So here's what I'm thinking.

I would argue that main bottleneck for people having an abundance of good potential partners is usually not their character or personality. Which sure can be improved, but it's not the main obstacle.

I believe the main issue is the bad matchmaking process. Meeting through friends, or chatting someone at the bar/club/restaurant, or at ones hobbies is just inefficient. How many new potential partners is the average person gonna meet in a month? 10? 50? 100? And most people are even too afraid to even start a conversation with a stranger.

We live in cities of millions... how many opportunities are being missed for amazing relationships?

So here is the question: How hard is it for an AI to be trained to match people? Is there anything happening in this area currently? And if not what are the main obstacles? If we managed to train it to respond with a certain amount of common sense (GPT4), even if it's 50% as accurate as a human at matchmaking, given the profiles of potential candidates, it would be huge massively more successful than Tinder.

NOTE: I'm assuming that the AI would be provided a wealth of information about the candidates: there could be a long questionnaire to be filled, an interview with the AI itself, access to socials, pictures and videos, etc.

UPDATE: While writing this I'm starting to realise the immense processing power required for something like this, and that we're not exactly there yet, but a simpler version could be made. Something where you just answer different questions about what you're looking for, what you want to do in your life, etc. Sure it would depend on trust, but incorrect profiles could be flagged and deleted.

And still, I think in a few years the AI will be there technology wise, to be able to conduct and interview and gauge the personality and characteristics of a person.

So again, what do you guys think? How hard is it for an AI to be trained to match people? Is there anything happening in this area currently? And if not what are the main obstacles?

New to LessWrong?

New Answer
New Comment
24 comments, sorted by Click to highlight new comments since: Today at 12:47 AM

I believe the main issue is the bad matchmaking process.

I don't think that's likely.  Compared to previous generations, modern matchmaking is orders of magnitude larger and more detailed than it was.  

The main issue (IMO, as an outside view, having been married and glad of it for many years) is broken expectations of participants.  Most young people seem to feel that they need to find the perfect partner out of millions, who's a great fit for them and will somehow know it.  Most DON'T seem to want to get to know a few dozen (at most) potential mates well enough to figure out how to change to match better, and how compatible they are in terms of communication and compromise.  

As to your direct question, this kind of questionnaire-based matching has been around forever.  AI doesn't really change the basics, though chatbot-elicitation of personality traits may have promise.  The problem is that people lie to themselves and others, and it gets Goodharted almost immediately, so loses much of it's objectivity.

It takes getting to know more than a few dozen potential mates, at least for some people

I don't know how the "average number of dating partners before marriage" has changed over time, but I suspect it spiked massively in the internet era.  Of course, a lot also depends on the threshold for "getting to know" someone, and whether that's a first date or the point of exclusivity, or sex.

I think plenty of guys would be willing to settle for less than perfect. 

@Dagon Indulge me in a hypothetical. Let's say the average person gets to know one new person a day (which is probably an exaggeration). 365 new people a year. half of them probably the wrong sex (182) 1/3 is the wrong age (120). 8/10 they don't find attractive (24). 2/3 don't like them back (6).

That's 1 person every 2 months that you like and likes you back. And that's even before you go an a date, and all the things that could mess that up.

Now imagine if you had 10 of those people every month. 10 people that you like, and like you back and want the same things as you, at least in theory.

I can't imagine anyone not wanting to go on dates with compatible people they find attractive.

Now sure, that wouldn't solve the psychological issues and traumas, or knowing to be a good partner, but at least people would be able to experiment and learn. I feel like now, people have kind of given up because it's so hard to find good matches.

How can you learn to play the guitar if you only have access to it once a month?

365/year is probably a bit low for most people, and it's bursty rather than continuous, but sure.  I think I'd disagree with any model for which only the most attractive 20% are candidates, and where there's a single point-in-time opportunity to date.

Almost everyone I know dated people they'd known for a while and NOT dated for some time before considering romance.  Each candidate partner has a full life of their own, and their circumstances will change even as yours do.  

Although the dating web sites, I'm sure, have accumulated a lot of data about people looking for dates, I doubt any of them have taken the trouble to collect accurate information on how well the dates turned out especially when the criterion for success is whether the date leads to a long-term relationship.

So (at least if you want to help those looking for long-term relationships) you can't just buy the data you'd need to train the AI from an existing dating site: you'd need to collect it yourself and ensure it is accurate, which is expensive.

Contrast that situation to the situation of training an AI to play chess: it is easy to program a computer to determine with 100% accuracy which opponent won a chess game, which keeps down the cost of training.

No yeah, you would probably need to collect the data yourself. But that should not be that hard. Do you think AI would currently be capable of actually matching people correctly, given the right data?

