News

S’pore volunteers training AI to crack down on illegal wildlife trade online

SINGAPORE - Exposing illicit wildlife trade that thrives on e-commerce platforms is like a game of whack-a-mole, even with the use of artificial intelligence (AI) tools.

SINGAPORE - Exposing illicit wildlife trade that thrives on e-commerce platforms is like a game of whack-a-mole, even with the use of artificial intelligence (AI) tools. Illicit online listings of the likes of ivory, pangolin scales or even live animals are often disguised using emojis, other languages and code words such as “rare collectibles” that make them difficult to detect, said Ms Jayasri Srikantan, who spearheads a volunteer effort under wildlife conservation group World Wildlife Fund (WWF) to flag illegal trade to the authorities. But an AI model is being developed to assist WWF’s volunteers – called cyber spotters – to evaluate flagged listings on Shopee, Facebook Marketplace and other platforms with greater accuracy. The system, trained on thousands of wildlife product samples and their estimated black market prices, helps to prevent false positives, ensuring the group’s credibility with law enforcement partners. The programme’s accuracy rate is around 90 per cent for now, said Ms Jayasri, adding that new code words and trends are regularly added to the system to keep it up to date with current trends. Deployed in 2023, the WWF’s Cyber Spotter AI model is part of an initiative to fund and kick-start AI projects in Singapore to spur AI adoption, led by national AI programme AI Singapore. The AI system, which has been in the works since 2022, is designed to help improve the accuracy of reports compiled quarterly by the WWF and by its pool of more than 400 volunteers in Singapore, who trawl local e-commerce platforms for hours each month for signs of illegal wildlife trade, said Ms Jayasri, WWF Singapore’s head of education and outreach. Images and text from postings flagged by the cyber spotters are uploaded to an AI model, which analyses text and photos for clues of illegally traded wildlife and their products. The team needs to avoid flagging replicas by mistake and must look out for clues in each listing’s images, text and prices to ascertain if it truly involves an exploited animal. Listings that have a high mix of these elements could be flagged by the software as genuinely illegal. “It’s trickier than it sounds,” said Ms Jayasri. Spotting ivory, for instance, is not as easy as it seems, as Schreger lines – unique patterns found within an elephant’s bones that guarantee its authenticity – are tough to discern through often blurry images online. Some sellers brazenly list protected wildlife products online, but others employ tactics to conceal their illicit offerings, said Ms Jayasri. These tactics include using emojis or code words that hint at the creature being sold, or employing emojis representing the animal, which all “make it challenging for our team to identify genuine listings accurately”. Ivory replicas can be sold at low prices while real elephant ivory is more likely to be listed in the hundreds of dollars, said Ms Jayasri. “But sellers can play all sorts of tricks to slip under the radar, like lowering the prices or using other code words so that they aren’t suspected. The AI needs to consider all of this.” The system does not replace the work of volunteers, who conduct final checks on the reports before they are submitted. Ms Jayasri said the team here flags tens of thousands of listings each year in an effort to prevent the illegal trade – the second leading cause of wildlife loss after deforestation worldwide, and among the world’s most profitable crime sectors. In Singapore, the illegal wildlife trade is , where buyers can purchase contraband items like ivory, or even live and exotic animals smuggled across the border. The Animal Concerns Research and Education Society reported that the number of banned wild animals it found for sale here on Telegram nearly doubled between 2021 and 2023, to 660 cases. The number of sellers offering prohibited wildlife for sale also rose from 66 in 2021 to 387 in 2023. The end goal, said Ms Jayasri, is to fully automate the system so that an AI model is able to trawl websites for illegal listings and compile a report on its own, but the technology has some way to go before achieving this level of autonomy. The international agency aims to introduce the system to its other branches soon, she added.