Bots Discover Tinder’s Algorithmic Scoring Arena

Everyone is constantly categorized by algorithms, but we don’t often question this process of categorization and what kind of biases are infused within these decisions. Dutch Artist Mark van Koningsveld’s work Hokjebots (2018) uses Tinder Bots to discover how people respond to the ways in which they are categorized and aestheticized by Tinder. As an anthropologist, Mark’s artistic research methods have intrigued me since we first met. His daring and critical use of digital tools and platforms seem to explore digital culture phenomena in ways that digital anthropologists still struggle with. Throughout various interviews and projects, I’ve gotten to know Mark and followed the development of his Hokjebots work. This written piece is an extension of our conversations about Tinder’s politics and aesthetics.

Mark van Koningsveld showcasing his work to some audience members during the HELLO WORLD! opening night.

Hokjesbots (2018) is Mark van Koningsveld’s newest work. Intrigued by the ways in which algorithms make suggestions to us, he has trained an AI to tamper with the algorithms used in the Tinder app. Mark developed a framework for recognizing types of people on the basis of their photos. The AI looks at the profiles of users using image recognition software and attempts to put them in one of six basic categories: Duckface meisjes (Duckface girls), Basic Bitches, Dansjes-en-drankjes meisjes (Party good vibes girls), Tjappies (Trashy guys), Normaaltjes (Booring Guys), and Mooiboys (Self-Loving Guys). If the bot is matched with somebody, they will ask them personal questions to see if the bot put them in the right ‘category’. Mark states that he wanted his Tinder bots to explore the digital infrastructure of Tinder, and how people would react once confronted with its categorization process. Algorithms can affect whether we get a job interview or how long we go to jail in some places, yet the processes by which they arrive to these decisions are often not transparent.

Mark’s Tinder bots all have different ‘personalities’ and preferences when it comes to their swiping strategies. The ‘Anna’ bot only swipes on guys who can help her fix her car, so she is looking for guys that are outdoorsy, or that can work with cars. ‘Becca’ is just looking for a normal guy; a guy who just wants to go on vacation with her and who is in a group photo on his profile picture. ‘Chloe’ is only on Tinder because of Instagram, and is looking for a guy that wants to party. ‘Chris’ is looking for girls whose profile has beach photos, or maybe a techno party, or club photo. ‘Mark’ even has subscriptions for some of his profiles that allow them more swipes a day.

“when you are liked by somebody who has lots more likes than you (…) you improve your score. If you do not like them, then your score improves even more. This creates a kind of arena where people are battling to get a higher score.”

Mark’s Tinder bots did not only try to categorize people in order to find their romantic partners, but also had chat conversations with the ones that they matched with. In these conversations, Mark is also trying to find out if the bot categorized them ‘correctly’. The user is then also confronted by the bot’s assumptions in their chat conversations.

Above are screenshots of the conversations that Mark’s bots had on Tinder with the profiles they matched with. In these conversations, the bots attempt to find out if their ‘categorization’ of the users is correct; asking them about their hobbies and how they define themselves.

Mark discovered throughout his research that Tinder scores profiles, and that this affects the ways in which users are categorized and paired, all without users knowing about these processes. He noted that: “when you are liked by somebody who has lots more likes than you, you also go higher, so you improve your score. If you do not like them, then your score improves even more. This creates a kind of arena where people are battling to get a higher score.” Furthermore, Tinder does not only score profiles, it also encourages a certain aestheticization of the way in which users present themselves on the app. For example, Tinder is constantly tracking their ‘success rate’ as some users swipe right on certain photos and not others. It then makes suggestions on what photos are more attractive or what order they should be in. Mark states that he found there to be many clichés on Tinder, ‘like people taking pictures with dogs, people taking pictures on vacations or in, like, ties and suits’. He states: “I found very interesting that these are, kind of like, archetypes coming together in a strange way, that after a couple of slides you get kind of the same person”.

We can all recognize the existence of these certain ‘archetypes’: the fit girl, the traveller, the self-obsessed selfie taker, the sarcastic topless guy… Yet, we don’t often think about why these specific archetypes emerge and how Tinder’s algorithms use these photos to make assumptions about who we are and what we like. Mark likens Tinder to a bar, where people present themselves in a performative manner: “So, people try to be adventurous, or try to show that they have a nice car or something else. A lot of things that I have seen come back is that they also want to be seen as a good person, doing volunteer work (…) Those couple of things that people just put in their profiles become their own templates after a while, and people also are starting to copy it”.

“You can tick all the boxes, but you can still be a very different person from the one that the algorithm thinks.”

As Mark explains, despite the ‘efficiency’ and ‘accuracy’ of these algorithms, they still rely only on data that Tinder thinks is important. There are many ways you can be put into a wrong category by accident or in the right category for the wrong reasons. Mark van Koningsveld talks about the ways in which algorithms can be wrong ‘for different reasons’ — because of a simple glitch but also a miss-categorization based on what the algorithm perceives to be the ‘right’ input.Its discriminatory algorithms are one of the main things that Mark wanted to highlight for users of the app. For example, Mark mentions that his profiles have been picked up as belonging to certain categories simply because of the lighting in a room, or because an object looks like a car. The image recognition software picks up on the ‘right’ variables in this case, but makes the ‘wrong’ conclusion: “I have one profile that looks like for people who like to party, but if you have different color lights in your room, you also think you are at a party. So, in these ways you can be put in groups that you may look like you belong to, but you do not actually belong to. And that’s also what happens a lot also with algorithm that work on where you live, or how old you are. You can tick all the boxes. but you can still be a very different person from the one that the algorithm thinks.”

About HELLO WORLD! and Mark van Koningsveld:

Mark’s work Hokjebots (2018) was part of Creative Coding Utrecht’s HELLO WORLD! exhibition, which took place in Utrecht from 2–4 November 2018. Hello World! Is the first overview exhibition of Creative Coding Utrecht and presents the work of the creative coding Utrecht community: artists, designers and hackers who question and redesign the world using smart technology, code, machine learning and generative design. The exhibition presents artworks by Frederik Vanhoutte, Saskia Freeke, Sylvain Vriens, Jasper van Loenen, Bram Snijders & Carolien Teunisse, Cristina Cochior & Ruben van de Ven, Art van Triest, Sabrina Verhage, Wouter Willebrands, Ren Yuan, Mark van Koningsveld, Vera van de Seyp & Dora Kerekes.

To read more articles about the exhibition and other artists that partook in it, visit CCU’s Medium page.

Following his studies in liberal arts and sciences at UniC Utrecht, Mark van Koningsveld is now a Master student in Communications and Multimedia Design at Leiden University. While exploring his research on the social implications of apps and algorithms, Mark is collaborating with Sensor Lab and completing his internship at WAAG and SETUP MediaLab Utrecht, where he is playing with mobile applications and Tinder algorithms to create interactive works and challenge audience expectations.