Algorithmic Online Dating and the Paradox of Choice
“Computer dating is fine, if you’re a computer.” — Rita Mae Brown
Dan Ariely at TED
Dan Areily is a brilliant new professor at Duke. He has thought deeply about classical questions in psychology [1] and formulated his own solutions and tools [2]. He gets invitations to deliver keynote speeches in front of VPs of marketing of major retailers like Target and he presents his ideas at TED [3].
I really like professor Ariely’s method of thinking. He has a profound understanding of statistical analysis and statistical significance. For example he points out that focus groups are not as useful as companies think [4], simply because the data you collect from your observation do not translate to valuable information. He is absolutely correct, with focus groups you cannot make a claim unless you have shown (from your data) that your claim is statistically significant.
A friend of mine forwarded Dan Ariely’s recent interview, in which he explains why online dating is so unsatisfying. It is a very thought provoking interview and I suggest you watch the video before continuing.
“There’s very little advice in men’s magazines, because men think, I know what I’m doing. Just show me somebody naked.”
— Jerry Seinfeld
The online dating industry is even bigger than the porn industry [5], and more than 30 percent of first dates end up in coitus [5] yet its customers are much less satisfied than the customers of the porn industry. Let me break it down for you, those who go out, partake in a social activity and perhaps later on, sleep with a living human being feel more miserable than those who look at naked women, with fake breasts, online, all day, and alone…. But why?
I am sure everybody in the field of computer science has also thought about making better dating websites. After all we all have friends who are single, good looking and perfectly qualified why can’t they just start seeing each other?
I, as well, have been thinking about this problem in the context of recommendation systems. Once you start reading the literature you soon realize that nobody has really figured out a solution for online dating yet.Allow me to provide you with an example: Randy Farmer and Bryce Glass have a fine book entitled “Online Reputation Systems” in which they talk about two reputations systems that they designed for Yahoo. they use similar methodologies for designing reputation systems for both websites (Flickr and Yahoo personals, which was Yahoo’s dating website but I guess they now use the engine of match.com). The Flickr system works great but the yahoo personal system, was not that successful! A solution that works for photos fails for humans.
So how can we write a code that finds people’s “the one” automatically and algorithmically.
What is wrong with current dating websites?
Here is my $0.02 on this issue. I beilieve the problem is not really our algorithms, it is not our number crunching computers. The problem is how we present our recommendations. I’d like to argue that the problem is the paradox of choice.
Barry Schwartz
The paradox of choice is professor Barry Schwartz’s thesis [see this video]. It basically says if you need a jar of jam and you go to a supermarket that has 3 types of jams you’ll leave much happier than when you go to a supermarket that has 30 different types of jams on the shelf. This is a beautiful dilemma, in the first case your chance of buying the best jam available is 33 percent and even if you fail to chose the best jam there is 66 percent chance that you will get at least the second best jam or better. In the second case the likelihood of buying the best available jam is only 3 percent. So you’ll be less happier knowing that you have probably selected the jam that is not the best.
Now back to online dating. Most online dating websites show you a lot of different options, they say “look we got 100 gorgeous ladies for you”. And perhaps you think “wow they really have a good collection that satisfies my taste” one of the ladies has a PhD, the other is wealthy, the otherone is athletic and pretty. you search for a couple of hours, message a bunch of them, perhaps stalk the top 50 and at the end of the day you end up having a date with another random woman who was not even in the list of top 100 gorgeouse ladies that the website recommended to you. Who can be satisfied with this date now?
Similarly, in TV shows such as “The bachelor [6]”, we have never seen a successful marriage even though the bachelor selects from a large pool of qualified bachelorettes [see this news article for example].
So what’s the solution?
“My philosophy of dating is to just fart right away.”
— Jenny McCarthy
[I am still thinking about a solution but in case you want to see a half-done solution please keep on reading]
The problem is that assortative mating is a very complex game theoretical problem. If you have watched the movie, beautiful mind, you probably remember that John Nash (Russell Crowe) talks about the best strategy to get the blond lady! Now add 5 million more online single guys to the pool of competitors and you got yourself an unsolvable game theoretical problem. (and let me remind you that you’ll never get that blond girl, and you’ll always remember that your current wife or girlfriend is not the hottest one)
For a long time online dating websites have been highlighting the good features of candidates, but I think the best solution is for the algorithm to point out the flaws of hottest girls/guys directly to the customer (but just if the customer is not among the top attractive people himself). Perheps to a lady customer that is ranked 5 out of 10 on the social scale (not attractive) it can suggest:
- This guy is handsome but he pees all over the toilet seat
- This gentleman makes six figure salary but he is rude to his girlfriend, farts at home and snores in bed
- This guy has a PhD in computer science, works in silicon valley but is socially retarded and wears striped shirts in parties
- .
- .
- and as we go down the list
- .
- .
- This guy looks relatively ok, makes good amount of money and has a college degree, want to meet him?
The algorithm is not changed but I believe people will be happier if they understand that the grass on the otherside is greener simply because of the smelly fertilizers!
Open Source Research 
Hi Mark,
Lots of interesting ideas.
