The model developed by the researchers is able to detect fake news in 82% of cases.
On April 3, 2020, researchers published their study at Microsoft in collaboration with Arizona State University. The artificial intelligence model developed by scientists relies, in part, on the commitment of a publication on social networks. Microsoft believes that this AI outperforms a large number of advanced models in detecting fake news.
AI that can spot weak signals
Fake news, is everywhere. Even in the midst of a health crisis, you have a good chance of seeing them pass. Despite the efforts of Facebook and Twitter (among others) to combat misinformation, miracle cures and conspiracy theories continue to rain down on the net. A survey conducted before the Covid-19 crisis in 2018 by the Brookings Institute revealed that 57% of Americans had seen fake news in the midtermmandat elections. More worryingly, 19% of them think it has even influenced their vote.
Conventional artificial intelligence models take time to detect weak signals, such as false comments. Some fake news may not be detected for several hours or even days. Which obviously makes their impact much stronger. Other en artificial intelligence can be influenced by relying only on false comments, sometimes published by fake profiles. This is where Microsoft’s AI could make the difference.
The hunt for fake news
Thanks to a module called the weighting network (LWN) label, Microsoft’s AI is able to model the different labels that regulate the learning process. . Several tests have been carried out and the researchers claim that their model was able to accurately detect fake news in GossipCop and PolitiFact in 80% and 82% of cases. Current models are around 73%. Microsoft used the open source Data set FakeNewsNet to test its system.
Social networks are trying at all costs to limit the spread of fake news. In the summer of 2019, Instagram announced that a feature was finally available to report fake news. For its part, Facebook recently launched with Reuters a tool to spot deepfakes, these videos designed using an artificial intelligence technique that consists of superimposing existing images and videos on other images and/or videos to make anyone say what you want.