Researchers at the American University in the United States have developed a new method for discovering which posts on social media contain fake news. Like so many others, these researchers use AI, but instead of machine learning, their method is based on statistical models.
The reasoning behind this is that machine learning can create a "black box" where the answer may be correct, but no one knows how the AI arrived at the answer.
"We would like to know what a machine is thinking when it makes decisions, and how and why it agrees with the humans that trained it. We don’t want to block someone’s social media account because the model makes a biased decision", says Zois Boukouvala's Assistant Professor at American University and one of the researchers behind the model, in a press release.
Anyone who publishes fake news should then be able to find out which posts contain fake news and why they are fake before being suspended. Something that will increase legal certainty as those affected will always know which accusations they need to respond to.
The researchers began by training the AI to recognize language patterns that often occur in fake news. Based on that, the AI was able to create statistical models of which posts contain fake news. Tests then showed that the AI came to the same decision as human assessors in 90 percent of the cases.
So far, this AI is most useful as a research tool, but researchers will continue to work to make it practical for the public as well. The user interface will be improved and the model refined. Hopefully, it will give people better opportunities to distinguish between real news and fake news and in the long run also stop spreading fake news themselves.
"We're designing a tool that will warn the public about fake news and teach them what fake news look like, but it is our belief that people must play an active role and not spread fake news to begin with", says Zois Boukouvalas.
