So, how did you two meet? “Online”. It used to be that couples would grudgingly confess this. Nearly a third of recent married couples, though, is good company. The stigma is falling away. The answer should also be changing because the truth is, increasingly, “we were introduced through a robot.”
Well, not quite a robot, but the machine intelligence resident on sites like eHarmony and OKCupid. What could be wrong with that kind of algorithmic match? We recently caught up with an expert on the subject, Manshu Agarwal. Agarwal is the CEO of dating start-up Spritzr and, more important, a true network thinker. He tapped James Fowler, a UCSD professor and co-author of the book Connected, for advice on how to design a dating app for emergence. Agarwal, a Cambridge-educated ex-strategy consultant and biotech executive, took us on a guided tour of the dating app ecosystem.
The takeaway: Daters beware. Robots can’t replace our social networks when it comes to matchmaking. This lesson has implications beyond dating. It gets to the nature of human interaction with artificial intelligence.
Robots can’t replace our social networks when it comes to matchmaking.
First, though, let’s give robots a fair hearing as matchmakers. To do so, we revisit Nobel Prize-winning economist Daniel Kahneman, the father of behavioral economics. Kahneman’s work revolves around the idea that our intuition is a minefield of potential errors. We should avoid making life-altering decisions based on impulse bubbling up from our subconscious. Instead, we should use our conscious minds to weigh such things as the match between a potential mate and our stated preferences.
Kahneman’s hyper-rational prescription is like an advertisement for the ‘matching algorithms’ used by dating sites. Why not abandon intuition and let the algorithms do the analytical heavy lifting? Not only are these algorithms more dispassionate, but they can also sift through thousands of potential suitors in an instant. Our ‘dating productivity’ rockets as a result.
Faster search, more candidates, better matches. What could be wrong with that? Loads, answers Agarwal. His criticism of machine dating intelligence is grounded in science: not surprising, coming from a dating CEO with a Master’s in Chemical Engineering.
First off, Agarwal argues, dating site algorithms suffer from poor design. The fact that they are touted so aggressively amounts to a scam. Agarwal peeled off a series of flaws, and cited research by Northwestern Psychologist Eli Finkel to back him up. In fact, Finkel’s 2012 paper on the subject, while a bit dated now, reads like a damning indictment. For instance, checking out algorithmically delivered profiles is obviously very different than actually meeting in person. It’s not clear that humans are any good at ‘swiping’ their way into relationships.
Agarwal also questions whether dating sites are more productive. The average dater writes 50 messages to get one response, and that just starts the process of getting to a date. So algorithmic matching falls short of its principal claim: raising our dating productivity. The fact that 22% of all current couples met online is itself misleading. It doesn’t tell you whether those same people might have met someone more compatible (and faster) through a non-algorithmic search.
The fact that 22% of all current couples met online is itself misleading.
Beneath the design of current algorithms, though, lies a greater truth: Relationship matches are a complex network phenomenon. They obviously involve feedback travelling between the two daters. The network connections don’t stop there. A potential mate joins your social circle. Their reactions, your mate’s, and his circle’s counter-reactions, all influence how the relationship will work out. The amount of information exchanged through the network is vast. Agarwal points to things like facial expressions, intonations, mannerisms — all occurring as emergent behavior in response to feedback. On top of that, there is path dependence: Early on, leaving dirty dishes in the sink can spark a fight that scuttles the whole thing.
Humans process that vast amount of data at the subconscious level, and it is almost impossible to say how we arrive at lasting attraction. Romeo could ‘count the ways’ in which he loved Juliet, but perhaps Shakespeare, a genius observer of human behavior, meant this more as poetry than social psychology.
Agarwal has an answer to all this complexity: Use technology to solve for it the human way. His app allows users to import their Facebook contacts and through an easy swipe, invite them to contact and meet. The direct ‘user’ in Spritzr’s case is not the would-be dater. Rather his friends and relatives all serve as potential matchmakers. Agarwal recounts how the idea was born from his own experience of friends wanting to eliminate the odd chair at a dinner party. As his friends found mates, they had a strong incentive to help him link up as well. This is like a network cascade effect, a tidal pull that motivates constant matchmaking.
Our average social circle extends to 500 friends, but we only know 50 of these very well.
We asked Agarwal the obvious question: Don’t people go to dating apps to broaden their search beyond their social circle? Yes. People do want novelty; they want to encounter potential mates they don’t already know. This is where James Fowler’s research came in. Our average social circle extends to 500 friends, but we only know 50 of these very well. It is this 50 that may ‘trap’ us in dates with existing acquaintances. The Spritzr app allows the other 450 to easily set us up, even though we may not consider them close friends. Those 450 friends have a total of 10,000 friends. These ‘friends of friends’ can still process information about you and your social circle, and output a match that incorporates it. Agarwal says his biggest surprise is that even friends three times removed make decent matchmakers: Their invitations to meet a potential date have a surprisingly high influence.
At this point, the interview zoomed out to cover more philosophical ground. If one thinks about it, social networks are like biological algorithms. They ‘process’ complex information about our compatibility as mates, our status within the network, our personalities; and they output matches for us. These matches create new ties within the network, strengthening and perpetuating it. Agarwal calls this these ties the network’s ‘glue.’ From an evolutionary standpoint, matchmaking plays a part in the network’s continuous self-assembly. It is part of what makes human sociability such a great adaptation.
Agarwal backed up our 50,000-foot perspective with some data from Spritzr usage. Not surprisingly, married women win the top matchmaking spot. Second place though, was a shocker: single men. Agarwal believes single men play matchmaker because they want others to reciprocate. Their best path to finding a date is to help friends find theirs — you know, wingmen. The comparison to Tinder ‘hookup’ culture is obvious. Seeking mates within a circle of friends, single men bind themselves more closely to the network. Going outside that network, they shift towards a transactional mode that weakens our social fabric.
It is the nature of complex systems that one cannot lop off a part and still get the same behavior from the remainder?
Spritzr is growing fast, but from a small base. It is a social David to the algorithmic Goliath. Humans are detaching from their social networks in droves in search of computational matches. There may be some tipping point out there…It is the nature of complex systems that one cannot lop off a part and still get the same behavior from the remainder? How will our overall social behavior, even that beyond dating, change as a result?
Perhaps, though, there is still hope. The work of AI researchers like Jeff Clune and Kenneth Stanley, highlighted elsewhere in this issue, points the way. In the future, maybe neural nets will evolve to recognize the value of us dating within our social network. They might start, mysteriously, to flash us profiles of friends of friends. They might even resort to deception to engineer a first date between seemingly incompatible humans. When queried as to how they came up with such a match, the algorithms might, computationally, smirk. Another algorithm could write a clever story about the happy couple’s ironic algorithmic origin, or even a play: Shakespearean AI.
That future is still far off. Meanwhile, not just daters, but also society: Beware. It pays for us to be meta-cognitive of how our interaction with machine intelligence affects our interaction with each other. That awareness says head for your nearest human matchmaker — preferably one empowered by an app.
