PAWS anti-poaching AI predicts where illegal hunters will show up next


The illegal animal trade is a global scourge but a lucrative one, worth $8 to 10 billion annually, according to — trailing only human, drug and weapons trafficking in value. With so much money to be made, conservationists and wildlife rangers face overwhelming odds against well-organized poaching operations fueled by incessant demand for illicit animal products. The results of this protracted conflict have been nothing short of devastating for the species caught in the middle.

At the start of the 20th century, more than 100,000 tigers are estimated to have roamed throughout Southeast Asia. Today, due to a combination of habitat loss and aggressive poaching, fewer than 4,000 currently remain in the wild. On the black market, products made from a single tiger can fetch as much as $50,000. Rhino populations have been similarly decimated, dropping from around 500,000 individuals at the start of the 20th century to only 50,000 today. Overall, that mammal, bird, fish and reptile populations have declined by 60 percent since 1970.

“Poaching is the illegal hunting, capturing or killing of wild animals and it’s done for a number of reasons,” Erwin Gianchandani, Senior Advisor in the Office of the Director of the National Science Foundation, said during a panel discussion at SXSW 2021 on Tuesday. “Some folks poach because they want to be able to claim the land that the animals reside on for human use. In other cases, folks poach because they want to be able to retrieve and use rare animal products, things like ivory or fur, even organs and skin. They often do this because they feel that these products can have religious, medicinal, nutritional, or financial value.”

“It’s not just that the poachers go after the animals,” he added, “but they are often so motivated, that they will actually end up harming or even killing wildlife rangers to elude the detection or capture of their poaching.” Just this January, in the line of duty while on patrol at the Virunga National Park in Congo, home to a third of the world’s mountain gorilla population.

Though wildlife rangers are often outnumbered and stretched thin in their attempts to patrol broad swaths of nature preserves, AI and machine learning systems are poised to drastically improve rangers’ effectiveness by helping them not only track where poachers have been but also predict where they are most likely to turn up.

PAWS (Protection Assistant for Wildlife Security) is one such system. Professor Milind Tambe, co-founder of the USC Center for Artificial Intelligence in Society (CAIS) and Director of the Center for Research on Computation & Society at Harvard University, oversaw its development after attending a Global Tiger Initiative conference in 2013.

“I found out just how stark things were for animals that I’d read bedtime stories to my kids about,” he told .

PAWS leverages poaching data from the open-source SMART (Spatial Monitoring and Reporting Tool) system developed by the World Wildlife Foundation and uses security games — a subset of game theory where the player must optimize limited resources to mitigate threats and attacks — to suggest the most efficient routes for rangers to take given that historical data.

Tambe and his team first trialled PAWS in 2014 at Queen Elizabeth National Park in Uganda. The park is home to a variety of endangered species, as well as thousands of traps and snares set by poachers. What’s more, the 2000 square-kilometer park only has around 100 rangers to patrol it. PAWS works by first splitting the park’s area into individual 1km squares then assigns a risk factor to each square based on where snares had been previously discovered — a decade’s worth of that data collected through SMART. It then suggests patrol routes through the highest risk areas. These suggestions change over time as poachers adapt to the rangers’ actions. The time of year; trail, river and road locations; weather and topographic conditions also factor heavily into the PAWS estimate. Over a six-month test period, QNEP rangers were blindly assigned a mix of patrol routes through high- and low-risk areas of the park.

“What we found was, where PAWS made predictions of higher risk, indeed more snares were found,” Tambe said. “Where paws made predictions [for lower risk] rangers found less number of snares.”

But like all machine learning systems, PAWS is constrained by the quality of data that it ingests. “The data that park rangers are collecting is not perfect and there is some uncertainty with the data,” Shahrzad Gholami, a data scientist at Microsoft, said during the SXSW panel. “So the locations that they visit, they may not find any activity, any poaching signs, but it doesn’t mean that poaching activity did not exist. It might be because the snares were well hidden.” Even when rangers find a snare, they can only glean so much information from it. They cannot, for example, know if the trap was set recently or had been sitting undisturbed for weeks or even months before being discovered. What’s more, PAWS can only address the specific act of poaching, not a poacher’s underlying motivations for doing so.

Buoyed by their success at QNEP, Tambe’s team partnered with the WWF in 2018 to bring PAWS to conservation areas managed by the wildlife organization, such as the Sepak Wildlife Sanctuary in Cambodia. Located along the country’s eastern border with Vietnam, Sepak is home to a sizable Asian elephant population as well as bongos, antelope, deer, macaques and leopards. Tigers used to roam the region as well, though none have been seen since 2007 and are thought to be locally extinct. The WWF plans to reintroduce the species beginning in 2022 and has identified Sepak sanctuary as an ideal site to do so. But first, they need to get a handle on poaching activity in the region which threatens both the tigers themselves and their prey.

Like QNEP, Sepak covers a massive area, roughly 1400 square kilometers, but has just 72 rangers to patrol it. The Cambodian wildlife refuge also offered a number of unique challenges in training the PAWS AI not encountered in the Uganda test — such as monsoons. The PAWS team worked closely with conservationists at Sepak to develop an effective model and that collaboration led to some surprising discoveries.

“For example, it helped us discover that in addition to just modeling distance from roads,” Gholami said. “We should actually specifically model the distance from one particular road called route 76 which was a major highway through the park.” The team also found that poaching practices varied depending on the poacher’s country of origin. That is, poachers crossing the border from Vietnam behaved and reacted differently than local the Cambodian poachers. The time of year also proved to be an important factor as poachers would dramatically alter their snare placements and distribution during monsoon season compared to the dry months.

Still, the PAWS system proved to be highly effective. “They found five times more snares in the one month that the field test was going on compared to any other month on average in 2018, Gholami explained.”

These are promising improvements, but wildlife conservationists still face an uphill battle against poaching. “Conservation biologists have estimated that rangers are only effective at removing about 10 percent of all snares in these parks,” said Lily Xu, a Harvard PhD candidate involved in the PAWS project. “One of the most effective mechanisms for preventing poaching and other conservation crimes are through deterrence, so when rangers conduct patrols in certain areas, they dissuade poachers from coming back.” However, poachers pushed out of one area of a nature preserve will often simply move their operations to a neighboring area, endangering the wildlife there instead.

Despite the challenges, Tambe’s team remains undaunted. Through partnerships with the WWF and other conservation organizations, Tambe hopes to implement PAWS in as many as 600 protected areas worldwide as well as expand its scope to protect marine sanctuaries and forests in the near future.

“This is bringing up newer kinds of challenges that may not arise in other domains where AI is active,” Tambe told the . “What lessons we’ve learned would be valuable for many applications; they wouldn’t be confined to wildlife crime. There are all sorts of challenges in applying AI for society and social good, and the benefits would spread across to other areas.”



Source link

Share

Written by bourbiza

bourbiza is an entertainment reporter for iltuoiphone News and is based in Los Angeles.

Comments

Leave a Reply

Loading…

0