Online Dating: Match Made in the Cloud

Written by Sara Tähtinen

Can AI revolutize online dating? Probably not, but it can make it more tolerable if implemented correctly.

The usage of online dating is increasing. According to a recent study the share of heterosexual couples meeting online rose from 2% to 39% between the years 1995 and 2017. For same-sex couples the number is even higher: in 2017 65% of couples met online. At the same time another study shows that Americans who have used some dating site or app in the past year report more often feelings of frustration (45%) than hope (28%). Could AI help us to make the experience more pleasant?

The data is abundant, at first look

In the era of Big Data we have gotten used to asking AI for help. Imagine this: you decide it’s time to get settled and you have a rough idea what you are looking for. Then you sign up to a dating site, upload some images, answer some questions and “Tadaa”: AI makes a perfect list of people that are searching for the same things as you are. Nice, clean and simple. Or is it?

First, let us take a usual machine learning approach. Dating sites always ask you to fill out some information. You tell your age, gender, sexual orientation and home town. Then you select some pictures of yourself (in the best possible angle of course!) and download them into the dating site. Perhaps you write a short bio telling who you are and what you are looking for. You might also be encouraged to share where you fall on the grid regarding your political, religious or other worldviews.

Then you select your “Looking for” criteria. How much older or younger partner you could consider? How far away can they live? Any additional preferences? If that’s all, you are finally ready to go into the next phase: browsing people’s profiles and giving out some love. Take out the popcorn!

The dating app will also keep record of your likes, matches and messages. From messages it can probably predict if the discussion is going well or not. Using a lot of smileys, writing back multiple times per day or sharing a mobile phone number can be considered as good signs! Questions left unanswered, very long answering times and short replies probably hint on one-sided crush. So it looks like we have a nice amount of inputs and we have a good idea of preferable output. Why not just feed these in some nice algorithm that could suggest us even better candidates?

There’s a catch: the quality of a machine learning algorithm depends greatly on the quality of the input data. And the data in online dating sites is not actually as accurate as we might think.

Digging into the user data

The world of online dating is led by impression management. To improve your luck it’s attempting to slip a little bit of lies here and there. And if you don’t lie per se, you select very carefully what information you share. Everybody knows social media: the life we show online is the idealised version of the truth. What would be the harm of it?

According to studies deception is very common amongst the daters. In fact, there’s even a term for people that seriously misrepresented themselves in the dating app. “Kittenfishers” often use heavily edited or decade old pictures of themselves, or they might lie about their age, lifestyle or interests. The goal is to attract more people and increase the amount of first dates by inventing a better version of yourself.

Unfortunately, trust is the most important element for a succesful, long-lasting relationship. Lying in your profile risks putting people off as the expectations don’t match the reality. Moreover it’s frustrating for the online daters to meet up with a person that barely resembles the version they played online. But for data science perspective this becomes a problem too: how could we use this data to train an algorithm to improve matchmaking if we cannot even be sure which parts are true and which ones are not?

It might also be that users are worried about the privacy and security of the data that they post online, and rightfully so. Would you like to share your first name, age, gender, education level, occupation, religion, your location and where you live with a site if you cannot be sure that the data is 100% safe? Nowadays it’s also very hard to avoid recognition even if you post very little about yourself in a dating app. Let’s face it, “Tindstagramming” – where a person that you rejected in a dating app (often Tinder), approaches you on other platforms by a private message (often Instagram) – is very creepy! But if you don’t share enough details about yourself, you diminish the app’s means of providing you with good matches. Not such an easy decision then!

The other problem is that in the dating world people don’t really know what they want. The decisions who get likes in the app are often based on looks and superficial criteria. For short-time relationships this is perhaps an understandable approach. But for long-time relationships softer qualities, like kindness, becomes more desirable. However, the dating apps can collect information only from the very beginning of the relationship, so there’s no way of gathering data on what makes long-term relationships successful. The risk is that AI aided apps will optimize the suggestions to maximize the number of first dates or short-time relationships. It could be biased and unfair towards a person who is looking for a long-time relationship. So as said: today’s dating sites are only as good as the data they’re given.

So where could AI help? 

Some of the dating apps (like Tinder, Hinge, OkCupid) have already taken steps towards using AI for advanced matching. The aim is to reduce your need to scroll through endless profiles by showing you first the most promising candidates. And if you are not sure what you are looking for, there’s even an app for that: The Artificially Intelligent Matchmaker. It starts a conversation with you about your preferences and gradually narrows down to the profile of your dream-come-true date. Very practical!

There is additional development done around making the apps more safe by identifying fake profiles and “catfishers”. Tinder uses AI to compare profile pictures  with the user’s real-time selfies to make sure the profile is not fake. So definitely a nice bonus from AI!

Tinder also uses AWS image recognition software to recognize what is happening in the user’s pictures. Then it uses text recognition to pick keywords from your text and matches this with the image categories. It is an interesting approach, but apparently the users of Tinder are a bit sceptical about the benefit of this. If you are looking for an athlete, this may work. But if you write in your bio that “Allergic to cats”, the algorithm might simply pick the word “cat” and start suggesting you persons that pose with a cat. Yikes!

Another use of AI is abolishing harassment or use of abusive language. OkCupid, for instance, uses machine learning to support the moderation team and flag potentially harmful language. Tinder has just released a feature that prompts the user with “does this bother you?”, hence leaving it up to the use to decide what is ok and what is not.  

But at the same time it’s important to keep in mind the pitfalls – the quality of the data being just one of the points to consider. Looking at the use of AI from an ethical perspective, lack of transparency and bias in how AI is making decisions are something to keep in mind. Biases in the data get inherited to the algorithm very easily, and thus concerns for racial bias and such have been raised. An ethical AI should alsol explain how and why the decisions are made, and ideally the user should have the possibility of disagreeing with the decision made by the machine. But as the algorithms are business assets, companies don’t want to reveal too many details of how they work. This leaves users puzzled at times when Tinder ‘shadowbans’ them without telling them the reason: the user can still carry on using the app without knowing that their profile is hidden for others.

And how about the phenomenon where people program chatbots to have those initial conversations with potential dates, to save time? Interaction anxiety is one of the problems in online dating, but isn’t it a bit too creepy that a bot can make the first move towards your potential date?

Should you trust AI to find you a Valentine? 

AI can surely improve some aspects of online dating, eg by identifying fake profiles and abusive language and eliminating those from the experience. And if you believe using shallow metrics to find a bit of fun, why not! 

But we would apply caution before having AI help with anything more serious. Things that are easy for AI to interpret from the data available may matter in short term relationships, but when looking for true love, it gets more complicated. AI has also tried to predict what people find desirable but there seems to be no alternative to the good old chemistry when people actually meet!

Only one question remains. AI – are you my Valentine?

 

 

 

Sara Tähtinen is a Data Scientist who is passionate about theoretical particle physics and her two cats.

Phone photo by Pratik Gupta
Heart photo by Photo by Alexander Sinn
Cat photo by Ramiz Dedaković