One of the more unsettling developments in recent months has been the phenomenon of ‘fake news’, where audiences consume and share news stories on social media which are not factually correct. The US election represented a peak in the dissemination of fake news, but the problem still persists today, in part because fake news stories make controversial claims that generate ‘hits’ and ad revenue, thus creating a financial incentive to produce more fake news.
The worrying implication of fake news is that it promotes false beliefs among members of society, which could result in bad decisions being made because they are based on fiction rather than fact. Fortunately, computing experts believe that AI technology holds the key to combating the fake news phenomenon, especially through machine learning and natural language processing.
It is possible to train a machine to detect fake news through machine learning. By exposing a machine to thousands of fake news articles, as well as thousands of real news articles, the machine can learn patterns from which to distinguish fake from real news 1. For instance, fake news may have more sensational headlines. Using the knowledge it has learned, the machine is then able to detect whether a future news article is fake, and will continue to update its ability to predict fake news based on new input.
Natural language processing
Another way to detect fake news is through natural language processing. For example, a machine could analyze the rhetorical structure of a news article, including the central argument, the supporting evidence of the argument, and the tone of the language, to determine whether the article is making a logical argument supported by evidence or is instead making spurious and illogical claims 2.
But AI is not invincible
Internet giants such as Facebook are developing the kinds of AI technologies described above to combat fake news 3. However, while AI can help curb the rise of fake news, it is not invincible. It may occasionally make mistakes, known as Type I (false negative) and Type II (false positive) errors. A false negative would result in a machine labelling a fake news article as true, and a false positive would result in a machine labelling a true news article as fake. Realistically, AI should minimize such errors, but cannot eliminate them entirely, especially if fake news producers discern ways to manipulate the machine into making mistakes (such as by altering the structure of their articles).
The dilemma of combating fake news may be analogous to blocking spam email. Over the years, new ways have been developed to block spam, but these have only resulted in new ways to get around the blocking, meaning that spam is here to stay, even if it’s not as prevalent as it once was. The same situation may end up being the case for fake news.