Meta – Facebook parent company has introduced Sphere AI – an open-source AI tool with the ability to scan various website citations and examine their authenticity. According to Meta, the Sphere AI model can scan the citations on the Wikipedia page, and verify their reliability.
The Sphere AI consists of a dataset of 134 million documents that are split into 906 million passages gathered from public web pages which makes this verification possible and more useful than the other knowledge sources.
Meta’s next goal is training the Sphere in assessing the retrieved documents’ quality, identifying possible discrepancies, and compiling authentic sources. Meta explained that though Wikipedia is precise and formatted properly, it is crowdsourced and as result, not all the knowledge accessible on the web is captured. And its ongoing expansion has made it difficult for editors to double-check every citation or unintended biases.
Meta fed 4 million claims from Wikipedia to its algorithms, training them to focus on a single source from a large pool of web pages to corroborate every statement. The model can then rank the cited source and retrieve alternatives based on their likelihood of supporting the claim.
Meta stated that, when put into use in the real world, the most consistent URLs will be provided by the model as proposed citations for a human editor to analyze and approve.
Meta went on to explain that, when users search for information online, AI models sift through a digital archive to find the relevant information. To ensure the accuracy of the information, knowledge platforms utilize knowledge-intensive natural language processing (KI-NLP). The model’s knowledge base’s comprehensiveness determines the accuracy of the information retrieved.
In contrast to other knowledge tools that rely solely on proprietary search engines, Meta claims that Sphere is a retrieval solution that uses the world’s largest library of information — the open web.
With the release of Sphere, AI researchers will be able to see and control the corpus, as well as experiment with scaling and optimizing different methods to enhance retrieval methods.
According to the company, these models could eventually deal with harmful web content and improve people’s digital literacy and critical thinking skills.
Despite the fact that the AI tool was tested on Wikipedia articles, Meta has stated that it is not working on this project with parent company Wikimedia. Meta clarified that Sphere is still in the research phase and is not currently being used to automatically update Wikipedia content.
Meta released an AI translation tool last week that the company claims works in 200 languages. It went on to say that the lessons learned from this open-source project are being applied to the translation systems used by Wikipedia editors.
Meta has confirmed that it does not use Sphere or a variant of it on its own platforms, such as Facebook, Instagram, and Messenger, which have long struggled with misinformation and toxicity from bad actors. It has its own tools for managing and moderating its own content.
The social network, which has over 2.85 billion monthly active users, has been widely chastised for not doing enough to combat fake news and misinformation. Internal Facebook documents leaked by whistleblower Frances Haugen last year also reveal that the social media company allowed misinformation and hate speech to flourish on its platform.
Meta also intends to discontinue CrowdTangle, the tool used for monitoring misinformation on its platforms, after the November midterm elections.
