5 ms·

It's no wonder; these apps aren't delivering on the actual value proposition of dating apps: finding great matches. Instead they trap people on the platform "sw

by jey 2y ago

It's no wonder; these apps aren't delivering on the actual value proposition of dating apps: finding great matches. Instead they trap people on the platform "swiping" endlessly, where a right swipe doesn't even mean real interest. How would a dating app work if it actually worked in the user's favor? It would focus on finding great mutual matches that are personalized to exactly who you are and what you're looking for. But this is a tricky problem because there's two competing criteria: 1) we want to give each user the ability to express their personal preferences and match with the people they're most excited about, but also 2) we need to prevent spam and encourage compatible matches, otherwise some people will get inundated while others will be left shouting into the void with no reply. This is something I've been obsessed with figuring out, and it turns out there's an elegant solution drawing on mechanism design (a subfield of game theory), combined with cutting-edge preference learning algorithms that are able to learn your personal tastes with high accuracy. My cofounders and I are currently building a new app called Rarebird based on these insights. We're currently looking for a Creative Director (Lead Designer) and an ML R&D Engineer. See more details here, including a deck and technical overview: https://docs.google.com/document/d/e/2PACX-1vRfo082XRQWUfIwQibk03vW6i7tF5ySW_gYqyaDCA-b8JovhOrUeDB37VLNFQQ1YbRqrFWcYpidIheC/pub https://docs.google.com/document/d/e/2PACX-1vRfo082XRQWUfIwQ... Really, there should only be one dating app if it actually worked. What has stronger network effects than a dating app? The only reason we see such fragmentation is because the current apps don't actually deliver better matches as the population increases, instead what users experience is congestion as there's more and more irrelevant potential matches on the platform. Which is why most new apps just try to carve out a niche segment of the population, e.g. based on religion, ethnicity, or a special interest. But an ideal app would understand you holistically and find the best matches for you as a complex and multidimensional person, without forcing you to collapse yourself into one particular niche as the determining factor.

mike_hearn 2y ago
Is it really an algorithms problem? Surely that idea has been well explored over the last 20 years. OkCupid was big into the idea of algorithmically matching people based on their personality and it got crushed by Tinder, same as all the others. The problem with dating sites is not insufficiently smart machine learning. It's more like human nature when faced with a large homogenous pool of candidates, combined with the shift to smartphones as the primary way young people interact with online services (significantly restricts how many words people can be expected to produce and therefore how much data you can get for ML). OkCupid explained the problems on their blog many years ago: 1. Men aren't that picky and are willing to make a lot of effort to find a mate. They would like a nice girl who will take care of them and not be too crazy. Ideally a pretty one younger than them, but most men will settle for plain and same age or older if they have to. This lack of pickiness combined with willingness to make an effort manifests in simply swiping right on / messaging every single woman ("spam" as you put it). They are also pretty fair judges of women's attractiveness: the ratings they give are normally distributed. 2. Women become extremely picky. Tinder is/was notorious for women putting minimum heights in their profile, some also like to specify minimum salaries, specific kinds of jobs, very specific age ranges (invariably older than themselves) and so on. You don't see men do this! They also become unwilling to put in even small amounts of effort. Tinder expects only a few sentences about yourself in a profile but many women don't even write that much. Why would they, they feel like they can pick men from a menu without needing to write a personal essay. Unfortunately, women's ratings of men aren't even close to normally distributed, as they rate most men as being far less attractive than average. There are only a relatively small number of 6ft5 35-40 year old men earning 250k+ a year with a great sense of humor, the problem is women don't seem to reflect that reality in the distribution of attention they're willing to give. Only a tiny sliver of men are considered as being worth replying to, these men get all the female attention and can then sleep around as much as they want, leaving lots of heartbroken ladies in their wake. Your description of Rarebird makes it sound like the service assumes men and women are pretty much the same and a good dating service would just use AI to make everything better. OkCupid tried that and although it had its fans, once Tinder came out it became something of a ghost town. There's simply no way to satisfy women's expressed preferences no matter how fancy your ML gets. It's a social problem, not a technical one. (I am so glad to be married and out of that scene...)
- bb88 2y ago
  Observations: 1. Both sexes are willing to lie. 2. Some just want sex. Others want companionship. And yet a third loves getting attention. Others are looking for financial help. And criminals are looking for victims. 3. Any metric the matchmaking algorithm uses will be gamed by someone smarter than you, willing to ruin the experience for everyone else.
  - jey 2y ago
    > 3. Any metric the matchmaking algorithm uses will be gamed by someone smarter than you, willing to ruin the experience for everyone else. Not necessarily; mechanism design studies how to construct systems that aren't gameable, even though everyone in the system is acting out of self-interest. https://en.wikipedia.org/wiki/Mechanism_design https://en.wikipedia.org/wiki/Mechanism_design
    - bb88 2y ago
      Does that include teaming together to game the system? E.g. The game of poker changes if two or three at a table are secretly working together. Counting cards in black jack is harder to detect if a there's a team behind it.
      - jey 2y ago
        Yes, in the sense that the amount of "distortion" they can achieve is bounded to be small. Frankly, I'm approaching this as an engineering problem, so I haven't tried to work out all the gory mathematical details to fully theoretically characterize the distortion achievable by coordinated action. If you want to work on this and can apply the theory of [1], [2], and [3] below, send me an email. 1. https://doi.org/10.1093/restud/rdy042 https://doi.org/10.1093/restud/rdy042 2. https://doi.org/10.3982/ECTA7260 https://doi.org/10.3982/ECTA7260 3. https://doi.org/10.3982/ECTA6995 https://doi.org/10.3982/ECTA6995
        bb88 2y ago
cykros 2y ago
To be fair, the swiping was something that if I'm not mistaken was brought to the scene by Tinder, which was never designed as a "dating" app, but rather, a hook up app, that people decided to use for more serious dating. Big surprise when it doesn't do the job it was never designed to do.

