Fishing for Fake News
Recently, fake news has become a news event in itself. Earlier this year in a workshop with a group of banking executives in Harare, one of them shouted at me from the back of the room, “What about fake news?” I had to admit I didn’t have an algorithm in my rucksack for that particular problem.
Attempts to misinform have been with us since the beginning of time. But in 2016, the Macquarie dictionary went so far as to award “fake news” the dubious honour of Word of the Year. So what is it about the current phenomenon of fake news that makes this so different from anything we’ve seen before?
The advent of the internet has seen many more people getting their news from social media as opposed to traditional news outlets. In the United States for example, the number of adults who have seen news on social media channels grew from 49% in 2012 to 62% in 2016. In the Philippines, most people with an internet connection reportedly get their news from Facebook - and only Facebook. It is not an exaggeration to say the incidence of fake news on social media channels, poses a severe threat to societal discourse and the decisions that flow from it. Indeed the UK House of Commons Digital, Culture, Media and Sport Committee published a report in July this year that argued that the UK democracy is at risk due to the “dissemination of disinformation on social media for the purpose of manipulating the public in election and referendum voting”.
Fake news has a few important characteristics: Firstly, it is both easy to produce and consume and can be disseminated very rapidly and in great volumes. As a result, there has been much more effort and focus placed on research into the algorithmic detection of fake news. Fake news however, can be very hard to detect.
Why is detecting fake news so difficult?
In a paper on the detection of fake news in social media, researchers in the United States remind us that fake news is “intentionally written to mislead readers to believe false information”. Furthermore, fake news exhibits several diverse styles and topics and may present verifiable facts within incorrect contexts in order to make non-factual claims. While automated fake news detection tries to extract tell-tale features from an article’s author, headline, content, linguistic style and images, separating truth from falsehood based on news content alone, is in fact quite difficult to do.
Researchers are now turning to auxiliary information such as a user’s social engagement with fake news in order to tell credible users from mischievous ones. Bearing in mind that fake news is likely to be spread by non-human bots, the paper assumes that the spreaders of fake news and real news may form different communities, each with its own unique characteristics.
How bad is fakes news really?
The impact of fake news in elections is serious enough for policy makers to not only sit up and take notice, but to actually do something about it. In a paper published this month, researchers Dorje Brody and David Meier simulated 100,000 two-candidate elections with and without fake news. They found that a candidate who would normally have lost an election in the absence of fake news, ended up winning it 30% of the time when fake news was introduced.
While their underlying models can be developed further, Brody and Meier’s work seems to suggest that in the presence of fake news, a simple majority is in fact too low a hurdle for success in an election or a referendum.
What about human bias?
Even if we were able to characterise and detect fake news accurately all the time, we would still be faced with the problem of human bias. We tend to believe that our views are the only correct ones and we gravitate to opinions that support our own world view. Gordon Pennycook, an assistant professor at the University of Regina’s Hill/Levene Schools of Business, has shown that only a small kernel of truth within a story is sufficient for repetition to increase perceived accuracy. Other psychological studies have demonstrated that when fake news is debunked with true, factual information, this is often regarded as unhelpful and can even increase misconceptions, and especially among ideological groups.
So the outlook for defeating fake news entirely remains rather grim. But the stakes are much too high for us not to try.
Here is some further reading on the topic of fake news.
1: Fake News Detection on Social Media: A Data Mining Perspective. https://arxiv.org/pdf/1708.01967.pdf
2: How to model fake news. https://arxiv.org/pdf/1809.00964.pdf
3: Human detection of fake news. http://www.dataskeptic.com/blog/episodes/2018/human-detection-of-fake-news