How YouTube is Addictive — Recommendation Systems & its Impacts
Each day users of YouTube watch over a 1 billion hours of video. That is over a million years of life per day, everyday. For many of us, YouTube has become somewhat of a daily ritual. We use it as a means to entertain ourselves throughout the day, learn something new and exciting and follow the lives and stories of content creators we have grown to like. Recently, the content YouTube recommends us at any given moment generally captures our interests very well. We have started to click impulsively on what the suggestions both quite old and new YouTube provides. This article will explore how the YouTube recommendation algorithm works, the implications of such a system and what the goals are that YouTube aims to accomplish by saturating our lives with accurately recommended content.
YouTube’s Recommender System
YouTube has a problem that only a few platforms share; how to provide personalized content to its billions of users with a limited amount of control over the dynamic and quickly growing upload count. The problem can be summarized to 3 categories; scale (how to integrate a model on a large platform), freshness ( how can it predict with new and relevant content) and noise (selecting the appropriate features from the data). Traditional recommendation algorithms tend to sacrifice predictive success and ability for scale in dynamic environments. Given modern distributed computing, a neural network architecture modification to these traditional algorithms (logistic regression and collaborative filtering) improves on the underlying problems. That is why YouTube has integrated a two-stage deep learning model for recommending its videos. The model is represented below:
It wouldn’t be surprising to say that YouTube is built on Google Brain. This allows for it to scale its model. Then the problem is split into finding good candidate videos and ranking those candidates. Videos with larger watch time and length are prioritized. The model for finding the candidates uses collaborative filtering with features such as searches, watch time and user history. These candidates are put into a list and assigned a score for which a regression algorithm chooses the highest ranked videos. The videos with the highest probability of being chosen are the recommended ones. The model is then tested for accuracy by an iterative A/B testing process with real users. As a result, YouTube is able to provide relevant content.
What I find quite interesting is the simplicity of this process. What YouTube is employing is not novel in the recommendation space. Perhaps, YouTube is an example of how a simple model that is well designed can have more predictive success then poorly designed new and complex models.
There is a lot more to building the model that is not covered here. I did not cover how the scores and weights are assigned, what features the algorithm uses and its ontology. I would suggest anyone who is interested to read the paper linked below.
A complete explanation of how the recommendation system works can be found here: https://storage.googleapis.com/pub-tools-public-publication-data/pdf/45530.pdf
How YouTube Videos Affect Us
Now that we have a general understanding of how the recommendation model work we need to understand how YouTube videos themselves can become so addictive.
To identify if something is addictive users must exhibit symptoms of addiction. Common symptoms of addiction are; an inability to stop using, negative health effects, obsessive behavior and it is used to cope with outside problems. YouTube could be addictive if the content provides enough stimulus for the users to exhibit addiction symptoms and a neurological reaction that is consistent with addiction. What sort of content can be addictive? Content containing information, and especially a lot of it. YouTube has a lot of information. Having access to relevant information can improve decision making. This is why new and relevant information are rewards for our brain. Rewards are treated in the mid brain with a dopamine response just as an intake of foods high in sugar, fat and salt is. This is both true for the consumption, but also for the anticipation of such a reward.
YouTube as a platform fulfills those requirements for addiction. The videos provide us with relevant information that stimulate a dopamine response. This process is constantly reinforced by consistently supplying us with more appropriately recommended videos. YouTube can also modify our behavior by carefully selecting certain videos by the means of our own perceived autonomy. What I would I also propose is YouTube allows viewers to engage intimately and anonymously with many content creators without being judged or feeling obliged to participate. This sense of connection, engagement with many communities, the ability to be anonymous and expressive with a stream of constantly relevant information and our recommendations being chosen for may serve as platform to develop an addiction.
There are occurrences of people developing some sort of an information addiction that is consistent with the symptoms described here. The PBS recently published an article in which they describe a young girl who became an obsessive user of the platform that indirectly resulted in her being hospitalized and admitted for rehabilitation. The girl, in search of a path to popularity, began watching YouTube videos in hopes of connecting with popular students who would often speak of videos they had watched. What was just entertainment, quickly became an obsession for her. She would spend hours isolated in her room. When problems arose in her life she used YouTube as source of escape or to find a solution. Problems in her life built up, and this ultimately ended with her attempt to commit suicide.
As of yet, there are no direct mental disorders listed in the Diagnostic and Statistical Manual of Mental Disorders (DSM) for information addiction.
Source for ‘Midbrain Dopamine Neurons Signal Preference for Advance Information about Upcoming Rewards’ : https://www.cell.com/neuron/fulltext/S0896-6273(09)00462-0
Source of PBS Article: https://www.pbs.org/newshour/health/compulsively-watching-youtube-teenage-girl-lands-rehab-digital-addiction
The Attention Economy
YouTube as a platform and the videos that are so easily accessible creating an environment in which can stimulate us in such a way that is consistent with addiction. These problems speak to a larger issue that is present with many large platforms using information and data as a commodity. The pressure for media companies to gain as much engagement for as long as possible is what is known as the Attention Economy.
Your attention is scarce and is also desirable. Large platforms compete for attention as that allows them to generate revenue through advertisements, the selling of data, goods and services. Large platforms like Google, Facebook operate as a medium for others to create, buy and sell. This can be simply selling you a funny video, or a tangible commodity such as merchandise. For these companies to thrive they need to not only provide a well designed and working platform, but must habituate you to and ultimately make you dependent on the platform. Sites like YouTube do this so well as there no other website quite like it and the cost for creating a directly competitive platform is too high. You can learn something new with visual representation, engage with creators, get the latest news etc… all on one free and easily accessible website.
Many argue that the attention economy keeps us from doing the work that gives us meaning and a sense of purpose. The constant distracting notifications and endless carefully selected content controls us by manipulating our examined behavior. The consequence of this is that we never have to fully embrace our loneliness, sadness and poor sense of self but we also feel pushed to engage with things, that upon reflection, we should not have.
Conclusion
YouTube has become somewhat of an addictive platform for many of us. It captivates our attention so well and recommends good content as too keep up revenue and gather as much user data. What I hope to have shown here is that from the ground up, YouTube can be addictive. This is evident given the pressure of the attention economy, its recommendation system, and a unique set of features in which its users show some signs that are consistent with the psychological and neurological responses of addiction.
YouTube is still a wonderful platform that many of us are grateful for using. It gave rise to celebrities in which can work independently of any restrictive studio, it provides us with a place to satisfy our interests, to get creative and and learn something new. However, we should also be aware of the pressures YouTube as an organization faces, how they recommend us content and what the implications of using YouTube are so as to use the platform in a way that enriches our lives.
If you enjoyed the read and found it helpful, please give it a clap and share it with others! Thank you :)
