I Like What You Like: The Impact of Recommendation Engines on Consumer Preferences and Decision Making

By Heather-Destiny Konan 

Image courtesy of Flickr

Imagine a time you heard a song you really enjoyed. Perhaps you were hanging out with friends. Or on your daily commute to work. Maybe, like me, you were finishing up a late night of homework. Whatever the situation, many people can cite a moment in which a random song they had never encountered before became an instant classic for them. And the beauty of it all? These discoveries are never intentional. Gone are the days when we would jump between dozens of radio stations, trying to find the perfect song. Now, not only does the playback function of many devices allow us to skip, shuffle, and repeat, but thanks to the rapid rise of audio streaming and media service in the past decade, the exact song we want to hear is only a search bar away. With 75.5 million Americans subscribed to various music streaming services as of 2020, companies like Spotify and Apple Music engineer the way we listen to music. What’s more, these companies have found a way to optimize the process of finding music, new and old, ushering in the age of digital playlists.

In the early 2000s, before its acquisition by tech giant Google, Songza launched an idea that would put it on the map: automated music curation. Since then, curated playlists have become a given for streaming services. However, as we have progressed toward an increasingly digitized and data-driven society, small start-ups like Songza have been unable to keep up—hence the acquisition by Google. In an effort to match the pace of the digital age, what started as DJs and music experts putting together collections of songs for various occasions has transformed into automatically generated playlists. Bigger services like Spotify now have thousands of pre-made playlists perfectly curated to accommodate every mood and moment imaginable. But how is that even possible? How is it that software can not only produce a customized playlist for me at the tap of a button but also proceed to adapt my music selection to match my taste as it evolves over time?    

The answer lies in a multi-faceted algorithm that integrates various data models whose primary purpose is to analyze a listener’s data against that of others. Spotify, for instance, uses a combination of collaborative filtering, Natural Language Processing (NLP), and audio modeling to survey and compare users’ behavior, song lyrics, and raw song audio, respectively. As a whole, online companies are increasingly employing artificial intelligence (AI) technology and deep learning machines to better cater to individual tastes and preferences. What we recognize as our “Release Radar” playlist on Spotify or our “feed” on Tik Tok is the work of what the digital world calls “recommendation engines”. Whether it be music, movies, or trendy gadgets, consumers are constantly hankering for the next best thing. And companies are aware of this desire. Not only do they understand that catering to our ever-evolving desires is the best way to maximize profit, but they also realize that they must go beyond traditional marketing strategies.   

Image courtesy of Flickr

Recommendation engines are ideal for so many companies since the AI system is designed to mine through user behavior data to make personalized product recommendations to consumers. In fact, there is a high chance that a recommendation engine led you to this very site. And while this digital age has made a global network of information widely accessible, with it comes the burden of decision making. Despite the plausible theory of dual-process thinking that attributes our decision-making to two categories of rationalization—the intuitive System 1 versus the more deliberative System 2—boundless choices have made it harder than ever to logically assess our options. Recommendation engines streamline the decision-making process for us, “[providing] users with a means for managing information in a rational way, thus helping them to overcome the information overload.” However, the data-driven power of an AI system like this is nothing more than a temporary fix. While touted as an innovative response to the widening range of choices,  recommendation engines seem to only perpetuate other issues. The system has the ability to optimize consumer decision-making and buying but only gives people the facade of a choice.   

Contrary to the preliminary assumptions made about recommendation engines at the turn of the decade, the design of the software does not necessarily bring niche products and content to the user's attention. Instead, they push products that are widely circulated among users of a service, thereby augmenting its already substantial popularity. Take the study conducted by Kartik Hosanagar, the John C. Hower Professor of Technology and Digital Business at the University of Pennsylvania. In assessing whether or not recommender systems help consumers discover new products, Hosanagar found that, because common designs for recommendation AI systems rely on sales and ratings data, the common phenomena of “‘the rich get richer’’’ occurs with products that are already popular among users. Therefore, companies like Amazon whose site supposedly aids browsing customers by showing them similar and related products alongside listings are likely only exposing people to other high-selling products in that category as opposed to the myriad of undiscovered products offered. Similarly, the hybrid design of Spotify’s recommendation engine assesses overall user interaction, such as the stream counts of a song, to generate playlists with similar songs on them. They then recommend these playlists to users with similar music preferences. Consequently, the collaborative nature of the data the recommendation engine uses, like Amazon’s software, is based on the songs that circulate the most among Spotify users. Thus, when Spotify suggests your Daily Mix Playlist on any given day, selected songs are featured because enough people with similar taste to you have previously listened and liked them. As a result, niche songs by lesser-known artists remain undiscovered. This creates a perpetual cycle in which users are only being exposed to a fraction of the tens of millions of tracks on the app.   

