AI-generated “deepfake” images are stepping into the world of conservation—offering a novel way to tackle the pervasive problem of data scarcity in monitoring endangered species. Researchers at Duke University’s Marine Robotics and Remote Sensing Lab are using synthetic whale imagery to train AI models that detect rare marine species from aerial and satellite footage, such as the critically endangered North Atlantic right whale, reports Phys.org.
These whales number fewer than 400 worldwide, making it tremendously difficult to gather ample photographic data for effective AI training. By deploying diffusion models, the research team generates realistic synthetic images of right whales—and even humpbacks—as additional training examples. Through a process called fine-tuning, they refine base AI models using a small, curated dataset, improving anatomical accuracy and realism in these deepfakes.
These synthetic images could significantly enhance AI detection tools, helping conservationists locate and monitor elusive whale populations more effectively. This approach supports broader efforts—including a space-based whale detection system developed in collaboration with Canadian agencies.
However, the research is ongoing. The next step is to validate whether these deepfake-generated images truly improve whale detection performance in real-world conditions. As environmental scientists explore integrating AI into their toolkits, questions about ethical, energy, and sustainability considerations remain—and will need careful attention moving forward.