by jey 2y ago

mike_hearn 2y ago

Is it really an algorithms problem? Surely that idea has been well explored over the last 20 years. OkCupid was big into the idea of algorithmically matching people based on their personality and it got crushed by Tinder, same as all the others. The problem with dating sites is not insufficiently smart machine learning. It's more like human nature when faced with a large homogenous pool of candidates, combined with the shift to smartphones as the primary way young people interact with online services (significantly restricts how many words people can be expected to produce and therefore how much data you can get for ML). OkCupid explained the problems on their blog many years ago: 1. Men aren't that picky and are willing to make a lot of effort to find a mate. They would like a nice girl who will take care of them and not be too crazy. Ideally a pretty one younger than them, but most men will settle for plain and same age or older if they have to. This lack of pickiness combined with willingness to make an effort manifests in simply swiping right on / messaging every single woman ("spam" as you put it). They are also pretty fair judges of women's attractiveness: the ratings they give are normally distributed. 2. Women become extremely picky. Tinder is/was notorious for women putting minimum heights in their profile, some also like to specify minimum salaries, specific kinds of jobs, very specific age ranges (invariably older than themselves) and so on. You don't see men do this! They also become unwilling to put in even small amounts of effort. Tinder expects only a few sentences about yourself in a profile but many women don't even write that much. Why would they, they feel like they can pick men from a menu without needing to write a personal essay. Unfortunately, women's ratings of men aren't even close to normally distributed, as they rate most men as being far less attractive than average. There are only a relatively small number of 6ft5 35-40 year old men earning 250k+ a year with a great sense of humor, the problem is women don't seem to reflect that reality in the distribution of attention they're willing to give. Only a tiny sliver of men are considered as being worth replying to, these men get all the female attention and can then sleep around as much as they want, leaving lots of heartbroken ladies in their wake. Your description of Rarebird makes it sound like the service assumes men and women are pretty much the same and a good dating service would just use AI to make everything better. OkCupid tried that and although it had its fans, once Tinder came out it became something of a ghost town. There's simply no way to satisfy women's expressed preferences no matter how fancy your ML gets. It's a social problem, not a technical one. (I am so glad to be married and out of that scene...)

bb88 2y ago
Observations: 1. Both sexes are willing to lie. 2. Some just want sex. Others want companionship. And yet a third loves getting attention. Others are looking for financial help. And criminals are looking for victims. 3. Any metric the matchmaking algorithm uses will be gamed by someone smarter than you, willing to ruin the experience for everyone else.
- jey 2y ago
  > 3. Any metric the matchmaking algorithm uses will be gamed by someone smarter than you, willing to ruin the experience for everyone else. Not necessarily; mechanism design studies how to construct systems that aren't gameable, even though everyone in the system is acting out of self-interest. https://en.wikipedia.org/wiki/Mechanism_design https://en.wikipedia.org/wiki/Mechanism_design
  - bb88 2y ago
    Does that include teaming together to game the system? E.g. The game of poker changes if two or three at a table are secretly working together. Counting cards in black jack is harder to detect if a there's a team behind it.
    - jey 2y ago
      Yes, in the sense that the amount of "distortion" they can achieve is bounded to be small. Frankly, I'm approaching this as an engineering problem, so I haven't tried to work out all the gory mathematical details to fully theoretically characterize the distortion achievable by coordinated action. If you want to work on this and can apply the theory of [1], [2], and [3] below, send me an email. 1. https://doi.org/10.1093/restud/rdy042 https://doi.org/10.1093/restud/rdy042 2. https://doi.org/10.3982/ECTA7260 https://doi.org/10.3982/ECTA7260 3. https://doi.org/10.3982/ECTA6995 https://doi.org/10.3982/ECTA6995
      - bb88 2y ago

cykros 2y ago

To be fair, the swiping was something that if I'm not mistaken was brought to the scene by Tinder, which was never designed as a "dating" app, but rather, a hook up app, that people decided to use for more serious dating. Big surprise when it doesn't do the job it was never designed to do.