In one night, Matt Taylor finished Tinder. He ran a script on his computer that automatically swiped right on every profile that fell within his preferences. By the morning, he had swiped through 25,000 people’s profiles.
Nine of those people matched with him, and one of those matches, Cherie, agreed to go on a date. Fortunately Cherie found this story endearing and now they are both happily married. If there is a more efficient use of a dating app, I do not know it.
Taylor clearly did not want to leave anything to chance. Why trust the algorithm to present the right profiles when you can swipe right on everyone? No one will be able to repeat this feat, though, as the app is more secure than it was several years ago and the algorithm has been updated to penalise those who swipe right on everyone. Or so people believe.
For those who might struggle with "packet sniffing" – the means by which Matt gamed Tinder – the tantalising promise that maybe, by putting our faith in an algorithm, an app or website might be able to find the right person is thoroughly appealing.
“It’s something that single people want to exist – it’s the romantic equivalent of an easy weight-loss plan,” says Dr Samantha Joel, assistant professor at Western University in London, Canada. “People want it because meeting one-on-one is exhausting. Like most things that we wish we had, I think it deserves particular scepticism when someone claims they can do it.”
位于加拿大安大略省伦敦的西部大学（Western University）的助理教授乔尔（Samantha Joel）博士说：“这是单身人士想要的，寻找伴侣就像是一份简单的减肥计划。人们希望能这样，因为一对一的见面让人筋疲力尽。这像我们希望拥有的很多东西一样。当有人声称他们可以做到时，人们应该保持怀疑态度。”
Lots of apps and websites claim to be able to use data to sort through profiles for better matches. By completing their personality tests, they say they can save your thumb the effort of swiping. The issue for scientists who might want to investigate their data, and journalists who want to fact-check their claims, is that the algorithms are the intellectual property of these companies, so they are not publicly available. Their entire business is based on developing smart match-making algorithms and keeping their formulas private.
So what do scientists do if they want to investigate predictors of attraction? They make their own.
In one example, Joel and colleagues asked people to complete a questionnaire about themselves and what they were looking for in a partner. Some of the questions were very similar to what you might expect on any dating website, and many more went way beyond. Daters were asked if they agreed with statements from “I’m an upbeat person” to “I worry about being abandoned” and “If I could live my life over, I would change everything”.
In all, they completed more than 100 traits and preferences. Then, after a series of four-minute-long speed dates, they were asked if they had romantic interest in any of the other daters.
Now, the researchers had all three things they needed to be able to predict romantic desire. The first is actor desire, or, on average how much people liked their dates compared to others. This captured how choosy each person was. Did they click with a lot of people or did they find it hard to feel chemistry? By comparing daters to each other on choosiness the researchers could control for people who might make a lot of potential connections mostly because they were quite open-minded about who they would like to date.
Second is partner desire, or, how much did people like you compared to their other dates. The reverse of actor desire, this is a measure of average attractiveness.
By subtracting choosiness and attractiveness from daters’ scores of romantic interest, the researchers had a more accurate measure of compatibility. “Some people are more attractive than others and we can predict who tends to get the most matches,” says Joel. “That is not the goal of these matching websites. They are not saying they will filter your pool so you only have attractive people to choose from.”
Joel found that her algorithm could predict actor desire and partner desire, but not compatibility. Not even a little bit. It could only predict negative percentages of variance – which is like being accurate less than 0% of the time. This might sound like a bit of a head scratcher, but, Joel says that her algorithm would have been better off using mean results for every dater rather than offering a tailored response. “It was completely useless,” says Joel. “It really should have done better.”
“My take is that when two people actually meet they form a shared dynamic that is more than the sum of its parts and cannot be predicted a priori,” says Joel. “Their individual preferences do not make up the substance of what they find attractive. My rating of whether I found you funny after meeting you will predict whether I like you, but my desire for a funny person and your measure of whether you are funny do not because we might not agree on a sense of humour.”
Finding a way to make accurate predictions is not going to be straightforward.
Another team of researchers seem to have successfully predicted romantic desire using an algorithm. Picture a house filled with potential dates. The higher up in the house someone is, the kinder they are. The further towards the back, the funnier. The further to the right, the more physically attractive, and so on until you have collected data on 23 different preferences.
Now, depending on your preferences, you can imagine your perfect partner is standing somewhere near the bathroom sink, for example. There might be other people nearby, who would be nearly as attractive. There might be someone even funnier and more beautiful than them, but a little less kind, stood in another room downstairs.
That is how Dr Daniel Conroy-Beam, an assistant professor from the University of California Santa Barbara, US, describes the algorithm. The distance between a potential partner and your idealised partner in your hypothetical house was the best predictor for attraction.
美国加州大学圣巴巴拉分校（University of California Santa Barbara）助理教授康诺伊-比姆（Daniel Conroy-Beam）博士是这样描述该算法的。在你假想的房子里，潜在伴侣和你理想伴侣之间的距离是最能预测吸引力的指标。
In this particular study the daters were presented with fake profiles of made-up people, not real potential dates. Although, Conroy-Beam points out, people judge online profiles before they have a chance to meet or even talk to their potential dates, so you could consider online profiles hypothetical, up to a point.
