Let’s be honest, what do you think of when you read this word?
A romantic comedy from the early 2000s? A product line of Rituals? The song from the popular K-pop band BTS? One of those trendy words that are all over inspirational Pinterest boards, but no one really knows what it means? Or do you immediately think of Alexander Fleming, whose controlled sloppiness led to the world-changing discovery of penicillin?
Serendipity is notoriously one of the vaguest words in the English language, and that's one of the biggest hurdles to "design for serendipity”. When people argue that algorithms should be designed for serendipity, what does this even mean?
Of course, you may wonder why one would want to design for serendipity?
Serendipitous experiences play a crucial role in shaping key moments in our lives. The renowned psychologist Albert Bandura recognized the importance of these chance encounters in his social cognitive theory. Like many of us, Albert’s life was marked by serendipity. On a Sunday afternoon, bored by a reading assignment, instead of finishing up his schoolwork, the usually always conscientious Albert Bandura went to play some golf with a friend. Due to a stroke of luck, they ended up playing behind two attractive women who played slower than them. So, they joined them in a foursome of golf. It was during this fortuitous Sunday that Albert met the wife of his dreams with whom he married and had two daughters. Many examples of such life transforming events which develop haphazardly can be found in people’s biographies. Think about it yourself? How did you come across the important people in your life? How did you find yourself in the job of your dreams or pursuing your favorite hobby? Chances are that serendipity played a crucial role!
In the context of recommender systems, the concept of serendipity has experienced a surge of interest thanks to Eli Pariser's filter bubble hypothesis. According to Pariser, a filter bubble occurs when algorithmic filtering and personalization results in people only consuming content that confirms their beliefs, aligns with their tastes and supports their perspectives. An extreme example of people living in a filter bubble are flat earthers. Filter bubbles capitalize on a common human bias, namely confirmation bias: people pay more attention and more readily accept information which aligns with what they already believe. Although there is much to say about this (highly debated) hypothesis, serendipity is often promoted as a means to burst these filter bubbles. In this understanding, serendipity is all about discovery, and hence the idea is that increasing serendipity will broaden peoples’ horizons.
But what does it mean to design recommender systems to burst filter bubbles? Is any design for serendipity one that would burst filter bubbles? And what does it even mean to “broaden one’s horizon”? Even though serendipity is generally seen as a positive and enriching phenomenon in our everyday life, and something that technology should cultivate rather than demolish, when it comes to designing technologies for serendipity numerous challenges remain.
This vagueness not only results in a design challenge, but also represents a fundamental threat to the concept itself. Those vague concepts that everyone endorses but no one really seems to know how to put into practice (let alone define) often end up on a pile with other initially promising but ultimately mere buzzwords. Is serendipity doomed to become the eco-label of algorithms? A characteristic that everyone values but due to the ambiguity of the concept and the wide range of (questionable) implementations, no one really knows what it really means?
We don't think so.
Serendipity Engine aims to move beyond this utterly vague word and turn it into a concept that is actionable within a specific context. Actionable, meaning that the systems that are being built can support users to experience serendipity within this application domain.
Will our findings be applicable to all contexts? Not at all.
What will be applicable to other application domains is our approach, our understanding of how to translate particular design intents into design decisions and evaluation metrics for serendipity.
That’s right, we are building a serendipity engine, but it is up to each and every one to decide how it should be implemented to suit your goal. Our blueprint can inspire others to design for serendipity themselves, but we caution against simply copying it. Think of it like this: Each engine in a car is designed specifically for the purpose the developers have in mind with the car. Similarly, serendipity engines are customized to fit the specific context in which they are applied. They are not one-size-fits-all, but instead tailored to the situation they are meant to serve.
How will we do this?
In terms of the conceptual and theoretical work, we will largely build upon the doctoral research of Annelien Smets, who has – together with the support of a great bunch of other researchers – developed a framework to study and design for serendipity within specific contexts.
The central thesis of this argument is to let go of the quest to find a one-size-fits-all definition of serendipity.
Rather, one needs to acknowledge the subjective and ontogenetic nature of serendipity, and the various ways in which it manifests itself.
In practice, this means that for every use case and application domain, we start by investigating what serendipity means in this particular context. Indeed, it is not hard to imagine that serendipity might mean something different when it is about recommendations for culture activities compared to serendipitous encounters to foster innovation in science.
Hence, our first question: We want to know what serendipity means to you. We want to know what it means to citizens, but also to policy makers, software developers or civil society. While this seems like a simple question, it's one we ask far too infrequently. Especially when it comes to serendipity in an urban context. Once we know what it means and why we want to design for it, we can – together with the other research groups in this project – start looking into how we can design for serendipity.
Why does this all matter?
Society, and the lives of the people living in it are increasingly mediated by recommender systems. Indeed, we book a bed and breakfast recommended by Airbnb, we listen to the music recommended by Spotify and we eat in restaurants recommended by TripAdvisor to name only a few ways in which recommender systems influence our daily decisions. The intended purpose of these recommender systems is to help us in making better decisions by filtering out the most relevant options and in this way saving us from the overwhelming number of choices available.
However, this power to shape our decisions also comes with great responsibility. Indeed, recommender systems play a key role in determining what content we are exposed to. Journalists have a similar responsibility: when they write news articles, they also determine what content we are exposed to. Just as we expect and train journalists to act ethically, we should expect the same from recommender systems. Therefore, incorporating serendipity into the logic of recommender systems contributes to their responsible use.
The project’s aim is to provide insights into
How to identify what serendipity means in your specific application domain
How to design your systems to foster such serendipitous encounters
How to evaluate if users indeed experience serendipity
Does this spark your interest?
We hope to meet you along our way, serendipitously or meticulously planned. You might increase the chances of staying informed about the Serendipity Engine project by subscribing to our newsletter or connecting on LinkedIn.
Any questions in the meantime? Feel free to contact us at email@example.com.
This post is written bij SMIT (Vrije Universiteit Brussel). SMIT is mainly responsible for the work package on Value of Serendipity and the overall coordinator of the SBO project. SMIT's main researchers on this project are Brett Binst and dr. Annelien Smets.
This text is intended for a general audience; if you are looking for a more in-depth discussion of some of the concepts, please refer to the linked academic publications. You may also want to check out the publications of our research team.