Once the subject of science fantasy novels, artificial intelligence (AI) is now a household name. AI is becoming increasingly common in our everyday lives, and it’s also being used in some life-changing applications.
In Australia alone, AI is making leaps and bounds in the realm of agriculture. Australian farming is rapidly shifting with the use of agbots and other tools. Despite a reduction in the population of farmers, these AI-driven agricultural robots are actually making farming easier and more efficient---and often more affordable.
AI is set to transform education, medicine, the job market and much more, in Australia and beyond.
What about wildlife?
In Australia, our wildlife is one of the things we prize most highly. We're known around the world for our amazing species---many of which are found no place else on Earth. It may come as no surprise then that the latest and greatest modern technology is being used to protect and preserve Australia's exquisite creatures.
There are many reasons that it makes sense to use technology in wildlife research. Traditional research methods are time-consuming and require often extensive manpower. These approaches are typically more costly and complex.
Revolutionary technology such as AI and drones change the game completely, allowing for more affordable solutions that are extremely fast and efficient. In the world of wildlife, AI can be used in counting and monitoring wildlife in a streamlined fashion that also brings with it incredible accuracy. For animals that are difficult to track this is even more important as the technology removes the potential disruptions that human interaction might create.
The potential outcomes of developing and utilising this technology are boundless. AI, drones, and machine learning could enable us to remotely survey animal populations that are in danger with far greater ease than ever before. Perhaps eventually drones could serve to track over a long distance, following wildlife populations on lengthy migrations or over a massive area. The technology could even pave the way for closer, more intimate observation of some species, allowing us insights into their behaviour in ways we’ve never accessed before. And all of this knowledge can be put to great use in efforts of conservation and preservation.
How Does it Work?
At the crux of artificial intelligence, wildlife technology is the machine learning that allows the technology to categorise animals via images or videos and identify them. A software program created by a nonprofit in Oregon, USA provides one such example of this technology. The software is called WildBook and it identifies animals using their unique coat patterns or other distinctive physical features. The program has already helped organisations such as the Giraffe Conservation Foundation to swiftly assess the giraffe population in part of northern Kenya. Across Africa, populations of reticulated giraffes have declined by as much as 40% over the past 3 decades. Traditional methods of monitoring giraffe populations are costly and time-consuming. Using software like WildBook offers a promising solution.
The various AI-driven surveying methods which are emerging are incredibly fast and accurate. Ali Swanson, a researcher from the University of Oxford estimates “that the deep learning technology pipeline...would save more than eight years of human labelling effort for each additional three million images. That is a lot of valuable volunteer time that can be redeployed to help other projects.” By creating opportunities for AI to take over some of the observational research, human energy is freed up for other efforts---efforts which may require more human-based influence.
Swanson was part of a team that produced an academic paper released last year. The paper describes in detail the cutting-edge methods of AI that have begun emerging with deep learning and how they can be of use in animal research.
Here’s how it works: In the field, researchers place motion sensor cameras in the animal’s environment. These unobtrusively placed cameras capture images of the animals in their natural habitat, over and over. The number of these captured images reaches into the millions, and the images are then converted into readable data for the AI system. The AI learns how to spot species, number of animals present, and can even identify a juvenile animal vs an adult. Once these images are labelled, machine learning takes place and new images can be constantly compared.
Each new image will then be automatically described by deep neural networks (artificial intelligence systems that require intense and repeated training data), contributing to the growing body of knowledge. The AI is able to identify the content of up to 99.3% of images correctly. If you're wondering if it's as accurate as humans, the system shows a 96.6% accuracy rate when compared with human volunteers. It’s definitely a reliable approach and it’s only getting better.
Jeff Clune, leader of the research team said, “This technology lets us accurately, unobtrusively and inexpensively collect wildlife data, which could help catalyze the transformation of many fields of ecology, wildlife biology, zoology, conservation biology and animal behaviour into 'big data' sciences. This will dramatically improve our ability to both study and conserve wildlife and precious ecosystems.”
Other AI Methods for Wildlife
It isn’t just traditional image-based technology that’s proving successful. Thermal imaging is one of the latest to emerge as a beneficial wildlife tracking method. A research team from Queensland University of Technology (QUT) has developed an approach that utilises drones, infrared imaging, and machine learning. Focusing on monitoring the declining koala population in Australia, this method has been shown to be less invasive to the animal population and increasingly reliable as well.
The accuracy of this method was tested by calculating error rates of automation vs. human ability. The study used a larger, statistically valid number of koalas and tracked how often a person (or an algorithm) correctly identified a koala. The results were promising. Dr. Grant Hamilton, co-author of the study said, “On average, an expert koala spotter is going to get about 70 percent of koalas in a particular area. We, on average, get around 86 percent. That’s a substantial increase in accuracy that we need to help protect threatened species.”
This may be especially impressive in that the tracking is effective even amidst the thick canopy of the Eucalyptus trees, where the koalas are most often found. Previous wildlife tracking attempts have often been a little more straightforward (monitoring seals on an open beach, for instance).
Drones circulating over the area moved in a “lawnmower” pattern as they tracked the koalas. Using infrared imaging, the heat signals of the koala could be observed. Compared with their GPS-connected tags, the researchers could confirm that the heat readings were in line with actual koala locations.
As this new AI-driven approach is just developing, there are likely to be major strides in the use of drones and thermal imaging in the world of wildlife conservation.
This is good news for Australian wildlife in general, but especially hopeful for koalas, who have seen declining numbers in the past two decades. Estimates by the Australian Koala Foundation suggest that there could be as few as 43,000 koalas left in the wild.
AI and Other Australian Wildlife
Koalas aren’t the only Australian animals who are being helped by AI. Researchers are experimenting with the use of drones for tracking dugongs as well. Dr. Amanda Hodgson, a marine mammal researcher from WA’s Murdoch University has been focusing on this study for several years.
Dugong spotting usually is done via plane, a method that is far more costly, time-consuming, and environmentally challenging. Using drones, the methodology could be dramatically changed.
Hodgson along with QUT’s Dr. Frederic Maire developed what they call a Dugong Detector, a tool that uses AI to recognise dugongs in images that have been captured via air by drones. Hodgson said that the results have already been positive. “Using Google’s TensorFlow machine learning platform, the Detector identified 70 per cent of the dugongs that could be found manually in 37,000 images from the Australian coast, reducing our manual review hours by 95 per cent.”
Not only can this technology be used to monitor dugongs, but it can be used to observe the health of the seagrass in the environment as well. This could prove an advantageous environmental tool for habitats not just in Australia but across the globe.
Drones are being used to observe another ocean animal: sharks. A system called SharkSpotter has been developed, combining AI and machine learning to identify sharks in the water. The goal is to spot sharks in areas where people are swimming, creating swift and early alerts to keep both swimmers---and sharks---safe.
SharkSpotter has already received praise; it was named the national AI or Machine Learning Innovation of the Year at the Australian Information Industry Association (AIIA) annual iAwards in 2018.
Over the next few years, it should be fascinating to watch as the world of AI expands in more ways, becoming a part of many facets of Australian life.