The dating sites actively do not want to know about outcomes because some of the outcomes will be rapes, and the sites wants to be able to claim that they had no way to know about any rapes. Some of the victims of serial rapist Dr Stephen Matthews for example are currently suing a dating site. So, no, it is not easy to collect the data.

Here is the (long) video by which I learned about that law suit. Sadly, there is no way for me to include a Youtube URL in a LW comment without the URL's "helpfully" getting replaced by a large image that is about 300 times more visually salient than anything else on the page, and I hate when that happens, so I'm going to give only the part of the URL after the domain name:

Note: Just press CTRL+Z after the auto-embed happens, and you should be left with just the Youtube URL. I edited your comment to include the full URL like that.

I believe, the fact that they might not want to know the outcomes, due to potentially complex legal procedures the effort they would have to exert to protect users better doesn't make it hard, it makes the companies selfish.

Looking the other way and claiming ignorance because you don't want to deal with the legal implications of your product potentially causing harm is really horrible.

EDIT: Happy to hear the disagreements here.

The problem is coordination. Suppose that the AI gave you the perfect matching algorithm. But still, you have no users to apply this algorithm to, so you can't prove that it works.

The existing dating websites do not have a reason to change -- their current websites generate profit. You need to make your own website.

Okay, so now you have a website with a potentially perfect algorithm, but zero users. You start advertising. The first few users come, but get a poor match, because there is little to choose from. So they leave.

Someone like Facebook would be in a better position to do this, because they already have tons of users, and can advertise to all of them at the same time for free. So they could add it as an extra functionality to Facebook, and feed the algorithm the data about the users they already have. You would just have to consent to receiving dating suggestions. Also, Facebook would have less of a problem with fake profiles. (There are many, but not as many - as a fraction of total accounts - as there would be on a dating websites. Also, most fake accounts on Facebook are for marketing or politics, so they wouldn't be interested in dating.)

Not sure about the incentives, though. Maybe people who are happily in love, spend less time scrolling Facebook endlessly? Then Facebook would kill the goose that lays them golden eggs.

Facebook has a dating service. It has a huge fake profile problem that essentially makes it unusable. 

Facebook's inability to effectively handle fake profiles is always remarkable.

The existing dating websites do not have a reason to change -- their current websites generate profit.

Existing dating apps (presumably) make a profit now, but they could be making more profit if they were better, through a combination of pulling market share from their competitors, and bringing more people into the dating app scene (if dating apps had less of a reputation for being dystopian hellscapes).

There could be disincentives if they were too good at it (instantly pairing up all of their users into perfect marriages so the dating market dries up), but if they were really that good they could just charge a lot and front-load their profits, and they'd still be incentivized to do it before their competitors do.

How would a prospective user tell which dating site offer is genuine and which are just scams that also are charging a lot?

It's not like anyone, even the bonafide sites, could provide hard evidence of successful long term relationships, since that would be a huge privacy issue.

Users would find out about apps the same way they do now: Hearing about the apps from friends and/or media. If one dating app was much better than the others, people would hear about it from their friends, and people who don't have friends in the dating pool would hear about it from the media.

I also don't think it's a privacy issue to provide aggregate data about this. A dating app could run surveys asking people if they're still dating a person they met on the app. The harder part would be getting people to actually answer, but there's incentives you could try (like take a deposit and refund it if the person answers the survey), and if your app is actually life-changingly good it will be easier to convince users to take 5 minutes to help you out.

Epistemic status: heard in Internet comments

There is another coordination-like problem when women think that saying that they're looking for dates is weird/low-status.

These are tiny issues you are mentioning. The hardest thing to make is the AI.

If you have the algorithm for "perfect" matchmaking. You can just advertise the hell out of it, put a limit for the minimum number of users for the platform to work (100K?) and once that's reached activate it, match people, everyone is ecstatic to meet the love of their life there -> more publicity -> more people sign up -> success.

I think people would be willing to pay a LOT of money, for "perfect" matches. 

Just because a match is perfect does not mean that people end up in a relationship. The dating process is still there and needed to actually get people to fall in love. 

What is your point?

Of course you will have to go through the rest of process, but you just solved the most critical step.

I don't think it's the most critical step for most people. Actually, going through the mating dance in a way that leads to feelings seems to me often more critical.

I think we might be valuing the importance of each step based on personal experience. But maybe they're both equally important. ex. What's more important for a cake, buying the ingredients or cooking it?

How can one improve the "mating dance" ?

The fact that arranged marriages out of pretty small polls of candidates can produce good marriages suggests that the match is not that central.

There's a lot that has been written on dating.