Conroy-Beam’s algorithm assumes that all preferences are weighted evenly, which might not be the case. If physical attraction matters much more to you than kindness then perhaps that person waiting downstairs is a better candidate after all. “The next step is to incorporate that weighting,” says Conroy-Beam. “I would be very surprised if weighting didn’t matter.”
Clearly, having a list of preferences makes things complicated. In what order do you rank them? Are your assessments of your qualities the same as mine? All of this makes predicting romantic interest difficult. Perhaps a more straightforward option is to look at deal-breakers – what would rule someone out for you?
In another of Joel’s studies, students were asked what they would consider an absolute deal-breaker in a potential partner – traits like whether they smoke or are particularly religious. Later in the semester they completed a dating profile and sifted through other people’s. After whittling their choices down to a favourite, the researchers offered to swap their contact details. However, at the same time they were shown a bit more information about their chosen partner, which included the fact that they had two deal-breaker qualities.
For 74% of people who thought they might get a real date out of the interaction, the deal-breakers became non-issues. They were prepared to overlook them. Even for people who knew that the date was only hypothetical, 40% still agreed. It turns out, when presented with an opportunity to meet someone who is supposed to be interested in us, we are much more flexible about who we are interested in.
“We wanted them to have some buy in first before we told them about the deal-breakers,” says Joel, “because often deal-breakers show up on the first date or the second or the fifth.” You might not find out that someone is a smoker, or that they have another horrible quality, until you meet in person, or even several dates down the line. We hardly broadcast our less desirable qualities at the first opportunity.
Why might we not strictly observe our deal-breakers? Joel has her own theory: “I think that people just aren't actually very choosy. People feel like they need to be choosy because that is our culture. But realistically people are pretty open to a broad range of partners.”
Putting your faith in an app
If in real life we are much more flexible than we say we are on paper, perhaps being overly fussy about what we’re looking for in someone’s dating profile makes it harder to find the right person. At one end of the online dating spectrum are sites like Match.com and eHarmony who, as part of the registration process, ask users to complete reasonably extensive questionnaires. These sites hope to reduce the amount of sorting the user needs to do by collecting data and filtering their best options.
“We look at core values, we decode those and we match those with people who are as similar as possible,” says Rachael Lloyd, the in-house relationship expert at eHarmony. “From all our years of research, the more you have in common the more likely a relationship is to be a success. We start with 150 questions, although these have changed and been refined over time based on machine learning.”
Lloyd explains that the goal of the eHarmony algorithm is to find ‘satisfying relationships’, which is slightly different to the goal when the company was founded in 2000. Then, marriage was much more important. This shift has reflected the slight change in attitudes over the past two decades.
Researchers from the University of Oxford analysed data from 150,000 of eHarmony’s subscribers and corroborated Joel’s findings on deal-breakers: generally, people are less bothered by things like smoking and drinking than they might predict.
“We also saw that people who are altruistic generally do well,” says Lloyd. “People who have conversations about charity and giving have 34% more interest in them. As our algorithm demonstrates, kindness is still really important. More than being highly sexualised – that tends to not work so well.”
The data also suggests that being very, very attractive as a man offers no advantages over being fairly average. Women like men who rate themselves as five out of 10 as much as men who think they are 10 out of 10s, whereas men would ideally date someone who self-rates their physical appearance as eight out of 10.
At the other end of the spectrum, apps like Tinder and Bumble ask for very little in the way of preferences before they start to show you profiles: usually, the gender of the person you are interested in, an age range and distance from where you live. These apps refine as they learn about the user’s preferences.
“I would argue Tinder is much better because they are showing you people and asking if you like them,” says Joel. “It seems to me based on the data that preliminary filters don’t work.”
“If [online dating sites] are going to match you with someone long term, that requires a lot of long-term data. This claim is exciting to me but to properly test it we would need to follow people for years,” says Joel. “Another possible reason that we might not have found something is that people don’t know what they want. I might not have a lot of insight into what I find attractive and what I am actually like.”
Long term success
We have different sets of preferences depending on whether we are looking for something long-term or short-term, Conroy-Beam says. Generally speaking, when were are only interested in short-term relationships we prioritise physical attraction, whereas for long-term relationships kindness and other signals that someone would be caring are a greater priority.
But, Conroy-Beam says that other preferences also imply whether we are looking for the one, and these preferences can be grouped into sets. So, in theory, you can make “a pretty good guess” whether someone is interested in a meaningful, long-term relationship by looking at what set of traits they are most interested in.
For Lloyd, the data collected from eHarmony’s users suggests that openness is a really important trait for long-term success. “The more genuine you are and confident you are, the better you tend to do,” says Lloyd. “That approach to dating really works. Online dating has given us so many benefits. But it has also created a sense that we are all superficial and shallow. The important thing to stress is that this takes time.”
Perhaps, then, romantic desire cannot be accurately predicted before you have a chance to speak to or meet your potential partners. We are still reliant on being able to pick up on intangible cues from talking to each other, but at least there is some evidence that good guesses can be made about who we might generally be suited to. “What is definitely clear,” says Conroy-Beam, “is that humans make diabolically complicated choices.”