An artificial intelligence system that can tell the difference between real and fake news — often with better success rates than its human counterparts — has been developed by researchers at the University of Michigan. Such a system may hep social media platforms, search engines, and news aggregators filter out articles meant to misinform.
“As anyone else, we have been disturbed by the negative effect that fake news can have in major political events [and] daily life,” Rada Mihalcea, a UM computer science professor who developed the system, told Digital Trends. “My group has done a significant amount of work on deception detection for nearly ten years. We saw an opportunity to address a major societal problem through the expertise we accumulated over the years.”
Mihalcea and her team developed a linguistic algorithm that analyzes written speech and looks for cues such as grammatical structure, punctuation, and complexity, which may offer telltale signs of fake news. Since many of today’s news aggregators and social media sites rely on human editors to spot misinformation, assistance from an automated system could help streamline the process.
To train their system, the researchers represented linguistic features like punctuation and word choice as data, then fed that data into an algorithm.
“Interestingly, what algorithms look for is not always intuitive for people to look for,” Mihalcea said. “In this and other research we have done on deception, we have found for instance that the use of the word ‘I’ is associated with truth. It is easy for an algorithm to count the number of times ‘I’ is said, and find the difference. People however do not do such counting naturally, and while it may be easy, it would distract them from the actual understanding of the text.”
The system demonstrated a 76-percent success rate at spotting fake news articles, compared to around 70 percent for humans. Mihalcea envisions such a system helping both news aggregators and end users distinguish between true and intentionally false stories.
The system can’t completely compensate for humans, however. For one, it doesn’t fact check, so well-meaning (but ultimately false) content will still slip through.
The researchers will present a paper detailing the system was presented at the International Conference on Computational Linguistics in Santa Fe, New Mexico on August 24.