Machine Learning vs Poachers

By using machine learning rangers of Uganda Wildlife Authority try to predict the next attack of poachers

Is wildlife conservation important?

The wildlife of our planet is endangered. But why do we care? The preservation of wildlife costs us millions of dollars, while there are many other problems of humankind that require immediate support such as fighting hunger or poverty.

One reason is that we want to: “many of us love the natural world. We think animals are cute, majestic, or just plain fascinating”[1].

Figure 1: Young elephant is playing

The wildlife is something we inherited and responsible for. Moreover, the decrease in the variety of spices has a direct negative impact on natural ecosystems and local economies[2].

Poaching is one of the most significant threats to wildlife. “Every day in Africa nearly 100 elephants are killed for their ivory”[3]. The global tiger population has dropped over 95% from the start of the 1900s and has resulted in three out of nine species extinctions[4].

Exhibit 1: Wildlife population is in decline[5]

Poaching also provides significant profits to terrorist organizations, “which are which are attracted to wildlife trafficking because of its low risk of detection, high profits, and weak penalties”[6].


Predicting poaching behavior

Fighting against poaching is not easy. A population of animals is spread across huge area making patrolling very difficult. For example, Uganda Queen Elizabeth National Park, a home of 5,000 elephants, is spread across 763 sq miles[7]. To protect animals on such vast territory “Uganda Wildlife Authority uses 50 to 90 percent of its budget on ranger patrols”[8].

In 2013 researchers from the University of South California (USC) in collaboration with Uganda Wildlife Authority (UWA) developed PAWS (Protection Assistant for Wildlife Security), an AI algorithm which analyses historical information on previous poaching activities and designs the most effective routes for patrolling by trying to predict where the next poaching activity will happen. The AI algorithm every time develops new routes for patrolling units to prevent poachers from predicting patrol’s activity[9].

Exhibit 2: PAWS Overview[10]

As in other AI platforms, improvement of the PAWS machine learnings algorithms relies on large data available to analyze. Before the launch of the program, the wildlife data from UWA collected by rangers over 12 years from 2003 to 2014 were analyzed[2]. When patrollers execute the patrol strategy generated by PAWS over a period, more information is collected and become part of the input in the next round[10]. After an initial launch in Uganda, in 2014 PAWS was rolled out in Malaysia to further increase the amount of data available for analysis and test adaptability of PAWS in other geographical zones[11].

Currently, the project team is working on:

  • Incorporation of complex topographic features in the algorithm to make route planning more effective.
  • Geographic expansion. The PAWS team considers rolling out the project in the Northen China and Cambodia.
  • Integration of the project with other tech wildlife conservation initiatives. For example, PAWS is being integrated with Air Shepherd, a service which uses drones equipped with infrared cameras to search for poachers at night[11].

Larger PAWS needed

In my opinion short-term the project team needs to focus on improving the predictability of the model, which can be done by deriving larger sets of data. One way to do so is to share learnings and data with other wildlife conservation organizations, such as The World Wildlife Fund or The Nature Conservancy.

Additionally, short-term the team needs to make PAWS easily scalable. Since 2013 the project team has rolled out PAWS in two countries only: Uganda and Malaysia. Low speed of expansion significantly limits the positive impact on wildlife conservation that the project can deliver. To speed up the roll out the PAWS project team should learn how to quickly optimize the algorithm for different terrain conditions, as well as design easy-to-understand interface to make transferring PAWS management to local patrol teams as smooth as possible.

Medium-term the project team should increase awareness of the impact that technology can bring in wildlife conservation. The success of the PAWS project should be highly publicized and, as a result, motivate tech companies and research labs engage in other wildlife conservation initiatives.

Technology is becoming increasingly important to help to protect animals. However, poachers as well have access to technology. How can we be ahead of them? What other recent technology developments can we use to protect wildlife? How a person who cares can help animals survive?


(719 words)




[1] “What if the point of saving endangered spieces?“ BBC, July 2015.

