AWS and Tehanu Use Generative AI to Give Wildlife a Voice in Global Conservation

While Homo Sapiens (commonly known as the human race) share our planet Earth with 8 million non-human species, our attention and conservation efforts reach only a small percentage of this vast biodiversity. Many remain unstudied, with only 2 million documented by science, leaving the majority unacknowledged or forgotten. Animals, trees, and other species play vital roles in sustaining the global economy, which cannot thrive if resources benefit only humans. Tehanu believes that these species must actively participate in our global economic framework to ensure their survival and the planet’s prosperity.

In 2024, Tehanu sought to launch a project focusing on a family of mountain gorillas in Rwanda’s Volcanoes National Park, on the border with Uganda and the Democratic Republic of Congo. While mountain gorillas represent a rare success story in conservation, many efforts to protect other species suffer from fragmented research, limiting the ability to generate implementable findings. With the objective to automate the aggregation of academic studies, such as the gorillas’ behavioral preferences, Tehanu hoped to enable informed policymaking that would support biodiversity. This innovative approach aimed to improve intervention accuracy—such as suggested interests on anti-poaching measures and veterinary care—for mountain gorillas while reducing negative impacts on other species within their habitat.

Using generative AI to process and extract patterns from a comprehensive behavioral dataset on the mountain gorillas, the project set out to demonstrate how AI can uncover the nuanced needs and preferences of non-human species, translating them into actionable insights for human-centered economic systems. By integrating the interests of non-human species into market dynamics, Tehanu aims to create a sustainable model that fosters coexistence and mutual benefit between humans and the broader natural world.

The project’s success relied heavily on the scalability and efficiency of the AWS Global Infrastructure, which formed the backbone of a solution capable of processing extensive unstructured data. The team partnered with conservation experts to refine its system, ensuring the complex needs of mountain gorillas were comprehensively captured. Adastra helped synthesize data from hundreds of academic papers, and analyzed them using a structured pipeline developed by using Amazon Bedrock—the easiest way to build and scale generative AI applications with foundation models. This pipeline utilized Large Language Models (LLMs) to extract, validate, and categorize gorilla preferences. Adastra’s meticulous process of iterative refinement, expert review, and clustering ensured both the accuracy and transparency of insights. The team also conducted human-led inference experiments using an interview guide, followed by thematic analysis, to code the responses and derive key themes— finding that AI-generated preferences offered more operational recommendations than those generated manually.

Adastra implemented AWS services, including Amazon Bedrock, Amazon Simple Storage Service (Amzon S3), and AWS Lambda to support the seamless storage and serverless execution of generative AI features. Using Anthropic’s Claude 3.5 Sonnet model in Amazon Bedrock, Tehanu accurately identified gorilla preferences at a level comparable to human experts, but with significantly greater efficiency. This breakthrough in operational efficiency enables conservationists to analyze and respond to wildlife needs more comprehensively, accelerating conservation efforts that traditionally required extensive manual analysis and expert consultation. AWS also offered scalable compute resources, including Amazon S3, Amazon Bedrock, Amazon Lambda, and Amazon CloudWatch. These resources enabled Tehanu to manage databases for the Internet of Species, infer species interests, and maintain application programming interfaces (APIs) for digital monitoring, reporting, and verification (MRV).

Leveraging Adastra’s AI expertise and the comprehensive cloud capabilities of AWS, the project revealed new ways technology can enhance our understanding of Earth’s biodiversity and support climate change mitigation efforts. Using Anthropic’s Claude LLMs, the model provided insights as accurate as human experts, but with far greater detail and actionable recommendations. Notably, the project pioneered AI-inferred interests for species, an unprecedented achievement in conservation science.

Amazon Bedrock enabled sophisticated experimentation with large language models, providing the foundation for Tehanu to develop a scalable approach to analyzing and understanding species behavior and needs. The initiative focused on inferring the preferences of Rwandan mountain gorillas by analyzing scientific literature and expert interviews, resulting in one of the most detailed valuations of a wild animal population. Covering 440 mountain gorillas, Anthropic’s Claude integrated diverse datasets—including census reports, genetics, health, diet, social behavior, and movements. This unique approach advanced data mining, developed an API for preference identification, and established a foundation for broader applications.

Through a collaboration with Gorilla Vision—an AI research project in Berlin, Germany designed to analyze the behavior of gorillas and identify better ways to protect them—the project achieved over 93% accuracy in gorilla face recognition using ranger observations, camera traps, and crowd-sourced imagery. The project also created an AI-assisted model to generate and validate species-specific interest data, integrated into a secure blockchain system. This system enabled verified conservation actions to trigger payments via Rwanda’s MTN MoMo mobile money network. Remarkably, this allowed gorillas to hold a digital identity, express preferences, and even spend money—marking a historic first for non-human species. The estimated financial value of Rwandan mountain gorillas now stands at $1.55 billion, equivalent to nearly 10% of Rwanda’s annual Gross Domestic Product (GDP).

As a comparison, AI-inferred gorilla preferences from interview transcripts with park rangers were generated in just 13 minutes, compared to 24 hours required by four human experts. With comparable results, generative AI was 110 times faster, highlighting its efficiency in analyzing data and comparing methods. This innovative approach has established a benchmark for using AI in biodiveristy conservation, providing a replicable and impactful model for future efforts involving other species worldwide.