Ever since the election, talking heads have discussed how “fake news” could have tipped the election to Donald Trump by disseminating false stories favorable to him or unfavorable to Hillary Clinton. Social media users can spread these fake new stories far and wide, creating an aura of credibility which can make fake stories hard to distinguish from the truth.
Social media websites like Facebook have begun to cooperate with human fact checkers to track fake news websites and notify users that a story is likely false, but it is impossible to trace every single news story as they pop up. There is just too much information on the internet for any human mind to process.
So maybe an artificial intelligence (AI) can perform better. Facebook along with researchers and hackers are examining whether artificial intelligence and big data can help track down fake news stories faster than humans can. But the unique nature of analyzing fake news creates problems which means that it is currently likely beyond the skill of an AI. But by working together, humans and AI combined with greater skepticism from the public can help end the fake news problem.
The challenges of distinguishing truth from lies
While Facebook wants to prevent fake news from leaking, it has also strongly emphasized that it does not want to be an “arbiter of truth.”
The challenge which Facebook faces is that truth is often subjective. It is one thing to shut down news articles which can be emphatically proven false such as one story which said that that Elizabeth Warren endorsed Bernie Sanders for president. But some people worry that Facebook could shut out different opinions and start taking a partisan stance, even if it is unconscious. The fact that Facebook allegedly suppressed conservative news stories last year is proof to some that Facebook cannot be trusted to remain unbiased.
All of this means that if Facebook and other websites want to shut down fake news websites, they face an incredibly high bar to avoid suspicion. But there are ways in which AI could help reach that bar, particularly through pattern recognition.
The idea is that an AI could detect patterns or words which would indicate a fake news story. For example, a 19-year old Stanford student named Karan Singhal recently developed a web extension called the Fake News Detector AI which looks at 55 different metrics. Singhal essentially took track of factors he would use to determine whether a news site is real or fake such as the wording and layout, and programmed those factors into a neural network which would determine whether it is fake or not.
While Singhal’s approach may be simplistic and thus could be beaten by just changing the website’s design, the idea behind it is how AI developers hope to tackle the fake news problem. Ideally, an AI could detect certain patterns of words which could reveal whether a news story is likely to be fake or not. We’ve seen this approach implemented on various comparison website, which use AI algorithms to help you compare the best product. Companies such as eBay and Amazon have implemented this on their own sites.
Mistakes will be made
But while some progress has been made, completely being able to distinguish fake from real news is a task that is currently beyond the grasp of modern AI.
While AIs can detect patterns, determining whether a website is fake or not is fundamentally a judgment call that would require a human-level understanding of society. After all, even humans cannot nail with certainty whether a story is fake without doing additional research. Dean Pomerleau, a Carnegie Mellon professor who has led the way in calling for an algorithm that can find fake news, admitted to Wired that if an AI could reliably detect fake news, “It would mean AI has reached human-level intelligence.”
On top of the judgment problem, there is the obvious fact that news are well, new. If a smaller website breaks a story which older, more established websites have not noticed, an AI may just reject it as fake even if it turns out to be true. Judging news requires a holistic view of the world which AI do not possess.
A lot of reporting around AIs in general is about the fear that AI will supplant humans entirely. But what is more likely is that both humans and AI will serve different purposes in society as they complement one another.
The field of fake news is one potential example. An AI can read many more articles than a human being ever can and thus use pattern recognition to inform the human that a news story might be fake. But it would be up to the human to make the final judgment call on whether it is fake or not.
The fight against fake news is a challenging one as our society has to figure out how far we want to go to quell fake news, the role of companies like Facebook in battling fake news, and what precisely truth is. Artificial intelligence can certainly help alongside humans. But it is no substitute for good judgement, skepticism, and the critical thinking which a democratic society needs.
This article is published as part of the IDG Contributor Network. Want to Join?