Page 224 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

The International Fund for Animal Welfare (IFAW), a leading nonprofit in animal conservation,
initially relied on manual monitoring to identify online illegal wildlife trade. However, this labor-
intensive process was inefficient and prone to errors. In 2019, IFAW partnered with Baidu to
launch AI Guardian 1.0, powered by Baidu's deep learning platform PaddlePaddle. The system
used image recognition to detect trade in elephant, pangolin, and tiger products.
By mid-2024, the tool had screened over 360,000 images and led to the removal of 7,853 illegal
postings. In November 2024, AI Guardian 2.0 was launched to target the growing market of
live exotic pet sales. Trained on a decade's worth of IFAW image data, the updated model
can detect 34 species and boasts an 86% accuracy rate. A zero-code interface enables non-
engineers to customize and deploy the model for real-time monitoring of online platforms.
Use Case Status: Operational

Category: Endangered Species Protection

2�2 Benefits of use case

The platform detects illegal wildlife products and support legal enforcement and thereby
promoting sustainable practices, reducing biodiversity loss, and protecting marine and
terrestrial ecosystems.

2�3 Future Work
• Data Expansion: Include more endangered species and items from different regions.
• Capacity Building: Launch training modules for enforcement agencies and NGOs.
• Regional Pilots: Expand AI Guardian deployment in Southeast Asia and beyond.
• Interface Improvements: Add multilingual support and simplify model deployment tools.
• Open Source Collaboration: Encourage developers to create custom biodiversity
protection tools using PaddlePaddle.

3 Use Case Requirements

• REQ-01: It is critical to address the scarcity of annotated image data for illegal wildlife
products.
• REQ-02: It is critical to handle class imbalance, as negative samples far outnumber positive
ones.
• REQ-03: It is critical to accurately differentiate between real wildlife products and synthetic
imitations.
• REQ-04: It is critical to detect altered features in processed wildlife goods (e.g., carved or
dyed).
• REQ-05: It is critical to enable fast scanning and flagging across digital platforms to
prevent illegal sales.

188