Ask HN: OkCupid, Online Dating, Machine Learning, the Future, Etc. | Hacker News Reader

Ask HN: OkCupid, Online Dating, Machine Learning, the Future, Etc.

8 pointsbaccheion10 comments

How many questions would it theoretically take to have a solidly complete match/enemy percentage? How do you think the questions need to be separated out to get full value from them (raw tolerance builder (bunch of questions to match on to see raw percentage hit or missed), dealbreaker (unacceptable answers filter, flag, significantly "ding", etc), elevators (acceptable answers positively flag, boost, etc))? What other bits of information would help round out, improve, complete, and refine the almighty match percentage? What else can be added to a "fictional" dating site's user experience or "data collection/sharing" ends in order to have the results be even better and more accurate? What was not possible in the past (but was needed) that is doable now?

If machine learning / AI / * neural networks / etc were involved with online dating, then what setup would allow maximum value from the results? That is, should it be used to, for example, just predict user's answers, then use those predicted answers until the user explicitly shares their answer, or should it be used to "sample the infinite" (have users answer questions, interact with the site, etc, such that it can "catch on" to what the user likes/needs and is like), or something else?

If there could be anything, then what should it be? What is needed? What is missing?

I'm really annoyed these sites aren't improving or aren't "up to snuff," but then perhaps they aren't so far from what's required, and it's just minor refinements that are necessary. I don't know-- what's missing from the online dating experience? What's not explicitly needed, but then would be hugely convenient or cool? What else do you think goes into helping infer overall datability/compatibility, and future success? Should there be more explicit sex-related questions in general (and for those looking mainly to date short-term or hookup)? Do the questions general need to be more specific, numerous, and detailed, or more generalized? Etc.

> what's missing from the online dating experience?

There are lots of interesting machine learning or other features that smart people could come up with to better group people who have similar personalities or are interested in common things, but I think they'd all be missing the main point of dating. How much stuff you have in common or whether you share a couple of personality traits are nice to have but really don't matter. It all still comes down to attraction.

Don't take that hipster vegan socialist seriously online who swears they won't date a Trump-loving meat eater. If there's attraction there, they won't care about their nonsensical list of "must-haves".

Rule #1: be attractive. If you think you're not attractive: shut up and lift and work on your life so you get towards making yourself more attractive. Don't let any dang folks spin you the politically correct Disney version of reality that you just haven't met the right person yet or some other nonsense.

From this regard, the core of Tinder is probably the closest to accurately quantifying the real human experience, though the number of spammers and bots on there make it not really worth the time.

As backwards as this comment sounds, it's the truth. You need to be attracted to someone first, and then you maybe share a sense of humour or shared interests (if that). End of the day, you just need someone you can easily spend time with, without ripping each others heads off.

The biggest problem with these sites is that people lie all the time, to themselves and others.

The things that people claim are dealbreaks ("She's a Republican!", "He lives with his parents!", "I would never engage in casual sex!") tend not to be in practice, once the opportunity presents itself.

But when asked point-blank? Oh yeah, I totally couldn't date somebody who supported cruelty to animals in lab test (one sec, gotta adjust my makeup), or who voted conservative (hold on, lemme claim my tax exemptions), or who's a virgin (well, everybody's gotta learn sometime).

There's AI and control theory that seeks to reason about an uncertain world--but in dating, it's active delusion and misinformation. People don't know themselves, and even if they do know themselves, they invent bullshit reasons why obviously good pairings aren't.

That's before you even get into the economics of those sites and why helping people date successfully isn't profitable.

I think being in a relationship requires a kind of giving up of control, a "surrender" to the flow of the relationship, and many people are just too inflexible and want everything to fit perfectly to their pet peeves. So they pass good opportunities while waiting for the "perfect match" which never comes.

That's a large part of my interest in this area. That is, using accumulated knowledge, global awareness, inference, etc to help buffer against lies, lack of self-knowledge, uncertainty, hidden dealbreakers (finding out 5 years down the road about some fundamental thing that ruins the relationship), etc.

Also, it would be cool to be able to understand fully many of the intricacies and nuances of relationships and compatibility, then being able to do some of the heavy lifting using information provided by the user. That is, maybe some things are known to users in general, but they are either too time consuming or too difficult to do, so they end up not doing it (or it's to the point it's clear the site should be taking care of such things).

throwaway420 is correct that attraction will get you more dates.

