How a Story About Using Algorithms to Find Love Puts Technology in the Proper Perspective


Reverse Engineering Relationships

The rise of multiple choice format in testing and market research is one example of the influence of technology in our practices—multiple choice rose in popularity in the mid-20th century when we devised ways to use scanners and data-processing to check the results.

As in many uses of technology to accelerate problem solving through data mining, we want to be careful not to confuse speed with accuracy, to be aware that our biases and assumptions might be giving us exactly what we're looking for rather than what we need.

In the summer of 2012 Chris McKinlay was going through his sixth year of a math Ph.D. program at UCLA and he was going through a sort of monastic phase. He had a foam pad he would roll out under his desk to sleep and he would go to the UCLA gym to take a shower. Every night, he would log onto this supercomputer called Yellowstone at the National Center for Atmospheric Research, which at the time was in the top twenty powerful supercomputers in the world.

He could only use it at night when he would run these distributed data mining algorithms he was working on for his dissertation. When his code was running or compiling in the cluster, Chris had a lot of down time to kill. Reddit and OKCupid became regular destinations. One night he came across this post on OKCupid blog, the largest free dating match website in the the world. The post says, “OKCupid is no more responsible for your match percentages than Microsoft Excel is responsible for your net worth.”

He took that as an invitation to program his way into a relationship and decided he was going to reverse-engineer a match on the system. The site has 14 million monthly active users and it generates from zero and a 100 percentage matching points between two users. All the activity and visibility on the site are based on this match percentage.

The math behind the percentage is based on multiple choice questions. There are about a half million of them, all user submitted to cover all walks of life, including dating history, lifestyle, religion, and so on. Chris wondered how the answers were grouped on the site. Was the process based on common belief systems? And if so, how much? To find out, he started using all his supercomputer power to download and analyze matching data.

Chris discovered that when people in Lost Angeles go online and answer data matching questions, they don't do it at random. They cluster rather tightly into these seven groups. One of the groups seemed datable. His next step was writing commentorial optimization software to figure out which of the half million questions he could answer truthfully but still maintain the highest possible percentage match with every single person at the cluster.

He wrote the code and ran it on the supercomputer for a while. Until it spit back the right questions to use and he entered them into his profile and then things blew up. All of a sudden, he became the top match for over 30,000 women. That meant that every time one of those women went to check OKCupid, he was up there at the top of their searches, their news feeds.

He was still this ratty looking student with a bad profile getting hundreds of profile views every day. So he decided to further optimize every last word and pixel to convert these page views into email messages. The result was 88 unsolicited messages per week. To put this into perspective, says Chris, we should consider the median message of straight males on OKCupid per week—zero.

That's how during the summer of 2012, he became the most popular of the seven million profiles on the site. What was happening was like Amazon Prime for his dating life. He remembers worrying whether OKCupid would notice and shut him down. Given the conversion rate and his new sense of urgency, Chris decided his second “equally hard core” phase of this project—he decided to go on at least one date per day until he either got shut down, or he met someone worth shutting the program down for.

His first date after making this decision was with a high-powered West Hollywood lawyer driving a convertible Mercedes who wanted to go to an expensive restaurant with him who had showered at the UCLA gym and left early to jump start his car. Over dinner he wondered about the woman's motivation when she asked whether they could be soulmates—she'd been on OKCupid since 2006 and she's never seen a 100 percent match before.

That freaked him out. But it didn't stop him. He continued going on dates—dinners, concerts, bike trips, even one funeral. He was trying to have these romantic storybook dates and it was killing him, not to mention putting more than 1,000 miles on his car. He was burning through time and money and it was clearly not scalable the way he wanted it to be.

Which led to the question, “What is the smallest, socially acceptable interaction I can have around about three minutes of face time? Because that is all I really needed to know to learn if I wanted to continue and I haven't really been able to write software to tell me that information.”

Then he devised a way to make the date part the shortest possible and to cluster dates. A small cup of decaf coffee consumed standing up would do. He found four coffee shops kiddy corner from one other so he could do one coffee date each hour on the hour. The common frame of the coffee date brought out the similarity of what attracted these dates, including their coffee order.

Given his propensity for optimizing, Chris got very good at these dates. He took the conversion rate from first date to email asking for a second date from 50 to 65 to 70 to 82 percent. But the problem was he wasn't going on many second dates because at any one time, he had six or seven first dates lined up. He was saying no to people he would have been excited to meet only months before—he had become desensitized.

Also, because of the high match, the women were bringing their “A-game” to these dates, had high expectations, and he was becoming disappointing to them. To avoid writing hard emails, he began trying to discourage them intentionally during the date. At first he would just tell the truth, “I've got three other dates lined up today.” It didn't work, cutting the email for second date only by 12 percent. But not more than that.

Chris adopted more extreme measures, like going to the rest room and putting high-liner on when the date was not going well from his perspective. Clearly something had gotten twisted and he began to question his motives, “are you truly trying to find someone or are you trying to optimize some kind of game from the sociological side, or is it the performance arts side.” He thought about gathering more data, refactoring the code and thought very seriously about shutting it down.

But he didn't because at that point he was going viral on OKCupid and placed on front of enough eyeballs that he was getting messages from all over the world—Argentina, Australia, all over LA and people who didn't belong to the group he had optimized for, so every once in a while on these dates, he would have a total surprise. Which he loved.

On date number 88, he met this woman names Christine who decided to amp up her profile after seeing the position of his—she had put new pictures on and optimized her profile. She was very different from the other dates. Chris decided to tell her what he had being doing.

“I basically did the same thing, I actually did much worse. I used some topic-mining algorithm to bottle up these things and reverse-engineered the match algorithm and I've been going on an average of 1.1 dates per day for the last 88 days, and I'm getting really tired and I'm confused. But you seem really cool. Is that twisted?”

She took a beat and said, “no, that's kind of what it's like being a woman and dating.” They had a twenty-minute discussion about the politics of gender and data. It felt like a good date.

When Chris tells this story, which was a couple of years ago, before he graduated, people experience “buyer's remorse” because they met their significant other in a bar or through friends and ask if he thinks they stopped dating too soon. They wonder if it worked out for him, if he met someone, or if it fizzled after putting in all that work.

Christine and Chris got engaged, so it did work. But it's not because he hacked OKCupid. They are engaged because they worked really hard at their relationship. Relationships, at least good ones, are the kind of things you cannot really get from Amazon Prime. He usually adds that he's fairly certain he conducted a thorough search.

People are much more complicated than their profiles. We all get more of what we want when we learn that technology cannot give us all the answers, especially since we know how to game the system. When we integrate technology with being in the conversation, we figure out the better question we were trying to answer in the first place.

 

[image and story via The Moth]