In a groundbreaking development, researchers at the Duke MaRRS Lab have created “deepfake whales,” which scientists believe could serve as essential tools for conservation efforts. This pioneering new method employs artificial intelligence (AI) to address urgent emerging challenges in ecological research. It uniquely tackles challenges such as data scarcity and species detection. New research published in Nature Ecology & Evolution shows how AI-generated images are helping us better understand whale populations. Most notably, it shines a spotlight on the constant peril faced by the critically endangered North Atlantic right whale.
The research team, led by Ph.D. student Henry Sun, prioritized cryptic coloring and other realistic details when generating images of whales. Unsurprisingly, their mission was to improve AI models for spotting these majestic creatures in actual aerial footage—as directed by the very experienced Holly Houliston and Juliet Wong. To rectify this, the team used cutting-edge deep-learning techniques to produce anatomically precise representations of multiple whale species. This included a rare two-tailed humpback, as well as an AI-generated North Atlantic right whale.
Advancing Conservation Efforts
The implications of creating deepfake whales are profound to the world of ecology. As Sun stated, “One of the big challenges in ecology is the idea of data scarcity.” Through this process, researchers can create synthetic images to supplement current datasets. This process helps fill the gap in the lack of examples for some species or age classes.
Sun’s research highlights a critical aspect of this work: “Max has been training deep-learning models using both real images and some of the fake images that I’ve made,” he explained. This combined to create a more powerful AI model, resulting in the effective identification of real whales in diverse habitats. The results of the study found that fine-tuning techniques created more precise images of North Atlantic right whales. In comparison, images produced only from text or image prompts tended to be less accurate.
The future uses of deepfake technology go far deeper than the creation of fake images. These artificial models can be used to train AI tools to identify endangered species in their natural habitats. We found this approach to be a powerful new tool for efficiently monitoring whale populations.
The Role of AI in Ecological Research
The intersection of CS and ES, also known as environmental informatics, is growing significantly. Researchers foresee a growing need for AI literacy among ecologists. Sun noted, “Something that I’m extremely interested in is capacity-building for natural scientists in the realm of artificial intelligence, because I think increasingly, these are skills that everyone needs.”
The ethical issues associated with using AI for ecological research are more urgent than their novelty may imply. Houliston cautioned that researchers must be clear about their ecological questions before generating synthetic imagery: “You have to be really clear on the ecological question you’re trying to answer.” This care and attention to detail helps make sure that AI generated images are used to meet targeted research needs, rather than perverting the course of scientific inquiry.
Despite these advances, the difficulty in producing anatomically precise illustrations still remains. Sun remarked, “Sometimes the diffusion model produces anatomically deformed whale images, like whales that are conjoined or whales with multiple sets of fins, which shows that it hasn’t exactly learned the most accurate representation yet.” These concerns underscore the importance of making ongoing improvements to AI models. Perhaps most importantly, they demonstrate the breaking point where collaboration between ecologists and computer scientists is absolutely essential.
Looking Ahead
The future of conservation may be greatly enriched through the developments made by deepfake technology. Dave Johnston emphasized the importance of remote sensing and big data in ecology: “We are truly in the age of big data when it comes to remote sensing in ecology and conservation.” This increasing use of technology has created countless new possibilities and equally as many new hurdles for environmental scientists trying to protect vulnerable species.
As researchers continue to refine their methods and explore new applications for deepfake whales, they underscore the importance of integrating innovative technologies into conservation strategies. This new research, DOI 10.1038/s41559-024-02623-1, is a prime example of how AI can help address ecological data gaps. It uncovers important new knowledge on the life history and distribution of imperiled species.