The Better to Hear You With

As apex predators and a keystone species, wolves are the only thing standing between a healthy ecosystem and catastrophic disruptions in many regions. But the presence of wolves is not welcomed by all. In addition to keeping the populations of other animals in check, they can also cause damage in nearby communities where they are known to kill livestock and occasionally even harm humans. Can’t we all just get along?

Left to our own devices, probably not. With the help of some cutting-edge technology, however, our complex interrelationship is looking a whole lot brighter. Grizzly Systems, a company based in Montana, has developed a tiny artificial intelligence (AI)-powered device called a GrizCam. These GrizCams, roughly the size of a cell phone, can be deployed in the wilderness where they can monitor the locations of animal populations of interest — such as wolves — to assist in shaping conservation practices, and also to serve as a deterrent when a wild animal strays too close to humans.

Each monitoring device runs lightweight AI algorithms onboard to identify the sound of a target species, like the howl of a wolf. When a sound passes this initial filtering step the GrizCam begins recording audio onto an SD card. These memory cards are periodically collected by conservationists who load the data onto a more powerful computing system for further analysis.

This computing system runs a convolutional neural network that was built using the PyTorch framework. Training was carried out in the Microsoft Azure cloud using NVIDIA A100 Tensor Core GPUs. When new samples are analyzed by this model, it can not only identify the species that are represented in them, but they can also be precisely located, even from several miles away.

Using this information, researchers can develop plans to keep animals where they belong, without risking unwanted ― and potentially dangerous — interactions with humans. And when plans fail and wolves wander too far from home anyway, the GrizCams can play recordings of dogs barking, gunshots, or other sounds that will send a wolf racing to get back home without putting anyone, or anything, in harm's way.

At present, GrizCams have been deployed around Yellowstone National Park and in the immediate vicinity. The hope is that the information they collect will help scientists to better understand the behavior of wolves and protect the people that live nearby. The team at Grizzly Systems is working to make this dream a reality sooner by continuing to innovate and add new capabilities to the monitors, such as the ability to identify individual wolves by their vocalizations.

Once the GrizCams are ready for production use, the team plans to roll them out for some additional use cases as well. One day they may also monitor remote oil and gas rigs or electrical transformers, which are too often the targets of vandals. GrizCams may even find a role in curbing poaching and illegal wildlife trafficking operations.