However, the outcome of a first date with any particular person also seems to have high variability (if you were somehow able to repeat it!). There are so many small factors that make a "great" date, e.g. both people's mood, their openness, the date activity, confidence/actions, other options at the time, etc.

Even if you are both wildly attracted to each other, some people are only looking for the novelty of new people. Others have high standards for a date that are only rarely met. Nowadays, many people line up so many dates that the odds of a single one working out are low. They've lost those "butterflies" that are often needed to even be open enough to match with someone.

So, given all of this variability, there is unlikely to be a "perfect" matching algorithm.

It's called chemistry and no, it usually does not start from apps and models, real life interaction instead.

Some friends I know wrote an algorithm where all it took as inputs were date/time/place of birth. Pretty cool, all the matched it made were matches amenable to both parties. Makes you wonder how much information really is needed and if there is a simpler way around this problem.

After analyzing many OkCupid profiles, it became clear that 16-20% of all available matches are usually good matches (above 75%). That is, if you were paired randomly with people (on the internet) of the gender and age range you were interested in (and that are interested in people your age and gender), then 1 in 5 of them, on average, would be acceptable matches.

This varies wildly per person, but the overall average was surprising. That is, it seems low, but when you think about it, it's actually quite profound. You could randomly pick 5 straight or bisexual women between the ages of 19 and 37 (from among all the women in the world), for example, and it's likely 1 of them would be a good match.

You seem to have the knowledge and skills to pull this thing off. I'd advise against wasting your time doing so. For several reasons:

Taking anything people say in their answers at face value is a mistake at best, naive at worst.

These are not technical specs. These are people describing themselves publicly. They'll most likely be inaccurate either by design or unconsciously and rightly so. Yes, rightly so. You can have a PhD woman who drinks wine in her bath tub while she reads Sartre and dissecting Stravinsky who'll be wanting to get her head banged against the head board at that particular moment. Do we think she'll put that thought on her profile? No. She'll refrain from saying that because society likes clear cut labels and will reduce that complex human being to one tiny aspect of her humanity.

That's one reason.

The other reason building all that AI for "online dating" isn't a worthy endeavor is because the point of online is discovery and initial contact. That's it.

Discovery is when you notice a human being you didn't know before. Initial contact is exactly what it says: making contact.

Once you discover their existence, you make contact. You chat for a bit and hopefully you like each other to a certain degree. Then you go to another platform, namely Skype, and have a chat to get a feel of each other's vibe, voice, body, etc. Yes, people are willing to get naked but only for people who're used to get other people naked.. One of those "free for those who can afford it" things that are thrown at those who aren't starving for it.

But maybe they're listening to some music they find cool (sometimes really cool music) and they want you to listen with them.

Then from Skype to exchanging phone numbers. Skype and not Facebook because Skype is a better firewall than Facebook. It's easier to eliminate someone from your life after a Skype chat than after a Facebook one because if you stumble on an unstable person, now they have access to your friends and your real name and can stalk you or cause a scene. Also, Skype is more discreet. It avoids them the questions from other people about you. Maybe it's too early to parade you in front of their friends. Maybe they're afraid their friend steals you. Maybe they're engaged. Facebook adds degrees of freedom people would do without at this stage.

After exchanging phone numbers, you meet. People decide if they want to have sex with another person during discovery and contact and if they're going to have sex with another person during a Skype session. In other words, you can in most cases directly meet somewhere suitable for this (taking a room) for you both want each other.

And there you have it, the beginning of a relationship.

From this, you see that the actual time spent with each person on the discovery/contact platform is tiny and doesn't justify building "an AI" for it because it'd be like building "an AI" for football locker rooms when the game is played on football fields. A better example would be building an AI to handle powering up a computer and finding the best ways to push the power button. The problems in software don't exist because people don't know how to turn on a computer, it what happens next.

Similarly, the problems in "dating" don't exist because people don't "find" other people.. It's because they lack the skills to take it from there to add to their lives and to the lives of those they meet.

As I said, you seem like a qualified person. If I had your skills, I wouldn't be doing that.

Reply on news.ycombinator.com