Facebook Meta asserts that artificial intelligence (AI) has the potential to improve the reliability of Wikipedia.
As a go-to source for nearly everything these days, Wikipedia is allegedly riddled with questionable and misleading references.
There is no need to be concerned, as the business’s AI, Sphere, has built a model capable of analyzing hundreds of thousand citations simultaneously to assess whether they genuinely support the relevant assertions. It is here to assist, the company assures us.
“An order of magnitude bigger and substantially more complicated than previously utilized for this type of study” is Meta’s claim for its new dataset of 134 million public web pages.
As an alternative to Google’s standard, proprietary search engine, Sphere has already gathered 134 million documents from all across the internet (opens in new tab).
Sphere was built using CCNet, a variation of Common Crawl, which Meta claims will benefit other AI researchers working on knowledge retrieval projects.
A long-term objective is to create a platform that will assist Wikipedia editors in finding and swiftly correcting problems with the accuracy of their in-text citations.
The program apparently flags problematic citations, enabling human editors to examine the instances most likely to be incorrect without having to trawl through hundreds of correctly referenced assertions.
According to Meta’s approach, it will recommend a more relevant source if one of the citations seems to be irrelevant.
The announcement comes at a time when Wikipedia is exploring for other sources of funding outside contributions.
According to a recent announcement from the Wikimedia Enterprise platform, firms like Google, Amazon, and Facebook that utilize Wikipedia as a resource will now be charged.
It’s possible to download the project’s source code on GitHub here, as well as read a comprehensive write-up of the project’s results here or see the demo here (opens in new tab).