Wikipedia, the free and collaborative encyclopedia used by millions of people, may contain false information, either by inadvertence of those who edit it or by intentional movements. Without going any further, we can point to the case of the Chinese woman who for more than a decade wrote misleading articles about Russia, leading many users to believe that it was real information.
The Wikimedia Foundation, the non-profit organization behind the digital encyclopedia, acknowledges this situation. Although it has verification mechanisms, the challenge of avoiding errors is increasing for volunteers, who see more than 17,000 new articles added each month. Now, an artificial intelligence (AI) model developed by Meta could help them in their tasks.
Sphere, an AI for Wikipedia
One of the tools that Wikipedia editors have to determine if the information in an article is accurate is sources or references listed at the bottom of the page. The problem is that as the encyclopedia grows, it becomes more complex to verify them manually. What if we could use AI to work with that huge amount of information and help publishers?
Meta thinks it’s possible. In this sense, he has developed a system that allows you to automatically review hundreds of thousands of references at the same time. According to the company, this solution, known as “Sphere” has two main pillars: a trained algorithm with more than 4 million snippets of Wikipedia text, and a data set fed by information from 134 million web pages.
To check the references of a new article, Sphere performs several steps. The first is a search within an index that classifies the information to avoid wasting time analyzing data that does not correspond to the subject in question. When the system finds a document that could potentially be cited as a source, it extracts the most relevant passage from it and processes it through a classification model, which compares the new text with the original citation.
The truth is that the changes will not be made automatically in the article. If the system detects an inconsistent reference, the alternatives it deems most suitable will be presented to a person who will be in charge of approving the changes. Meta says that Sphere will be able to work in several languages and not only process text, but also images, videos and data tables.
Of course, we will have to wait to see this solution in action. The company says the system is still in the research phase and is not yet used on Wikipedia content. However, he is confident that later it will ease the burden on editors, not only of the world’s largest digital encyclopedia, but of other projects that require verification of huge amounts of data.
It should be noted that Sphere is an open source project, which will allow the community to examine how the algorithms work. On the other hand, the company led by Mark Zuckerberg has not revealed whether it plans to implement this system on its social networks Facebook and Instagram to avoid false news. Following the controversy surrounding disinformation, Meta had promised to invest in new tools so that users of its platforms know what information to trust.
As for artificial intelligence, we can say that it is a very promising discipline. Not only does it allow us to increase the possibilities of autonomous driving, create photorealistic faces, images from text and win at chess, to name a few examples. It could also help make Wikipedia a much more accurate information page.. It will work? In time we will know.
Images | Oberon Copeland (Unsplash) | goal AI
In Xataka | The paid Wikipedia is already a reality: this is what companies like Google or Amazon will charge for the use of your data