The great thing about the internet is that there is a lot of data to begin with.
Catching fake news before it has real consequences can be difficult, as many websites and news outlets are heavily reliant on human editors who often can't keep up with the constant influx of news. A step up from human curators, is to use machines for automation.
To make computers smart and capable of performing beyond its directives, is by using AI. But AI needs a lot of data to learn, and that before it can prove itself useful. This is because AI needs to find common patterns so it can differentiate things and do what it is supposed to do.
This is what researchers from the University of Michigan and the University of Amsterdam are doing. They created an AI for detecting fake news on the internet by crowd-sourcing hundreds of fake news and misleading articles, and fed all that to the AI they were creating.
The algorithm uses natural language processing (NLP) to search for specific patterns or linguistic signals that indicate a particular article is fake news. This approach is very different from fact checking algorithms implemented on tech companies, including Facebook.
While the method isn't at all groundbreaking, the researchers were capable of building the AI, and made it to understand and detect fake news better than humans.
To create this AI, the researchers had to first decide what fake news are. They need to describe what they are, so the machine can figure things out automatically.
For this, they turned used the "requirements for a fake news corpus" developed by a team of researchers from the University of Western Ontario. These were needed because if the researchers want to make the AI capable of detecting fake news, the AI should first know how to detect non-fake news.
It also needs to be able to verify the ground-truth, and all that by accounting external factors such as developing news, different languages used and cultural interpretations.
There are several probabilities of "fakery", and they include: intentionally false information (serious fabrication), hoaxes (created for the intent of going viral on social media) and those that are created as humor or satire.
After setting out the rules, the researchers crowd-sourced their dataset by asking Amazon Mechanical Turk workers to reinterpret 500 real news stories as fake news. Participants in the study were then asked to imitate the journalistic style of the original article, but alter some of the facts and information to ensure the result was fake.
The network learns the definition of fake news, by feeding on both fake news and real news. Once finished training in a controlled environment, the team then fed it news straight from a dataset containing both real and fake news gathered from the web.
The result is outstanding, as the AI was able to perform better than humans at figuring out which was which.
It’s not perfect, as it successfully found fakeries up to 76 percent of the time. But it's indeed a great start since it outperforms humans which have a success rate of 70 percent.
"You can imagine any number of applications for this on the front or back end of a news or social media site," said Rada Mihalcea, one of the researchers on the project.
"It could provide users with an estimate of the trustworthiness of individual stories or a whole news site. Or it could be a first line of defense on the back end of a news site, flagging suspicious stories for further review. A 76 percent success rate leaves a fairly large margin of error, but it can still provide valuable insight when it’s used alongside humans."
In this "fake news era", the most likely solution to spot fake news is by relying on people that swear to never lie and news outlets that create news based on thorough research from multiple sources. But in real life, things can go far from this.
Mihalcea said that an automated solution could be a step up from previous attempts. It can be an important tool for sites that are struggling to deal with an onslaught of fake news stories, often created to generate clicks or to manipulate public opinion.
Even as computers can be programmed and taught "comparable to human ability to spot fake content," it's not a guarantee that it can eliminate the problem entirely. But still it's indeed showing a foreseeable future where AI can help humans in combing the multitude of data the internet has to give.
The paper by the researchers is titled "Automatic detection of Fake News." The research was supported by U-M’s Michigan Institute for Data Science and by the National Science Foundation