[2] “CAPTURE: A New Predictive Anti-Poaching Tool for Wildlife Protection”. Thanh H. Nguyen, Arunesh Sinh. University of Southern California, USA, Wildlife Conservation Society, USA, Microsoft Research, Uganda Wildlife Authority, Uganda.

[3] “Wildlife poaching has a huge impact on Africa, but our leaders are silent”. The Guardian. December 2014.

[4] “PAWS: protection assistant for wildlife security.” University of South California.

[5] Air Shepherd website.

[6] “Poaching and terrorism. A national security challenge”. Committee on Foreign Affairs – House of Representatives. April 2015.

[7] The Wildlife Conservation Society official website.

[8] “Rangers Use Artificial Intelligence to Fight Poachers.” The National Geographic. June 2016.

[9] “Rangers Use Artificial Intelligence to Fight Poachers.” The National Geographic. June 2016.

[10] “Deploying PAWS: Field Optimization of the Protection Assistant for Wildlife Security.”  Fei Fang, Thanh H. Nguyen, Rob Pickles., University of Southern California, USA Panthera, USA, Rimba, Malaysia, Nanyang Technological University, Singapore, Columbia University, Kenyir Research Institute, Universiti Malaysia Terengganu, The Netherlands Institute for the Study of Crime and Law Enforcement (NSCR).

[11] “This AI Hunts Poachers”. IEEE Spectrum. January 2018.


Open innovation in the Public Sector: The Argentinean case


Streamlining Humanitarian Aid with Additive Manufacturing at Oxfam

Student comments on Machine Learning vs Poachers

  1. PAWS is truly an inspiring mission and one that needs to be implemented universally, however, there are a few points that I would like to question. As we saw in the IBM Watson case, the quantity and quality of the data inputs that are “fed” into machine learning algorithms are extremely important for the predictive analysis to be accurate. How can we trust that the historical data will be able to correctly asses the way poachers and terrorists may adjust their poaching methods once they realize that their routes are being intercepted? Has PAWS actually been effective since its origination in 2013? Is there data to support that the machine algorithm is actually working?

    Additionally, I agree with the content of the short-term and medium-term strategies, but I believe they both need to be implemented in the short term for the product to be successful. This is a race against time. If the awareness of the product is not recognized globally, we may unfortunately see a number of species become extinct.

  2. As someone who deeply cares about our nature and environment, I find PAWS’ mission extremely important and inspiring. In addition to the motivation you are describing in your essay, i.e. our responsibility to maintain and respect nature and the animals within it, another crucial aspect to think about are the many human lives taken due to poaching. Last year, more than 100 rangers died trying to protect the animal wildlife in Asia and Africa. Therefore, even for someone less motivated by the loss of environmental diversity, applying machine learning successfully to this cause also has an important positive impact on people’s lives.

    Building up on the previous comment, an interesting challenge I see in this context is to ensure high quality data input for the algorithm – not just due to regional differences or route changes of poachers, but also given the resource restrictions many of the affected areas are facing. When visiting South Africa and interacting with local rangers, I learnt that their access to digital tools to track and access data is often limited. Therefore, we need to think about solutions across the whole “supply chain” when aiming to apply machine successfully learning to PAWS cause.

  3. PAWS is a really interesting example of applying machine learning in an environment not normally thought primed for it, yet very much so in need of it. Particularly given the large swathes of land that Wildlife Services and anti-poaching officials cover, using this technology and process could make their jobs both easier and safer. What other types of data do you think would make this more scalable and applicable? Do they need more geomapping, or, on the darker side, more instances of poaching in order to predictively map their patterns? On a regional level how could this be scaled to allocate resources across countries?

  4. This is an interesting look at the use of innovative technology to help solve a broader societal issue. This article made me consider impact opportunities for machine learning and other new technologies beyond simply improving business outcomes. An additional technology that could play a pivotal role in wildlife protection is open innovation through social media and digital engagement. Using these tools, the Ugandan Wildlife Authority and other groups can tap into the ideas and support of a broad international population.

    These same technologies can be applied by other nonprofits or social groups. For example, the Red Cross utilizes open innovation to coordinate disaster response efforts. The next step in the dissemination of technologies such as machine learning, AI, open innovation, and additive manufacturing is to explore ways in which they can be used not only to improve business outcomes but also global societal development.

