[Science] Facebook’s fact-checking process is too opaque to know if it’s working – AI

Facebook is trying to combat fake newsInk Drop / Alamy Stock Photo By Donna LuFacebook is attempting to stop the spread of fake news on its site, but it needs to ramp up it’s efforts and be more transparent about if it’s working. That’s the conclusion of a report released by UK fact-checking charity Full Fact on Tuesday, detailing the first six months of a partnership with Facebook. Full Fact is one of 50 fact-checking firms working with Facebook to review and debunk false content on the site. The tech giant first began partnering with fact-checkers in the US in December 2016, after criticism about its failure to stem the spread of fake news in the lead-up to the presidential election. Under the program, Facebook provides independent fact-checkers with content to review and rate. Fact-checked content is automatically marked on Facebook, so users are informed if they share a post that has been reviewed. Stories, images and videos that have been rated as false are still shareable but are shown lower in news feeds by Facebook’s algorithm in an aim to reduce their reach. An analysis by Buzzfeed News last year, found that the top 50 fake stories on Facebook in 2018 generated around 22 million total shares, reactions, and comments. Advertisement Read more: Fake news generating AIs could be the best weapons to fight fake news Between January and June, Full Fact’s editorial team of seven people published 96 fact-checks chosen from a queue of content provided to them by Facebook. The posts they reviewed contained potentially harmful content, such as false statements about vaccines, and some had been shared over 100,000 times. But most of the queue provided by Facebook contained content that couldn’t be fact-checked – such as statements of opinion and random links including a swathe of Mr Bean videos – pointing to the ongoing difficulty in monitoring the more than one billion pieces of content posted to the platform daily. “Facebook’s algorithms are not yet at a stage where they can reliably identify information that is inaccurate,” says Will Moy, director of Full Fact. This is a blow for the company as they, as well as other tech companies, have said that artificial intelligence should be used to help tackle the problem of fake news. But it doesn’t seem ready yet. Read more: Singapore’s ban on fake news sets a dangerous precedent Julia Bain at Facebook in the US says Facebook’s algorithm sends content to fact-checkers based on metrics such as users flagging a post and disbelief expressed its comments. After content is rated as false by fact-checkers, its reach on Facebook decreases by more 80 per cent, says Bain. But Facebook has not provided Full Fact with detailed information about what impact their fact-checks have had on reducing false content. “We need more information to evaluate how well the program is working,” says Moy. Firms such as Snopes have quit Facebook’s fact-checking program with similar criticisms about the platform’s lack of transparency.Facebook has begun providing partners with reports on the number of fact-checks conducted and the number of notifications users see as a result, says Bain. Moy says the partnership has been a positive first step, echoing the views of other fact-checkers, but for it to have an appreciable effect, significantly stepping up both review speed and content volume is necessary. “Misinformation travels a lot faster than the corrections do,” says Michelle Amazeen at Boston University in the US. Fact-checking can be effective, but the risk is that people may believe false content to be true because it hasn’t yet been labelled as debunked, she says. More on these topics: social media