Imagine having your browser prevent fake news from reaching you.
That could be possible sooner than you think. Recently published research from MIT reports the use of AI to detect language patterns that sort the false from the factual.
That said, the average internet user finds hardware or software topics more in their search engine history than fake news warnings or where to find accurate news. This is set to change as more people consider fake news a serious concern. It’s recently been highlighted by The Times as a high-risk area in need of further investigation and funding.
Government reports on risk in Fake News
High-risk is now a common collocation with fake news, from social media posting to elections. Indeed, when a report of the Interdepartmental Group on Security and Ireland’s Electoral Process and Disinformation was published in July 2018, it highlighted areas of concern.
At that time, the Government report found two risk areas: low and medium. The former was of the electoral system (at polls), traditional print, and broadcasting, in which tighter controls exist on where information is from and how it spreads.
Medium risk was found in cyber security and funding. The think tank suggests that due to the nature of the internet and the speed of technological advancement, it’s essential for any policies put in place in this area are to be reviewed on a regular basis.
Further news on fake news in Ireland suggests the youth – teenage girls in particular – are most at risk in this internet age, hence Safer Internet Day being an annual event on February the 5th. It recognises the need for advancement in security online and isn’t just a “token day” if recent MIT research is taken into consideration.
It’s the vulnerable who can be protected most by AI technology developed to sort the fact from the fake, the original from the modified.
Fake News Speak is detectable
Research is underway at MIT in the use of AI to detect language patterns that sort the false from the factual. Automated fake-news detection systems are currently machine-learning models designed to find tiny differences in language patterns that could be used to find false/factual distinctions in news articles.
The question is: is any deep-learning fake news detection model reliable? What the model attempts to do is examine which kinds of words come up more often in fake news articles.
Thus far, models have revealed some key and consistent differences that can help everyone trust we’ve been accurately informed. In particular, researchers O’Brien and Boix found more exaggerated language. Prone to dramatic language, fake news articles seem to contain more superlatives, like “the best” and “the worst.” Genuine news uses less extreme, and thus more conservative, language.
How is this research being done?
Meet Kaggle, a fake news research dataset that uses 244 different websites as its sources, taking in 12,000 sample articles. Using 11 thousand real news samples mostly from the New York Times and The Guardian, the model scans layers looking for markers that indicate groups of similar words.
The model doesn’t scan for topics, but rather lexical patterns. When fed news, it’s trained to examine articles in layers, and able to retrace its steps allowing for analysis of how it reached its conclusion.
Protecting the mind from manipulated content on social media
What does this mean for the average internet user? Wouldn’t it be helpful to have your browser bin fake news before it even reaches you? It would sift through any news to see whether it’s genuine, or designed to mislead you, then flag it for you.
Fake news isn’t just inaccurate. Its purpose is to distract or disturb, and its ability to destabilise foundations for safer internet usage is not a far-flung idea.
What could come from this research is less sifting, and more sign-posting. Your browser could warn you the page you’re on contains information likely to be fake, or flag that the content is written with intention to mislead. We’re used to seeing security warnings in relation to cyber-hacking, but protecting our minds is as important as protecting our assets.
When tested for accuracy, the model testing for fake news was 87% accurate. The issues researchers face at MIT include whether the databases were taught writing styles rather than language patterns; further studies are underway.
Provenance plans to provide safety
Closer to home, further research at Dublin City University is taking place. The project, called Provenance, is funded by the EU. Its predominant aim is to track and flag online disinformation in social media. What this means is it should help in distinguishing factual and fake articles or revealing when information has been manipulated.
The team at Dublin City University’s institute for Future Media and Journalism (FuJo) use advanced digital technologies for multimedia analytics and image forensics to find modification of content. Not only can this be used for the flagging of disinformation online, but it will also be able to track content, which could be useful for content creators, though software like this already exists and is used daily.
In this project, verification indicators allow for information to be contextualised because researchers can track it to first registration and when updates or amendments have been made to it. The issues at hand are not only the speed at which misinformation can spread, but the sheer volume of it, as noted by the head of the project, Dr Jane Suiter, Associate Professor at FuJo.
Though this local study has different methods, both groups of researchers at MIT and Dublin City University are involved in using AI and advanced multimedia technology to understand how fake news spreads, and what we can do to stop it from spreading.
It may soon, then, be a reality to have your browser tell you the content you’re taking in has been modified or is otherwise false. For the most vulnerable members of our society, this progress can only yield vital protection strategies in an effort to create safer platforms for everyone online.