Wildlife crime is costly, not only for ecosystem conservation but also for local and national economies. The World Wildlife Fund defines wildlife crime as "any breach of national, regional, or international legislation that protects wildlife species." This includes crimes like poaching, but also illegal smuggling, poisoning, or killing of wildlife, as well as the unauthorized alteration or destruction of habitats. A report by the World Bank in 2019 estimated illegal logging, fishing, and the wildlife trade have a combined cost of one trillion dollars when accounting for ecosystem services and cost source countries seven to twelve billion in potential lost revenue lost annually. Traditional methods of combating wildlife crime include regular patrols and camera traps, but these are also costly in manpower and resources.

Yet new tools using artificial intelligence (AI) are advancing the fight against wildlife crime. AI and machine learning (ML) tools are already an active, impactful, and globe-spanning tool in the fight against wildlife crime. Applications of ML have become a powerful tool for detection and identification both of the wildlife they seek to protect and of crimes being committed. Following three case examples from around the world, one commonality is that these tools are particularly powerful in conjunction with traditional methods and human action.

Brazilian Habitat Protection

In the remote rainforest of northeastern Brazil, the Tembé people have a problem: how to shield their 300 square kilometer territory from illegal loggers and poachers. A few years ago, the only practical solution was regular patrols. However, finding criminals using sight and sound in such a large, dense, and loud area proved ineffective. The cacophony and cover of the rainforest mask evidence of human presence. Patrols were also dangerous; with little to no intelligence on the location and armaments of potential criminals, tribe members put their lives at risk for only a small chance at stopping wildlife crime. 

One nonprofit, Rainforest Connection (RFCx), partnered with the Tembé to develop small microphones that can be placed high in tree canopies. According to the  2023 RFCx white paper “Harnessing the Power of Sound and AI to Track Global Biodiversity Framework (GBF) Targets”, it sends real-time alerts to rangers who investigate the area of disturbance. The microphones are low-cost and low-maintenance, powered by solar panels and connected to the internet through satellites. Alongside the microphones, RFCx developed software that uses ML to recognize noises like gunshots and chainsaws through the intense background sounds of the rainforest. The AI can detect illegal activity imperceptible even to a trained human ear. 

The technology boasts a strong record. In 2020, an impact assessment by JG Global Advisory for a grant from USAID found the program had a false positive rate of .337%. The use of real-time AI reporting has led to more strategic reporting, more accurate reporting, and more information sharing between RFCx, the Tembé, and the government. Since 2015, alongside their partnership with RFCx, the Tembé have retaken nearly 15% of their total land area, previously occupied and controlled by illegal loggers and settlers. 

AI Cameras for Indian Tigers

The Dudhwa Tiger Reserve in northern India is 20 kilometers from the border with Nepal and covered in lush meadows and dense forests. Outside its borders, human settlement is dense. Forest rangers have trouble preventing tiger-human conflict across the reserve's difficult-to-manage 1,284 km2, an area where poachers shoot and poison tigers to sell their skins, bones, or meat for profit. Camera traps are often the go-to resource for such a problem, but they have two main faults: an SD card can only be collected long after a wildlife crime has been committed and they often are falsely triggered. For instance, a journal article by Dertien et al. explains that one SD card could have thousands of false trigger photos for every 300 usable images. 

Dudhwa Tiger Reserve has used Nightjar’s TrailGuard technology to solve this problem: a camera the size of a large Sharpie marker that uses computer vision, a type of AI, to identify animals, humans, and vehicles. The camera’s ML model is trained using three-dimensional species renderings, and validated with camera trap photos. When it detects a person where one should not be, TrailGuard sends a notification to forest rangers within 30-42 seconds after detection. During testing, captured images of people carrying guns and knives were used to identify and investigate an alleged wildlife crime. According to Dertien et al., in the future, the AI could be used to warn people of tigers straying close to human settlements, in an effort to prevent retaliatory killings.

Identifying Illegal Fishing in the Congo from Space

Conkouati-Douli National Park in the Republic of the Congo boasts an enormous marine protected area (MPA) dealing with illegal, unreported, and unregulated (IUU) fishing. Park management relies on a program called Skylight by the Allen Institute for AI to patrol their MPA. Usually, relying on satellites to identify criminal acts on boats is difficult. How do you know if one boat is illegally fishing, sightseeing, or just passing through among complex marine traffic?

To solve this issue, Skylight analyzes satellite imagery with advanced computer vision models to identify anomalous boat behaviors common among illegal fishers, such as suspicious rendezvous in the area. Headquarters can then dispatch interceptors to catch unsuspecting wildlife criminals in the act. In one case, Skylight’s technology led to the apprehension of illegal, unreported, and unregulated shark fishers operating at night—something neither land radar nor officers in boats could have otherwise detected.

However, implementing AI technology poses additional risks to vulnerable rangers. Many rangers in less industrialized countries, like those evaluated in Srepok, Cambodia during a test of the AI system Protection Assistant for Wildlife Security (PAWS), according to the Harvard Business Review, are deeply familiar with their local area and highly motivated, but also poorly trained, equipped, and paid. If AI technology significantly increases confrontation with wildlife criminals, as it has in these examples, then there is a demonstrated concern that a higher number of less equipped, less experienced park rangers will be harmed in the process. A higher casualty rate could reduce ranger willingness to use AI or dissuade recruits in an already deadly occupation. At the same time, more confrontations, and therefore disruption of wildlife crime, have marked impacts on biodiversity conservation and ecosystem services.

Conclusion

Today, the use of AI in fighting wildlife crime protects the habitats and lives of many important species. If AI in this field is dismissed entirely by wildlife services, many advantages stand to be lost. As this case study has demonstrated, a variety of tools exist to fight a variety of crimes in a variety of circumstances. 

Still, there are valid concerns with some forms of wildlife crime-fighting AI. For example, a concern is the implementation of AI as a 'quick fix' solution at the expense of people. Resources need to be allocated not just to buying the technology itself, but also working with and training the local volunteers or law enforcement in their use. As much as AI tools can help, these tools are only as good as the people that they support. To be effective AI tools must be supplemented by well-equipped, well-trained, motivated, and uncorrupt rangers.

With greater attention, and with more technological breakthroughs likely, the future looks a little brighter for our most critically endangered wildlife thanks to specific AI technology.