Where AI, content policy, and elephants merge: How Meta and WWF are working together to protect the herd

For anyone who has witnessed first-hand a large tusker elephant in its natural habitat, you know just how awe-inspiring this moment can be. We have both had the privilege of seeing these amazing creatures up close and personal in African game reserves. Unfortunately, the experience also serves as a reminder of just how rare this opportunity is becoming.

Live animals and products made from hundreds of different species like elephants, pangolins, and tigers are in demand around the world for uses ranging from ornaments and jewelry to alleged medicinal cures. The UN estimates that this illegal trade in wildlife has an annual value of over 20 billion USD and is one of the top five illicit markets in the world.

And like everything else, this trade has moved online.

WWF researchers can find endangered species products as well as live, exotic pets like primates, cheetah cubs, and threatened parrots for sale online in just seconds. Many of these are protected and illegal to trade, having been smuggled from the most remote corners of the world in horrible conditions.

But just as technology can enable such nefarious purposes, it can also serve as a tool that allows us to explore the world around us and help to protect it. Nowadays, anyone can have their own awe-inspiring moment from the comforts of their living room through tech innovations like virtual reality, transporting them directly to the African savanna. You can even help scientists sort through and label wildlife images to advance conservation. The proliferation of AI, in particular, has emerged as a game-changer. It is seemingly everywhere, and, if you believe the hype, it can do anything. One area that interests us is deploying AI to help save nature by identifying prohibited wildlife sales hidden across billions of social media posts.

For years, WWF and the conservation community have manually searched online platforms to identify content for company escalation and review. This is not only incredibly time consuming but also creates unintended bias based on the researcher’s knowledge and language skills, which have their limits.

Through their participation in the Coalition to End Wildlife Trafficking Online, which unites technology companies with WWF and conservation partners IFAW and TRAFFIC to reduce wildlife trafficking online, Meta has leaned into an industry-wide and cross-sector approach to tackling this issue.

Meta recently reported to the Coalition that they removed 7.6 million pieces of content in 2023 alone that violated their endangered species and live animal policies.

The Coalition supports the tech sector to (1) harmonize prohibited content policies, (2) train company moderators to identify prohibited content, (3) encourage users to report suspicious posts, (4) incorporate known tactics and search terms into AI block filters, (5) and at the core, share learning with other companies to accelerate impact.

Through this approach, WWF and Meta have worked together to:

• Regularly review monitoring data and incorporate insights surfaced by WWF and TRAFFIC experts into Meta’s AI models to more accurately identify prohibited wildlife trade at speed and scale.
• Review and update prohibited content policies for wildlife to ensure the latest scientific recommendations are incorporated in the company’s advertising policies and their overarching Community Standards.
• Train moderators to identify policy-violating content on Facebook and Instagram.
• Launch a pop-up alert on Instagram and Facebook to help users understand that these animals and their parts may be trafficked from the wild.
• Host an industry panel at TrustCon to invite other policy teams to collaborate on illegal wildlife trade.
• Host an internal Meta hackathon to identify additional policy and product solutions for prohibited content on Facebook and Instagram.

Despite rapidly evolving tactics by traffickers to avoid detection, these counter-efforts are working.

Combined with the 16 million offers for sales and accounts blocked by Coalition company members from 2018-2023, this brings the sector’s response to 23.6 million sales and accounts directly blocked. And this would have been impossible solely through manual, human review.

Content moderation continues to rely on a combination of technology and human review to remove violating content. We are excited that advances in AI technology have greater potential to advance efforts to protect endangered wildlife globally from online trafficking.

For example, a few years ago Meta unveiled a tool called Few-Shot Learner which can adapt more easily to take action on new or evolving types of harmful content quickly, working across more than 100 languages. Whereas before Meta needed millions of examples to train an AI model, Few-Shot Learner only requires a handful of examples.

More recently, Meta sought to augment these advancements by training Large Language Models (LLMs) on its Community Standards to help determine whether a post violates its policies. These types of innovations hold the potential to dramatically improve content moderation. And it is partnerships like the Coalition to End Wildlife Trafficking Online that will allow us to combine forces and hone in on this particularly challenging policy issue, which, if unchecked, will continue to threaten the future existence of countless species.

One day, we both hope to return to the stunning landscapes of Africa’s game reserves, perhaps to see the same tuskers and be thankful their tusks did not end up as carved statues, for sale online.