  5. I think this is a great application of machine learning! I think the species that PAWS could help preserve is more than enough reason to invest more into it. With this being a time-sensitive issue, I wonder how quick is the turnaround time for improvement iterations of the algorithm.

    Also, what preventative measures should conservationists put in place to make sure that poachers cannot reverse engineer this tool for their needs?

  6. I really liked this article. It shows that there are many applications of machine learning technology beyond those one would typically think of. I have one major concern with the article though, and that is whether it is effective or not. Before I would recommend rolling this out to other regions, I would want to understand exactly how much better this PAWS algorithm is than current methods for poaching, such as following large herds of elephants and protecting against poachers that way. My concern is that over-reliance on the new fancy technology, such as machine learning in this case, will distract from the more important mission of reducing poaching.

    Thanks for the great read!

  7. I liked this article and completely agree that AI can help to solve the issue of poachers. You mentioned that data volume is one of the key drivers to build good predictions of where poachers will be. In terms of getting more data, I see other alternatives. While the most evident one would be installing GPSs in some of the animals to track their movements and make sure that patrols are nearby, other startups are using alternative data sources, such as audio recordings ( ). I really believe that going beyond the use of drones can be very helpful to make the AI predictions more accurate.

  8. Really amazing what PAWS is doing for animal conservation – excited for the impact!

    I agree that larger sets of data provide more accurate patrol paths but collecting this data and truly understanding animal movements can be a time consuming process, especially in natural parks which have historically not had access to many resources and have previously not documented animal movements or have had consistent patrol efforts. Additionally, I recognize that there are only so many factors, which can be controlled. Has the company considered internal corruption and corporate governance structures, given a significant amount of conservation parks are in developing markets. The impact of a few bad actors could have severe consequences on not only the mission of animal conservation but also the data collection process, which could be applied broadly across all users of the product.

    In thinking through extreme scenarios: floods, earthquakes, storms, and other adverse weather events could significantly reduce the predictability of animal movements, how does this technology ensure that patrol paths would still be viable under these circumstances. This risk is elevated as global warming continues to be on the rise.

  9. What an inspiring use of machine learning technology! This article reminded me that technological advances like machine learning can have broader applications beyond increasing revenue and reducing costs for corporations and can be a force for social good. Perhaps PAWS could expand its data set by partnering with governments around the world, particularly those who are struggling with the poaching proble. Many governments already collect topographical and environmental data (including wildlife migration patters), regardless of whether they have a problem with poaching, as part of their weather, environmental conservation, and national security efforts. Could feeding this data into the algorithm, including data from those nations not plagued by poaching, could provide more “learning material” for the algorithm.

  10. What an interesting read – thanks for sharing this unique use of machine learning in protecting endangered wildlife species. I certainly agree that accumulating and disseminating larger sets of data will help wildlife conservation agencies stay ahead of the poachers. In fact, I think this is the main way they’ll be able to do so. While I know very little about poaching, my intuition is that that poachers operate rather independently and do not share many data-based learnings between one another, both within country and (even less likely) internationally. As a result, creating a centralized database of poacher behavior will afford animal protectors visibility into common trends across groups and identify best practices in stopping the spread of poaching. Furthermore, should a poaching group attempt to expand beyond their current territory, these wildlife agencies will already know their poaching patterns and should be able to anticipate their moves and counteract them, thus discouraging expansion and containing and localizing each group’s poaching activity. As an aggregated platform competing against (presumably) less organized and more silo-ed groups, PAWS will create great informational advantages that will benefit wildlife conservation efforts.

  11. I definitely love the idea of using AI to help stopping poaching, but the thing that I am concerned with is if we use the historical data to predict the future poaching, and if the poachers know that, they would always find ways to change routes to avoid to be predicted. With AI, can it provide real time prediction of the poaching? For example, can we tag each animals with a chip, so all of them are monitored. Once the system shows that there are other unusual moving items around the area, they can immediately send out the patrol to protect the animals.

Leave a comment