In a recent collection of documents, leaks from Facebook revealed that social networks have been less successful in understanding the different languages used across the Middle East and North Africa. As the Arabic language is the third most common language used on Facebook with the people of the Middle East and North Africa using Facebook services on a large scale than any other country. But when it comes to understanding the Arabic language, Facebook is failing and getting warnings about how the company’s human and computerized reviewers struggle to understand the various regional language used across the Middle East and North Africa. Which resulted in, the company wrongly editing a benign post for promoting terrorism and exposing Arabic speakers.
Therefore the Meta’s Facebook owner has decided to establish a new Artificial Intelligence (AI) moderation system as it will require much less training data for some tasks that will adapt to new enforcement jobs more quickly than the predecessor and named this new AI system as ‘Few Shot learner’. This AI system works in more than 100 languages and can operate on images and text. Back in September, ‘Few shot learner’ helped to impose a rule introducing banning posts that discourage people from the Covid-19 vaccine.
Facebook also shared that Few shot learner was established earlier this year and contributed to decreasing their record on the worldwide common issue of hate speech but didn’t share many details on the performance. Facebook is claiming to bring people together but according to United Nations, it contributed to the genocide against Rohingya Muslims in Myanmar.
Percy Liang, the Stanford center’s director, stated that Facebook’s system seems to show some of the impressive power of these new models, but will also exhibit some of their trade-offs. It’s exciting and useful to be able to direct an AI system to do what you want just with written text, as Facebook says it can with new content policies, Liang says, but this capacity is poorly understood. “It’s more of an art than a science,” he says. Liang also stated as the system requires less training data, it’ll show some drawbacks.