Facebook launches AI tool Sphere to help online publishers fight misinformation

Facebook parent company Meta Platforms has released a new artificial intelligence (AI) based tool called Sphere that can help online companies tackle misinformation more effectively. Meta has open-sourced Sphere and made it available on GitHub.  

Sphere will help researchers train retrievers to handle a wide range of documents and build automatic systems to filter out misinformation, noise, and incoherent text, claims Meta.   

Sphere has been tested on Wikipedia to scan entries and verify if a citation on the platform is accurate or not. According to Meta, Wikipedia is accurate and well-formatted, but it is also crowd-sourced and doesn’t capture all the knowledge available on the web. 

“Its continued growth has made it challenging for editors to double-check every citation or inadvertent biases,” Meta said in a blog post.  

Meta explains that when users seek information online, AI models comb through a digital archive to retrieve the relevant information. To ensure the information is correct, knowledge platforms use something called knowledge-intensive natural language processing (KI-NLP). The accuracy of the information retrieved depends on the comprehensiveness of the knowledge base used by the model.  

Meta claims that Sphere is a retrieval solution that uses the world’s largest library of information — the open web, unlike other knowledge tools that rely only on proprietary search engines.  

“When we use such proprietary search engines, we don’t know what we can’t see. Reader models might miss relevant information because the search engine algorithms rank it too low in the results,” Meta said.  

According to Meta, Sphere has 134 million documents split into 906 million passages and can access far more public information than other knowledge sources used in current KI-NLP research. As a result, it can provide more useful information that they cannot. 

Meta said that its next goal is to train Sphere to evaluate the quality of retrieved documents, detect potential contradictions, and prioritise reliable sources.  

Open-sourcing Sphere will allow AI researchers to see and control the corpus and experiment with scaling and optimising different methods to improve retrieval methods.  

The social network with over 2.85 billion monthly active users has been widely criticised for not doing enough to tackle fake news and misinformation. Facebook’s internal documents leaked by whistleblower Frances Haugen last year also show that the social media company allowed misinformation and hate speeches to grow on its platform.