The echo chamber. An enclosed space where all I hear is a reverberation of sound. This is what media has become. A personalized space where I am repeatedly being fed the information that enforces my view of the world. This individualized media bubble is tailored by both myself (who chooses the social media accounts to follow) and the AI systems (that enforce and support my beliefs). Instead of empowering me to be more informed and open minded, media urges me to develop stronger, more close minded opinions about my existing beliefs. I often see examples of how toxic this paradigm can be.
Here is a screenshot from the Wall Street Journals’ “Blue Feed, Red Feed,” feature which puts liberal and conservative Facebook feeds side by side.
Inspired by the same event, these two articles present entirely different stories on opposite ends of the spectrum. This is incredibly dangerous. Facebook’s AI will only show the right wing news sources to conservative readers and will rarely, if ever, show these stories to liberal readers. It becomes effortless for a Republican to rally behind the Conservative Tribune and even more so for a Democrat to back the Daily Kos. Because of this, users will listen and share these reports, polarizing the two sides even more.
The global implications of this are immense. Echo chambers have created what is colloquially called a post-truth world, where the line between opinions and facts is blurred. Political leaders can reinforce any agenda that is supported by their polarized news publisher (on both ends of the spectrum). This allows severe scientific issues like climate change to get tangled up in political rhetoric, leaving over 30% of Americans refuting human-induced climate change against nearly all scientific evidence. When a person’s only source of information is primarily concerned with clicks and emotional appeals rather than truth, he/she can believe anything.
Often times for news sources like Facebook and Apple News, the more ridiculous the headline, the better. This is solely because more clicks and engagement translate to more ad money for them. Naturally, media companies are responding by adjusting a portion of their headlines in order to get more views. These inflammatory headlines are now the new form of click bait.
In a world with polarized political echo chambers, this type of clickbait thrives. Because of aggregators, the more extreme the headline, the more likely it will be seen. Many media companies push content that drives clicks because that is the metric they need to increase. However, clicks are only describing one time views, and are not an accurate description of a customer’s loyalty to a publication. Oftentimes, it places the article’s quality and a user’s satisfaction on the back burner.
Currently, business metrics come before user metrics. A strategy that works in the short-term but will backfire in the future. As consumers continue to get smarter they become more and more aware of click bait. Soon, everyone will be sick of the bullshit that is being thrown their way.
So change it. And change it now.
Here is a screenshot from All Sides, a news source that was created as a tool for users to escape echo chambers and understand all perspectives of the same story.
All Sides is an excellent tool for the people that know it exists and that are willing to put in the work to read multiple perspectives, but echo chambers are still very influential in each individual's collection of news.
An obvious proposal to breaking down echo chambers is ending personalization. But what kind of world would that be? A world where YouTube is just a random database of videos? A world where spotify cannot create personalized playlists? A world where a newsletter is the same for everyone on the subscription? That’s no world for me.
Why not use AI to create more informed users rather than more ignorant ones? Of the hundreds of daily articles published, I should see those that I’m topically interested in. Instead of pushing me articles that align with my political views, show me environmental and economics articles with different political biases. It should be easy for me to find interesting articles -- like New York throwing away a lawsuit against companies using fossil fuels and wildfires raging in the Arctic Circle -- without siloing me from breaking news. News should be a place of discovery.
Publications should be on a mission to empower all people to find quality news. They should be placing user value before business value. By focusing on user experience you advance your company passed short term trends like clickbait while indirectly increasing the metrics that correspond to reader value, such as retention rate, Daily Active Users (DAU), and session length.
Looking at the most recent Trump and Putin dilemma, publications and news sources need to put quality first. Extremist articles that are displayed in the WSJ’s “Blue Feed, Red Feed,” should be headlines that are difficult to find. Quality articles that properly represent a story, such as the ones displayed in All Sides, should be the ones pushed to the consumer. These are the articles that empower the user to learn and develop more educated opinions about the topics they care about.
Even though I am blaming business for the existing problems, it is not completely their fault. Until recently, the tech to create the optimal personalized experience has not existed. Historically, there are two types of ML systems: content based and collaborative.
Content based ML only uses topic to come up with recommendations. Since it does not use metrics to measure quality, the ML cannot grab a more semantic understanding of the article. It just treats each article as a bunch of words which usually results in poor quality suggestions.
Collaborative ML tries to solve the problems of content based ML by taking actions into account. Generally, clicks is used, which is a terrible representative metric of the user experience for the reasons listed above. The system does not think long-term, and when left alone, it creates echo chambers and click bait in an attempt to optimize for clicks.
A hybrid approach can use the benefits of each method to solve the presented issues. Until recently, there had not been a sufficient amount of machine learning research to do this well. With all of the information, and clever modeling by the ML scientist, the model can be made to stop thinking short term and instead focus on user satisfaction, maintaining serendipity, and breaking echo chambers.
Through hybrid machine learning, you can use both content and user focused metrics such as read time, and click/read ratio to make recommendations that ensure your users are not just here to pay a visit, but instead are here to stay.
Just as AI is the cause of the creation and polarization of echo chambers, it is the only solution that is truly capable of tearing them down.
At Triton, our vision is to do just that. To learn more, send us an email at simba at triton dot cloud or visit our website at triton.ml.