First, I really like the idea of showing people the downsides of individuals and not just the positive traits. The question of course is how you get this data and how to get people to present it on an honest way (there used to be a site where old girlfriends wrote about the men they dates, so maybe this is one approach).
But the more difficult question in my mind is how do we present people online in a way that is informative for “consuming” people in the physical world. What would thee attributes be and how would we get the information about these?
Irrationally yours
Dan
These are some really interesting points. I’d make a few comments back:
(1) I think your introductory statistics are a little suspect. The valuations of the porn industry are notoriously unreliable (http://www.forbes.com/2001/05/25/0524porn.html) and I suspect that the valuations of the online dating industry also call for skepticism. And the comment about coitus seems like a very large extrapolation from a line, “One out of three women who meet men online have sex on the first encounter,” that’s really ambiguous and odd. (E.g. which first encounter? All first encounters?) My intuition about people tells me that this claim is a little off.
(2) I’m not sure why showing negative traits about people would solve the paradox of choice, which presumably results from excessive choice set size in the example you cite. I think showing negative traits, if it could reliably implemented (cf. Dan’s ideas above), might add some value, but only because it might shift the type of choice framing that often makes online dating problematic. The 100 Beautiful Ladies situation you describe seems so problematic because it promotes a type of choosiness that’s totally counter-productive to falling in love with someone. In real life, you never decide between 100 people and say, “which of these do I like the best?” You meet a single real person by accident in some random social setting, and you only ask yourself, “do I like this person?” In real life, we’re not solving an optimization problem, but simply a threshold problem: is this person worth dating at all? I think the optimization approach leads people to discard lots of viable options and invest too little emotional energy into those that could turn out well if a little effort and patience were invested.
With that said, do you think the ideal approach is to mix in negative information, or basically replace positive information with only negative information? Also, why focus on universally negative traits, rather than traits that might be negative for some people, but really positive for others? My impression from most people’s profiles is that they totally avoid any information that’s even conceivably negative, which makes everyone sound the same. For example, I’m vegetarian and generally prefer dating vegetarians. Most people would avoid admitting something like that, since it’s likely to lead to you losing a few dates, but might also lead to higher quality dates from the people who passed through the filter. Is there a way to encourage that sort of directness?
Just a short note: If you fail to chose the best jam there is a 50% chance to get the second-best. Without knowing whether you failed to get the best one, it‘s 66%.
WRT Dan’s comment: a post-date questionnaire would be a great source of honest information, especially if 30% of first dates involve sex. If dating sites are statistics games anyway, they may as well be populated with lots of useful, observed data.
Are striped shirts bad? That would explain a lot.
Bill R
Software Engineer
Maine
I’d go on a date with Bill R cause what he said was funny.
Blonde Girl
NYC
When you go to a porn site to buy porn, you pull out your credit card, do the transaction, and reliably receive porn. Whether receiving porn is or ought to be “satisfying” is not the issue. In order to earn the low satisfaction levels that typify online dating, a porn merchant would have to do business by taking your credit card info, charging you a monthly fee, and explicitly making no promises whatsoever about whether any actual porn was going to be forthcoming.
Online dating has a ROI profile for the customer that, in any other business besides casino gambling, would result in fraud charges against the merchant. It’s no wonder customers aren’t satisfied.
Bring back the days when dating was simple. Strike up a conversation with some good looking dude on the subway……
“This guy has a PhD in computer science, works in silicon valley but is socially retarded and wears striped shirts in parties”
Is it bad that I actually think that’s cute/hot/sexy in some ways? I mean, I don’t see anything wrong with this guy. I’d date him. ❤
The 3 jam example is entertaining but not very accurate in a world where almost every choice you make is between a huge number of alternatives. Having three glasses of jam to choose from doesn’t mean they’re actually good and little satisfaction is derived from the knowledge that you get “the best sample from this local supermarket” if it ends up tasting bad. I’d rather have 3 million different jams to choose from and going through a great number of them over time actually sounds like fun.
As someone who has flat-out no dating success, neither in the real world nor online, I don’t believe excessive alternatives are the problem with online dating. Instead I think people are frustrated because, much like in real life, the game is rigged in a number of ways:
People go to online dating sites because they can’t find partners in real-life interactions. What they don’t realize is the same reasons behind the dilemma also apply online: Looks still matter tremendously online, social awkwardness still comes across online and people you meet online are still lying jerks, no matter how often you write on your profile page that you don’t want anything to do with them.
“Alpha” men still send a huge number of cloned emails indiscriminately to any women they find, much like they flirt with everything that moves in the real world. That means if you’re just some guy who instead sends out a personal message once in a while, you still don’t get a shot, your signal is simply drowned out by all the other voices in the same way you’re overpowered daily by good-looking loudmouth alpha men in real life.
People who are successful finding partners in real life will also have success online (which is probably exactly the same 30% crowd that manages to actually have sex after an online-induced date), but they’ll still be just as unhappy about their partners as before. The rest of us, who are simply too flawed and too picky to get a date, also remain unhappy because the same rules apply online that prevent us from having success offline.