Image courtesy of PixaHive

With this design, recommendation engines are not accounting for diversity in preference or perspective. More significantly, because the recommendations being made act as driving forces behind many of our decisions, we run into the issue of our individuality as consumers. If an AI system is constantly creating an algorithm of choices for us based on other people’s choices, how do we know that the decisions we make are what we would have originally made if the process was not optimized for us? An angle to consider is that the practicality and convenience of AI predicting people’s preferences undermines consumers who want to explore their identity through preferences and choices. In the past, researchers have highlighted how Netflix’s algorithm, in comparison to other streaming services, tailors their consumer’s preferences and taste to align with Netflix content. Netflix users are not intentional in their choice of entertainment. Instead, they are passively choosing Netflix content recommended to them without the chance to “train their own entertainment preferences or develop cultural tastes and capital”. This only bodes poorly for people who seek diversity in thought through what they consume.   

In addition, the effect recommendation engines may have on our free will in particular, is disconcerting. While in the short term, the recommendation engine takes away the hassle of searching for what we like, it could potentially strip us of our agency in decision-making processes. Recently, research has been done on “learned helplessness” to demonstrate what a lack of self-determination does to consumers. In the most realistic scenarios of the studies, researchers found that marketing algorithms that threaten consumer’s free will impacts them both behaviorally, and psychologically. Essentially, the lack of a sense of personal responsibility have consequences of “reduced helpfulness, higher levels of aggression, and lower levels of self-control in intertemporal choices”. While consumers may not necessarily be aware of these changes, the small sacrifice of autonomy in decision-making, could lead to an even bigger sacrifice in our behavior.   

Recommendation engines, like most AI decision-making software, are encoded to optimize the most ideal output. If that output is always based on popular opinion, we as a society are bound to move towards a single, homogeneous way of thinking. Our choices will only reflect the preferences of the majority. This majority bias could mean anything from a national racial majority to global one, in that alternative societal values and beliefs would be entirely overshadowed by dominant, Western culture. And considering our vulnerability to implicit biases, big data-based recommendation engines cannot and may never be able to control the unintentional consequences encoded biases will have on us. While this does not mean we should ignore Netflix’s homepage suggestions or stop listening to Spotify’s “Mood Booster” playlist, maybe giving the old, antennaed radio a chance every now and then isn’t such a bad idea. 

References 

1. André, Q., Carmon, Z., Wertenbroch, K., Crum, A., Frank, D., Goldstein, W., & Yang, H. (2018). Consumer choice and autonomy in the age of artificial intelligence and big data.
2. Customer Needs and Solutions, 5(1), 28-37. Darlow, A. L., & Sloman, S. A. (2010). Two systems of reasoning: Architecture and relation to emotion. Wiley Interdisciplinary Reviews: Cognitive Science, 1(3), 382-392. 
3. Fields, B. (2011). Contextualize your listening: The playlist as recommendation engine (Doctoral dissertation, Goldsmiths College (University of London)). 
4. Giacaglia, G. (2020, May 14). Spotify’s Recommendation Engine - DataDrivenInvestor. Medium. https://medium.datadriveninvestor.com/behind-spotify-recommendation-engine-a9b5a27a935
5. Hosanagar, K. (2015, December 5). The Real Impact of Recommendation Engines. Knowledge@Wharton. https://knowledge.wharton.upenn.edu/article/recommended-for-you-how-well-does-personalized-marketing-work/ 
6. Montaner, M., López, B., & de la Rosa, J. L. (2003). A Taxonomy of Recommender Agents on the Internet. Artificial Intelligence Review, 19(4), 285–330. https://doi.org/10.1023/a:1022850703159 
7. Sisario, B. (2014, July 2). Google in Deal for Songza, a Music Playlist Service. The New York Times. https://www.nytimes.com/2014/07/02/business/media/google-buys-songza-a-playlist-app-for-any-occasion.html

______________

Heather-Destiny Konan is a third-year undergraduate student at Emory’s College of Arts and Sciences, majoring in Neuroscience and Behavioral Biology with a minor in History. While she had always been interested in neuroscience, it was not until her second year of college, that she was she was first introduced to neuroethics. As a student in Dr. Paul Lennard’s Applied Neuroethics course this past spring semester, she was able to further enrich her knowledge in and engage in discussions on controversial topics in neuroethics. 

Want to cite this post?

Konan, H. (2019). I Like What You Like: The Impact of Recommendation Engines on Consumer Preferences and Decision Making. The Neuroethics Blog. Retrieved on , from http://www.theneuroethicsblog.com/2021/05/ilike-what-you-like-impact-of.html