AI-generated images and videos depicting fictitious animals presents significant challenges for conservation efforts (Campos et al., 2023). Every biological and ecological trait of any animal can be manipulated by AI. We present specific examples of how AI-generated content can uniquely distort society's knowledge and perception of animal biology and conservation (additional examples in Appendix S1). Content generated by AI may enhance social media's existing bias toward portrayal of affiliative behaviors. Unrealistic interspecies relationships are often fabricated, and emotional expressions are exaggerated. These synthetic portrayals, indistinguishable from real footage, contribute to misinformation by reinforcing fictional narratives about animal social structures and cognitive abilities (Stumpf et al., 2024). Together, these portrayals and anthropomorphic perceptions of animals' social and cognitive abilities (Callahan et al., 2021; Eddy et al., 1993) may strongly influence people's perceptions of wild animal behavior. For instance, an AI video depicting different bird species shielding chicks from rain is often described by users as "true mother love." The misperception is that female animals exhibit human-like love of their offspring. However, in 90% of bird species, males also participate in caring for young (Wesołowski, 1994), and in many reptile, amphibian, and fish species, there is no parental care at all. Some AI videos depict interspecific interactions that are implausible, for example, affiliative behaviors between animals of different species, including between predators and preys, and between parasites and hosts. Some videos depict friendships between humans and wild animals, which can be dangerous if, in an encounter with an animal, a human fails to recognize it as a potential threat. One AI-generated video portrays a polar bear being rescued by fishers displaying exaggerated expressions of gratitude. This video exemplifies the unrealistic narratives AI-generated videos present about human–wildlife interactions. These false portrayals can lead to misconceptions about animal behavior and affect attitudes toward wild species and conservation. Videos generated with AI depicting wild animals in domestic or affectionate interactions with humans may contribute to an increased interest in acquiring these species as pets, exacerbating illegal wildlife trade (Clarke et al., 2019; van der Meer et al., 2019). Realistic representations of animals can distort public perception of wildlife, leading to the misconception that endangered species are more abundant or less vulnerable than they actually are (Thomas-Walters et al., 2020). By saturating digital spaces with realistic but fabricated portrayals, AI-generated photos and videos may further weaken the emotional and cognitive connections people have with real animals, complicating efforts to foster conservation support. This technology allows for the rapid, large-scale, and automated creation of realistic but false representations, which can be tailored in real time to audience preferences and social media trends. As AI tools become more sophisticated, distinguishing real from fake content will become increasingly challenging, further complicating conservation communication efforts. Moreover, if the public is unable to distinguish between actual threats to biodiversity and fictionalized narratives, the perceived urgency to act may diminish. This misperception could result in reduced financial contributions and volunteer engagement for conservation projects, which would directly affect initiatives aimed at protecting vulnerable species. In an era where biodiversity loss demands immediate actions, the unintended consequences of AI-driven social media posts further complicate efforts to bridge the gap between public understanding and action (Wearn et al., 2019). Another risk is related to the incorrect identification of animals; some AI-generated images have very distinct morphological characteristics that differ significantly from those of the real species. Accurate identification of species is a fundamental step to understanding their abundance and distribution, for which citizen science is essential (Cavadino et al., 2024). However, AI-generated images that do not correctly depict animals could affect the ability of people to identify species. At the farthest reaches of the visual information generated by AI, there are creations of cryptid species, such as tiger-fish, snake-bird, and hippo-lizard. These are often easily identified as AI generated, but in some comments the social media users ask if the images are real or not. These funny videos and pictures could be problematic, especially for younger audiences who have not yet fully developed critical thinking skills and have limited biodiversity literacy (Hooykaas et al., 2019). In this case, AI-generated images blur the line between reality and fantasy, potentially leading them to believe that cryptid creatures or mythical animals, such as unicorns, exist and causing confusion and misunderstandings about the natural world. The prevalence of such AI-generated content can contribute to broader misinformation issues, creating distorted perceptions of reality. The spread of misinformation on