The title reminds me of the amazing, adorable CLRS Introduction to Algorithms, but trust me you do not need a degree in CS to continue reading the post. All you need is the love for design.
I recently happened to chance upon an article that discussed how Donald Trump won because of “Facebook” . The article talked how social media influenced the recently concluded US elections. And the main case in point here was the recommender systems that popularized the juicy untrue stories of Trump more than the true stories.
Before going into the main point of the post, let’s discuss two terms that were highlights of the post – Clickbait posts and Recommender systems. Have you ever read such a post in your Facebook timeline? “She was cornered by four black clad people and what happened next will shake you!”. Or “Previously unknown facts about Brangelina breakup, #14 will shock you!”. And falling to their neat little trap, you find that #14 was Brad left Jennifer for Angelina or the people were her sons who were there to surprise her. These types of posts are called as clickbait posts, where the user is lured into the article with fancy titles and false promises and rewarded with mediocrity or in many cases false stories.
The second term we are gonna discuss is “Recommender Systems”. So, I recently came across an article on “Beauty and the Beast” and saved it to my Pocket account. Pocket instantly suggested me stories on Frozen, The Lion King etc. Similarly, I was browsing through some bohemian neckpieces on Amazon last week and now there is a section on the site which recommends me similar pieces. You would have also noticed, once you click on an article on Facebook, you find many more similar posts on your timeline the next time. All these are examples of some type of recommender systems. Recommendation algorithms learn from your click behavior what kind of articles you like to browse and finds items from the web of a similar topic. While Amazon was doing it internally to show its products, Facebook was relying on the world wide web for the articles.
So how does this tie up to the US elections story? In the article, Max Read suggests that once people clicked on an untrue story of Hillary Clinton being criticized, many more stories (usually fake) of similar content popped up on your timeline. Similarly, articles that praised Donald Trump kept cropping up on your timeline and very slowly the voters’ mindset were being made. Even if the voters did not outright believe the stories, the cropping up of multiple similar stories tend to provide weight and some stamp of genuineness to the content. About a hundred fake clickbait stories are written in a day to attract traffic to their site or promote a specific agenda . Adding in the recommender system of Facebook that shows your other fake stories when one of them is clicked and you have some pretty good traffic.
With the last couple of posts we have come upon the realization that every object around you has an inherent design. Take a spork, a bag, an app, a programming language or a chair. There is a design behind the way the product looks, the way the product works and even the way the product fails. There is a design behind every workflow you encounter everyday. Similarly, any algorithm in any of the systems have a design on how they work. Recommendation algorithms at present rely on the main context of the article – the US elections, the main participant of the article – Donald Trump, the main sentiment of the article – good and powerful. Or Hillary Clinton in the US elections and the sentiment being deceitful. So once you read a couple of such articles, you are bombarded with similar articles. Many of the articles in the references were not found by me, but by Pocket for me once I read Max Read’s article. Such a classic example of recommendation algorithms. With these ideas floating around and gaining popularity, even FB, fake news writers have had a sudden slap to their face .
This brings us to the point of the post. The evolution of design. A spork was designed when we realized it was not sensible to use a spoon and fork to devour a simple waffle or pancake. A combination of the two made much more sense and hence the design evolved. What one has to understand is that systems are not black boxes and algorithms are not the magic spell that runs the box. They all follow a design. Algorithms are designed to provide results given some inputs. They are designed in such a way that they are time efficient. But, they evolve as well. Now, the time has arrived to rethink the design of some algorithms prevalent today. The main results that are expected out of current recommender systems are just to increase engagement in terms of time spent on the site, number of times an advertisement or paid post is clicked, number of times it is shared etc etc. The monetary aspect of the output drives the design of the algorithm and now we are seeing the consequences of these greedy methods. Talking about recommender algorithms, we should not let popularity decide the importance of an article (i.e. if a story appears on your Facebook newsfeed or Instagram timeline) and the let content lead the matrix. The algorithms can be tweaked to include analysis of the comments, or well known news sources should be given a higher weight over a lesser known source/ blog post. Assign importance weights unequally and you popularize the reliable posts and lower the spread of clickbait. We need to start spending time on the user needs and choices without relying on the popularity meter. It’s time to think outside the box of algorithms written about a decade ago and move forward keeping the current use cases in mind.
Everyone in the CS world are familiar with the century old conundrum – Security vs Usability. Do we make it more secure with a ton of security features, or do we make it usable and easily accessible for the users. The problem has been studied so much in the past and continues to be a VERY popular topic. But, now we are faced with a new dilemma, Privacy vs Genuineness. Facebook cannot guarantee fact-checking with their intricate and confusing sharing and privacy settings. And thus begins the ultimate fight between the two. Considering more than 40% of people rely on Facebook for news these days, this is going to prove to be a very very hard problem to solve and it’s time for change. It’s time that algorithms developed in pre social media era be reevaluated, rethought and redesigned. Because, #designMatters.
That was some serious talk today!
Articles mentioned in the article