Conserv-AI-tion: How Artificial Intelligence Will Protect Wildlife

Image from https://www.iucn.org/story/202307/computer-conservation

In the past, Artificial Intelligence was heralded as the downfall of humanity (see Terminator’s Skynet or 2001: A Space Odyessy’s HAL 9000). Fast forward to current times, AI has been ‘domesticated’ in our everyday lives, from tools like ChatGPT or artistic time-wasters like Midjourney. But with strides in machine learning pushing the boundary of what’s possible with AI, wildlife conservationists are progressively embracing artificial intelligence as a cutting-edge technological solution to address the biodiversity crisis and alleviate the impact of climate change. Applied properly, scientists hope AI can make faster and more accurate calculations, helping the limited manpower of conservationists to maximise their efforts. The ways in which AI can assist conservation efforts can be broadly sorted into three categories.

A graphic depicting the three broad categories of AI use in wildlife conservation, and examples of questions each category could address.

Predictive models- “what is this?”

This is the most straightforward use of AI. Rather than using poorly trained undergraduates for grunt work, machine learning systems use samples and historical data to learn and adapt, creating a model to accurately predict labels of unseen data. Essentially, this does the job of many people while getting faster and more accurate results- for a fraction of the cost! Another advantage of these predictive models is that they operate regardless of conditions, leaving habitats where they operate relatively undisturbed. 

Predictive models have been used in the Congo basin to assess biodiversity. Home to the world’s second-largest rainforest, the challenge of preserving species on the brink of extinction is substantial, to say the least. In 2020, Appsilon, a data science company, collaborated with the University of Stirling and Gabon’s national parks agency to develop the Mbaza AI image classification algorithm (1). The goal? Large-scale biodiversity monitoring in Gabon’s Lopé and Waka national parks.

Automated cameras capture images of elephants, gorillas, chimpanzees, and pangolins, but manual identification from millions of snapshots isn’t an easy feat. Enter the Mbaza AI algorithm which analyzed over 45,000 images from camera traps across 2,500 square miles of forest in 2020. This AI powerhouse classifies up to 3,000 images per hour with an accuracy rate of up to 96%. Conservationists now have fast and accurate tools for monitoring, tracking, and detecting anomalies or warning signs, key in a country losing over 100 elephants to poaching each month. A bonus- the algorithm’s offline capability on a standard laptop is particularly valuable in areas without internet.

A chimpanzee family in the Congo basin, captured on camera. An AI algorithm enables analysis of up to 3,000 camera trap images an hour. Photograph: ANPN-Panthera (1)

Decision making- “what is best?”

Another application of AI is to aid complex decision making. Optimization algorithms can consider many factors like budget constraints, limited resources, and manpower to weigh up trade-offs and recommend optimal solutions. While considering so many constraints can take time by humans, AI can design methods for the most effective allocation of resources in a fraction of the time- without the headache!

The software CAPTAIN (Conservation Area Prioritisation Through Artificial INtelligence) integrates biodiversity data, conservation budgets, climate change, and human impact, developed by scientists at Kew Gardens (2). This software uses a brain-like thinking system to assess trade-offs between costs and benefits for area protection across multiple metrics.

Distinguished by innovative use of reinforcement learning (RL), CAPTAIN plays a simulated game where the reward is measured by spared species. After some iterations, the program learns optimal protected area placements in this artificial environment. In simulations, CAPTAIN consistently outperformed a widely-used conservation planning software Marxan, with an improvement in preventing species loss of up to 18%. Tested on a database of over 1500 native Madagascan trees, CAPTAIN didn’t just outperform Marxan within defined budget constraints- it also increased the average protected range per species by 50 percent, beating established conservation targets. In a world where ambitions are high and funds are low, solutions like CAPTAIN are key to paving the way forward in outlining conservation plans.

Causal inference- “what could happen?”

Finally, the most advanced applications of AI in conservation feature causal inference models. These models go past primitive algorithms, into the intricate web of cause-and-effect dynamics within natural ecosystems. They figure out how various factors connect, helping conservationists understand the deeper impacts of environmental changes and human intervention (3). These models are like the Sherlock Holmes of the conservation world, answering questions about the intricacies of wildlife management and habitat restoration that go beyond just correlations by analysing extensive datasets. They provide a sophisticated understanding of the intricacies between nature and our potential interventions, beyond what you or I could calculate!

An example of casual inference being used in conservation was shown by Harvard scientists who aimed to make better patrol policies to limit adversarial behaviour of illegal fishers, loggers, and poachers in Uganda (4). The proposed AI solution, MIRROR, uses reinforcement learning to minimise loss planning under environmental uncertainty. Derived from real poaching data, a later study confirmed the deterrence effect of patrols- it really works! This solution, the first of its kind, aims to inspire further research in conservation, using realistic models and MIRROR-like solutions to design optimal policies (5).

Despite these technological innovations, it’s key to remember that AI is just a tool. While applied AI can provide insight and direction into conservation efforts, they’re not the saviour of wildlife that we might hope for. People involved in conservation, from scientists to the public remain the key agents in wildlife conservation. Much like students using ChatGPT, we can’t leave AI to do all the work- we have to use these tools to the best of our abilities to ensure a greener future.

References

1. Green, Graeme. “Five Ways AI Is Saving Wildlife – from Counting Chimps to Locating Whales.” The Guardian, 21 Feb. 2022, www.theguardian.com/environment/2022/feb/21/five-ways-ai-is-saving-wildlife-from-counting-chimps-to-locating-whales-aoe.

• Antonelli, Alexandre . “Harnessing the Power of Artificial Intelligence for Conservation | Kew.” Www.kew.org, 24 Mar. 2022, www.kew.org/read-and-watch/artificial-intelligence-conservation.

• Runge, Jakob. “Modern Causal Inference Approaches to Investigate Biodiversity-Ecosystem Functioning Relationships.” Nature Communications, vol. 14, no. 1, 6 Apr. 2023, https://doi.org/10.1038/s41467-023-37546-1.

• WWF Fuller Fund. “Fuller Seminar – Artificial Intelligence and Conservation: Systems/Causality (Lily Xu, Harvard University).” Vimeo, 13 Apr. 2023, vimeo.com/817385079.

• Xu, Lily, et al. Robust Reinforcement Learning under Minimax Regret for Green Security. June 2021.