Why making recommender systems transparent won’t stop FAKE NEWS
Recent story “As Germans Seek News, YouTube Delivers Far-Right Tirade” investigates the role of YouTube recommender system in spreading fake news and radicalizing society.
Recommender systems are helping people not only to discover new content, but also stay in their information bubble. Here is how recommender systems work, so by recommending content previously visited by similar people (in terms of reading habits), one can easily discover content shared within an information bubble, but it is hard to find news written from different perspective for different audience.
How we read and search for news
It is a human naturel to construct theories and models of the world that explain everything that we see and read. When there are more news supporting the evidence, we are fine and enjoy reading them — but when news present something inconsistent with the model then it is easier to not to accept the news (it might be fake) than to think about it and redefine (complexify) our theory how world around us works.
Thousands years of human evolution support the fact that people generally prefer simple models of the world over more complex ones. People even tend to accept very strange dogmatic (but simple) explanations of how the world around us works instead of being humble enough to accept that explanations are beyond our reach. In this environment, black and white perception of the world around us and simple explanations win over reality and more complex views of the world.
Even when you are strong enough to be able to read about some event from different perspectives, willing to understand motives of participants and accept reality in its complex shape, it is often hard and almost impossible because of lack of data.
Where there is no objective reporting and diverse news from different perspectives, there is nothing to recommend and even the best recommender won’t help you to discover true story. One of the example might be the story of Ryszard Siwiec who sacrificed himself in protest of Soviet invasion to Czechoslovakia in 1968.
Although his act was captured by a motion picture camera, Polish press omitted any mention of the incident. A story invented on the spot was that it was an accident caused by drinking vodka and smoking, or spontaneously combusting, aiming to divert interest to gossip and portray Siwiec as an irresponsible drunk. He was also declared mentally ill.
Therefore it is important to support independent journalism so journalists can spend time investigating true reality and report from their perspective.
The economy behind news
Income from printed news decreased significantly over last decade. Media houses invent new models to finance quality journalism. Subscription based model is on rise, but still vast majority of people is not willing to pay subscription to a single medium for quality news and reporting. Micro-payments for single stories are also not very popular so far.
Majority of online news portals are financed from advertisement on their pages. Advertisers often pay for an impression so portals are motivated to increase number of page views regardless the time spent on page. This favours cheap content such as galleries of images, quizzes and short stories with appealing headlines over more quality in depth reporting. Due to widespread and invasive advertisements many visitors started to use adblock leading to further decrease of advertising revenues. Media houses often responded by adding more video content, where ads are more difficult to block and their per-view-revenue is much higher, so they can use it to subsidy quality news and articles. The situation can differ, because there are several types of news producers.
- Public independent media: Publicly funded media that can afford to publish quality content even without advertisement or other income channels.
- Privately funded quality media houses: Private media houses are often owned and financed by media moguls and reporting can be slightly biased in their interests, but it is hard to tell. Some private investors just like to support independent journalism and some just make some profit. Most private media houses are pushed into compromise between producing expensive quality news and cheap entertaining articles or product placement articles generating much higher advertising revenues.
- Cheap news portals: Some of the private media houses discovered the model of cheap news. They sacrificed quality reporting and produce fast news (often adopted from unverified source). They typically have army of external contributors and no editors to guarantee quality and verify credibility of information from more sources. This is the environment where fake news can flourish mixed with half truths and adopted news.
- FAKE NEWS outlets: Some news portals took it to the extreme — they even fabricate fake news to keep visitors engaged to grow audience and generate revenue from advertisement or to spread propaganda.
- Individuals and hobby journalists: It is increasingly easy to build your own audience and use platforms such as Medium to publish news/articles free of charge or even get paid when your audience is large.
In this environment with millions of news published everyday, there are personalized news aggregators such as news360 or big platforms such as Twitter and Facebook that use powerful recommender engines helping people access news from alternative sources.
It is increasingly hard for readers to find quality content and to tell if the content is credible and if it was verified. One must admit that many of them do not even care.
Influencing recommender systems to boost quality content
How quality journalism gems can find their way to the audience when there is so much content with more appealing headlines? What happens when we boost the probability of “right” news to be recommended?
In Recombee we work with media houses to help them personalize their content, improve reader experience to complete with big platforms where personalization is a standard. Apart of personalization, we give them power to bias recommendations towards quality content. Typically, authors and human editors can select up to N articles they are proud of and wish to promote. Then we increase the probability that these articles are being recommended by boosting. This simple feature helped editors to accept machine generated news, because they feel they are not being replaced by AI, but they can leverage AI to generate personalized news reflecting their views.
For big platforms and news aggregators, the situation is much worse than that for media houses which can guarantee quality of their news and articles being recommended. Third party fake news can spread easily through the platform or aggregator and recommender systems help to find the audience that is most likely to read and believe such news.
To suppress fake news in the platform, Facebook, for example, can employ independent journalists-editors that will select articles to be boosted and recommended to bigger audience. And flag or filter-out articles containing false information. But there is a dark side — such human editing is quite costly and editors can be biased. Even if AI detects fake news, it can be biased due to labeling bias in the training data. So influencing recommender systems can be dangerous.
Every government present their versions of reality and more power they have to filter out inconvenient content, the higher is the chance that people support their fight against that enemy on the other side — independent nature of youtube recommender here helps people to find information that official media try to suppress. Controlling recommender engines can help totalitarian governments filter out inconvenient news and suppress any opposition.
Is there any reasonable way how to fight fake news then?
Here is my recipe:
- Education to critical thinking — teach people how to read news, recognize fakes, how to live outside information bubbles and to be open to alternative views.
- Supporting independent high quality journalism — every democracy should have budget for journalists cultivating the society.
- Public news aggregators supporting critical thinking, fake news reporting — for those interested in balanced news, there should be public news aggregators powered by advanced recommender systems capable of identifying and presenting alternative views and penalizing fake news reported by independent editors.
- Encouraging big platforms to detect and flag fake news in collaboration with local independent journalists and fast checkers.
