Interview with AAAI Fellow Tanya Berger-Wolf: AI for ecology, biodiversity, and conservation
Summary
AAAI Fellow Tanya Berger-Wolf, a Professor at Ohio State University, discusses her pioneering work in AI for ecology, biodiversity, and conservation. Her research focuses on advancing computer science frontiers while addressing critical environmental challenges, such as monitoring the estimated 30 to 50 million species, of which only two million are named. Berger-Wolf leads the NSF-funded Imageomics Institute and the AI and Biodiversity Change Global Center, developing tools like BioCLIP and BioCLIP2. Released in 2024 and 2025 respectively, BioCLIP2 scaled to 214 million images, enabling new species discovery and revealing unlabelled biological traits like age, sex, and health within species embeddings. The model achieved over 90% accuracy in identifying disease-carrying ticks for the Ohio Department of Health after fine-tuning. Additionally, her co-founded Wildbook project uses AI for individual animal identification across 200+ species, contributing to conservation decisions, including the reclassification of whale sharks from vulnerable to endangered on the IUCN Red List.
Key takeaway
For research scientists developing AI for biodiversity and conservation, you should prioritize integrating domain-specific structural knowledge and large-scale data into your models. This approach, exemplified by BioCLIP's success in revealing latent biological traits and enabling new discoveries, will enhance model accuracy and scientific utility. Consider fine-tuning these foundational models for specific, high-impact applications. This can include disease vector identification or individual animal tracking to drive tangible conservation outcomes and inform policy.
Key insights
AI, by integrating biological structure and massive datasets, can uncover hidden biological traits and drive scientific discovery for conservation.
Principles
- Incorporating biological taxonomy and phylogeny into ML architectures improves model performance and interpretability.
- Large-scale biological image data enables foundation models to discover latent, unlabelled traits like age or sex.
- AI models can detect subtle biological features that human observation often misses.
Method
Develop domain-grounded foundation models by integrating biological structural knowledge into machine learning architectures and scaling data to uncover latent biological information.
In practice
- Apply foundation models to generate novel hypotheses in evolutionary biology, such as linking sexual dimorphism to mating systems.
- Fine-tune general biological models for high-accuracy, fine-grained classification of specific species, like disease vectors.
- Utilize AI for individual animal identification to inform conservation policy and resource allocation.
Topics
- AI for Conservation
- Biodiversity Monitoring
- Foundation Models
- Imageomics
- BioCLIP
- Ecological Data Science
Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Policy Maker
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by ΑΙhub.