Understanding Elephants through AI

Helping humans better understand how elephants communicate could improve the animals’ protection. A joint research project of the St. Pölten University of Applied Sciences and the Austrian Academy of Sciences uses artificial intelligence (AI) to analyse the animals’ sounds and interpret their meaning.

Elephants are social beings who live in varying habitats of different sizes. The animals’ calls are an important means of communication with their conspecifics. Research shows that elephants convey important information with their calls. However, in the past, the large variety of calls has made it hard for researchers to decode the animals’ communication patterns.
 
So far, there is only very little research into elephant communications. “The animals use a broad range of sounds such as the well-known trumpet. The most frequently used sound type is the low-frequency rumble which has a basic frequency that reaches infrasound”, says Angela Stöger, project lead and head of the field of mammal communication at the Austrian Academy of Science.

Angela Stöger, Credit: Angela Stöger

Using AI to Understand Elephants

“In our project, we seek to employ artificial intelligence in detecting acoustic signatures in elephant communication that give us valuable insights into specific behaviour. In this process, we adopt a completely novel approach to bioacoustics: based on hypotheses about the meaning of sounds that are produced through AI-based data analyses, we synthesize elephant calls designed to carry different meanings for the animals. We play these sounds in playback experiments to elephants in the wild. By monitoring the elephants’ reactions, we can put our hypotheses to a test. If this approach proves to be successful, we could open up entirely new possibilities for data-driven analyses of animal communication”, says Matthias Zeppelzauer, head of the research group Media Computing at the St. Pölten UAS’ Institute for Creative\Media/Technologies.
 
The project examines the communication of African Savanna elephants by using the largest existing databases on elephant sounds ever to be recorded in the wild. It is the first time that AI is used to systematically analyse these data. This allows for an in-depth analysis which could not be achieved manually.

Understanding Animal Behaviour 

Research on animal sounds provides valuable insights into the perception and the behaviour of animals and the population dynamics. This helps to assess biological diversity. The application of AI-based methods to analyse animal communication is currently becoming ever more popular. To learn from the complex data, researchers use deep learning methods, in other words, complex neural networks. So far, these methods were mainly implemented to analyse the communication of sperm whales and zebra finches.
 
“We do not only use AI to identify an animal’s age, sex, and social status and a group’s dialect, we also use AI to determine new patterns in acoustic data and encode information. This will help us understand how elephants communicate meaning and content”, says Zeppelzauer.

Links:

• Elephant sounds on Angela Stöger’s website
• Recordings of elephant sounds from an older project with acoustic camera “displaying“ the sound (under Supporting material)