Artificial Intelligence and Big Data Could Help Save Wildlife
A group of experts in artificial intelligence and animal ecology has proposed a new interdisciplinary approach designed to improve research on wild animal species and make more effective use of the vast amounts of data currently being collected thanks to new technologies.
Animal ecology has entered the era of big data and the Internet of Things. Today, an unprecedented volume of data on wild animal populations is being collected thanks to advanced technologies such as satellites, drones and ground-based devices (automatic cameras and sensors attached to animals or placed in their surroundings).
This data has become so easy to obtain and share that it has reduced the distance and time for researchers, minimizing human presence in natural habitats. Today, many AI programs are available for analyzing large datasets, but they are often general-purpose and poorly suited to accurately monitoring the behavior and appearance of wild animals.
A group of scientists from EPFL (the Swiss Federal Institute of Technology in Lausanne) and other universities has outlined a pioneering approach to this problem and to developing more accurate models, combining advances in computer vision with the expertise of ecologists. Their findings, published in Nature Communications, open new prospects for using AI to protect wildlife species.
Building Interdisciplinary Know-How
Wildlife research has grown from local to global. Modern technologies now offer revolutionary new ways to obtain more accurate estimates of wild animal populations, gain a better understanding of animal behavior, fight poaching and halt the decline of biodiversity.
Ecologists can use AI, more specifically computer vision, to extract key features from images, videos and other visual data in order to quickly classify wildlife species, count individual animals and gather specific information from large datasets.
Common programs currently used to process such data often operate as black boxes and do not make use of the full body of existing knowledge about the animal world. Moreover, they are difficult to configure, sometimes suffer from poor quality control, and may raise ethical concerns related to the use of sensitive data.
They also contain a number of biases, especially regional ones: for example, if all the data used to train a given program was collected in Europe, the program may not be suitable for other regions of the world.
“We wanted to interest more researchers in this topic and bring their efforts together to move forward in this new field. AI can serve as a key catalyst in wildlife research and environmental protection more broadly”
If computer scientists want to reduce the bias of an AI program trained, for example, to recognize a particular species, they need to be able to draw on the knowledge of animal ecologists. These experts can point out which characteristics should be taken into account in the program, such as whether a species can survive at a given latitude, whether this is critical for the survival of another species (for example, through predator-prey relationships), or whether the physiology of the species changes over its lifetime.
New machine learning algorithms can be used to automatically identify an animal. For example, the unique stripe pattern of a zebra or the dynamics of its movement in video can serve as signs of individual identity.
“This is where the fusion of ecology and machine learning is key: a field biologist has deep domain knowledge about the animals being studied, and our work as machine learning researchers is to work with them to create tools to find solutions”
Informing About Existing Initiatives
The idea of strengthening ties between computer vision and ecology emerged when Tuia, Mathis and other scientists discussed their research challenges at various conferences over the past two years. They saw that such collaboration could be extremely useful in preventing the extinction of some wildlife species.
Several initiatives are already underway in this direction, some of which are listed in the Nature Communications article. For example, Tuia and his team at EPFL have developed a program that can recognize animal species from drone images. It was recently tested on a seal population.
Meanwhile, Mathis and her colleagues have released an open-source software package called DeepLabCut, which allows scientists to estimate and track animal poses with remarkable accuracy. It has already been downloaded 300,000 times.
DeepLabCut was designed for laboratory animals, but it can also be used for other species. Researchers at other universities have also developed programs, but they find it difficult to share their discoveries because a true community has not yet formed in this field. Other scientists often do not know these programs exist or which one is best suited to their specific research.
Nevertheless, the first steps toward such a community have been taken through various online forums. However, the Nature Communications article is aimed at a broader audience of researchers from around the world.
“The community is steadily taking shape. So far, we have relied on word of mouth to build the initial network. We first started two years ago with the people who are now the other lead authors of the paper: Benjamin Kellenberger from EPFL, Sara Beery from the California Institute of Technology in the US, and Blair Costello from the Max Planck Institute in Germany”
