How Deepfake Animals Are Aiding Real Conservation

You have likely seen them on social media: unbelievable videos of wild animals, like a wolf befriending a house cat or a bear on a trampoline. Most are fakes, created with artificial intelligence. But what if this same technology could be used for a good cause, like aiding wildlife conservation?

That's the question driving a new initiative at the Duke Marine Robotics and Remote Sensing Lab. Scientists there are using AI to generate synthetic images of endangered species to help fill crucial data gaps in monitoring systems. While ecologists are increasingly using AI to analyze wildlife data, some species are so rare that there isn't enough information to train the AI effectively.

Dave Johnston, director of the lab, and doctoral student Henry Sun decided to flip the script on deepfakes. "Instead of always worrying about the negative aspects of [deepfakes], let’s flip the script. … How do we actually use this to actually make our work better?" Johnston said. Their goal is to use these synthetic images to train AI-powered satellites to track and protect endangered species from threats and the impacts of climate change.

This approach is part of a growing trend, but it comes with its own set of challenges, from the reliability of the data to the environmental footprint of the technology itself. As AI use grows, scientists must figure out how to leverage it for conservation without worsening the very problems they're trying to solve.

The Challenge of a Data-Scarce Species

Despite being as large as a school bus, the North Atlantic right whale is incredibly difficult to find. Once common along the eastern coasts of the U.S. and Canada, their population has dwindled to around 370 individuals. The whales are frequent victims of vessel strikes and entanglements in lobster gear.

To protect them, governments have established no-fishing areas and vessel slowdown zones. But for these measures to be effective, officials need to know where the whales are. Both the Canadian and U.S. governments have invested millions in satellite tracking programs. AI can process the vast amounts of satellite data much faster than humans, but it needs a large library of reference images to work correctly.

"A big problem … is the lack of good-quality training data and, more importantly, the lack of good quantities of training data for such rare species,” said Sun.

Creating Whales From Scratch with AI

To overcome the data shortage, researchers often alter existing images by cropping or recoloring them, but this has limited effectiveness. A whale can look dramatically different depending on its age, location, and behavior. A human can easily identify the animal in various contexts, but an AI needs to be trained on as many variations as possible.

Johnston and Sun used a generative AI technology called a diffusion model. These models are trained on billions of image-text pairs to learn how to create a picture from a description. The researchers fine-tuned a pre-trained model with a small set of real right whale images. This allowed the system to generate over a thousand new, synthetic images of whales in different positions and settings.

While many of the generated images were highly realistic, the AI also produced some strange results, a common issue with these tools. For instance, one of the synthetic whales had two tails and no head. Despite these quirks, the researchers found that their models successfully fooled Google's reverse image search, indicating a high degree of realism. The team is now preparing a paper on their findings, which could be applied to other endangered species.

Beyond Whales AI in Broader Conservation

The work at Duke is not an isolated case. Other scientists are also using AI to augment biodiversity data.

Experts at the University of Moncton in Canada developed an AI tool named ECOGEN that generates realistic bird sounds. These synthetic songs can supplement audio libraries for rare species, improving the accuracy of AI identification tools like BirdNET and Merlin. A 2023 study found that adding artificial sounds improved bird-song classification by an average of 12 percent.

However, some researchers remain cautious. Irina Tolkova, a postdoctoral fellow at Cornell University, expressed hesitation about trusting black-box models trained on small datasets of endangered species sounds.

In another application, computer scientists at Carnegie Mellon University are helping ranchers in Montana combat leafy spurge, an invasive weed that is becoming more common with climate change. By creating synthetic images of the weed in various weather conditions, they significantly improved an AI system's ability to detect it with drones.

The Environmental Catch-22 of AI

Artificial intelligence is a double-edged sword for environmental research. The technology itself contributes to the problems ecologists are trying to solve by consuming vast amounts of energy and water. A 2024 study found that generating a single AI image can use as much energy as fully charging a smartphone.

When this energy comes from fossil fuels, it exacerbates global warming—a major threat to wildlife. For right whales, warming waters are reducing the availability of their primary food source, forcing them into new areas with fewer protections. Furthermore, AI data centers require immense quantities of water for cooling, raising concerns about their impact on local ecosystems.

This creates a difficult dilemma for researchers. "As a young person who really wants to feel like I’m making a positive contribution, it becomes difficult to grapple with a lot of these sort of cost-benefit-analysis kind of decisions," said Sun.

Kasim Rafiq, a wildlife biologist at the University of Washington, echoed this sentiment. "You’ve got all these benefits, but then you also have the huge energy and data infrastructure costs... which puts you in this catch-22 situation." One solution, he suggests, is to develop smaller, more energy-efficient models.

Navigating the Future of AI in Ecology

The potential for misuse is another concern. In 2023, a fossil-fuel-funded think tank reportedly used a fake, AI-generated image of a dead whale on a beach to spread misinformation and discredit offshore wind energy, a claim scientists say has no supporting evidence.

Despite these downsides, many scientists believe AI has the potential to "revolutionize conservation," offering new ways to monitor the wildlife trade and project the impacts of climate change.

Cornell's Tolkova, while having reservations, remains open-minded. "I like to hope that there’s a lot that machine learning can do that isn’t just useful for creating chatbots, but also for meaningful conservation impact," she said. "I think it’s worth exploring those opportunities."