social media is a major threat to current societies due to its rapid spread and potential to cause harm across various topics, including health, politics, environment, climate, education, and agriculture (Allcott et al., 2019; Treen et al., 2020). On social media, misinformation can spread farther and faster than true information (Vosoughi et al., 2018), and AI can contribute to increases in social inequality. Use of AI can exacerbate these issues by enabling the rapid creation of realistic fake content—for instance, videos that amplify false narratives about human–wildlife conflicts, targeted to specific audiences (Bontridder et al., 2021). Deepfake videos might show animals attacking humans or livestock or invading urban areas, which can create unnecessary fear and lead to harmful actions against wildlife. Posting videos and pictures of animals and mentioning they were captured in locations where the species do not occur is another undesirable facet of biodiversity misinformation that can produce confusion concerning which animal species are naturally present in a certain place. If a significant portion of a society already has serious difficulties discerning which animals are native species and which are not (Ballouard et al., 2011; Genovart et al., 2013), such AI-generated videos can exacerbate this confusion. It is well established that social media can motivate people to visit some specific locations (Gretzel, 2019), increasing tourist pressure on natural areas with undesirable adverse effects (Bergman et al., 2022; Larson et al., 2016). Consequently, posting fake videos and pictures of wild animals, particularly viral content, and falsely claiming they were captured in specific locations can exacerbate overtourism and its detrimental impacts on ecosystems. Probably AI-generated content will be biased toward mammals because images and videos of mammals tend to be the most engaging in social media (Jain et al., 2019; Shaw et al., 2022). This bias can influence public perceptions and behaviors toward these animals (Martín-López et al., 2009), affecting public opinion, scientific research, and conservation funding, with conservation projects aimed at taxa that are underrepresented in social media receiving less funding. Finally, information available on the internet, including digital data, social media posts, web pages, news, and video-sharing platforms, is a common source of information used by wildlife researchers to answer key ecological questions (Jarić et al., 2020) related to, for example, species occurrences, distribution patterns, and animal behavior (Correia et al., 2021; Soriano-Redondo et al., 2024; Vardi et al., 2024). Because this information needs to be filtered and processed before use in scientific research, AI-generated images and videos can hinder this preliminary step. Therefore, the amount of online data sources created by AI could make it very challenging to use these data to address ecological questions. The potential for confusion that AI-generated photos and videos on social media may provoke is likely to intensify with the creation of increasingly realistic images and videos. This underscores the necessity of indicating whether content has been generated by AI, thereby preventing the spread of misinformation and preserving the integrity of scientific knowledge about conservation. This issue is far from trivial because it is likely that AI-generated content will become more common than content created by humans. Therefore, it is urgent to implement initiatives that educate children and adults on how to critically engage with digital content. The fostering of media literacy skills could empower individuals to question the authenticity of images and videos posted on social media. This would reduce the spread of misinformation and help ensure that people are better prepared to distinguish between reality and fiction in an increasingly digital world, thereby reducing the influence of AI-generated distortions. Integrating these educational efforts into curricula and awareness campaigns will promote a more informed society, one that is more discerning in its consumption of AI-generated content and that views nature and wildlife conservation with greater objectivity and supports more informed conservation decisions. Nevertheless, the principal question remains unresolved: who decides the boundaries of information sharing? In this context, the filtering of AI-created fake content related to wild flora or fauna may appear to be another problem related to consumption of online content. It seems unlikely that common regulations worldwide can be implemented in the near future. Therefore, we strongly believe that educational content on wildlife should alert users to the fact that not everything they see on the internet is real. We thank the students who contributed to this research by providing valuable AI-generated content on social media. We are particularly grateful for the support of Project FCT-22-17727, funded by the Fundación Española para la Ciencia y la Tecnología (FECYT)—Ministerio de Ciencia, Innovación y Universidades, and Project BIOD22_00033_17_PPCB, funded by the Junta de Andalucía (European Union—NextGenerationEU). Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.