Let the AI teach you how to flirt

post by AllAmericanBreakfast · 2020-09-17T19:04:05.632Z · LW · GW · 11 comments

"It's Not You, it's Me: Detecting Flirting and its Misperception in Speed-Dates" is a fascinating approach to the study of flirtation. It uses a machine learning model to parse speed-dating data and detect whether the participants were flirting. Here's a sci-hub link. I found three key insights in the paper.

First of all, people basically assume that others share their own intentions. If they were flirting, they assume their partner was too. They're quite bad at guessing whether their partner was flirting, but they do a bit better than chance.

Secondly, the machine learning model was about 70% accurate in detecting flirtation. It's much better than the speed date participants themselves, despite having far less information to draw upon and the fact that the authors used a more forgiving standard of success for people's detection rates than for the detection rates of the machine learning model.

Thirdly, storytelling and conversations about friends seem to be the strongest signals of flirtation. Talking about the mundane details of student life (this was on a college campus) were the strongest signals of non-flirtation.

Finally, men and women have quite different approaches to flirtation:

Men who say they are flirting ask more questions, and use more you and we. They laugh more, and use more sexual, anger [hate/hated, hell, ridiculous, stupid, kill, screwed, blame, sucks, mad, bother, shit], and negative [bad, weird, hate, crazy, problem*, difficult, tough, awkward, boring, wrong, sad, worry] emotional words. Prosodically they speak faster, with higher pitch, but quieter (lower intensity min). Features of the alter (the woman) that helped our system detect men who say they are flirting include the woman’s laughing, sexual words [love, passion, virgin, sex, screw] or swear words, talking more, and having a higher f0 (max).

Women who say they are flirting have a much expanded pitch range (lower pitch min, higher pitch max), laugh more, use more I and well, use repair questions [Wait, Excuse me] but not other kinds of questions, use more sexual terms, use far less appreciations [Wow, That’s true, Oh, great] and backchannels [Uh-huh., Yeah., Right., Oh, okay.], and use fewer, longer turns, with more words in general. Features of the alter (the man) that helped our system detect women who say they are flirting include the male use of you, questions, and faster and quieter speech.

This paper has changed the way I think about skillful heterosexual flirtation. I used to think that flirting was a unisex behavior, and that men and women were decently skilled at detecting it. In much the same way that it's harder to write a novel than to read one, I thought that the hard part was signalling your own intentions, not interpreting theirs.

Now, I think that a strategy for skillful flirtation is to get the other person to broadcast their intentions, and learn to interpret their signals correctly. Men and women have different flirting styles. Each person knows when they themselves are trying to flirt. But they're bad at guessing when their partner is trying to flirt. This suggests that if you can get your partner to engage in their own natural flirting style, and get good at detecting it, then you can guess their intentions with much more confidence than the average person is capable of.

Both men and women should try to make each other laugh, let their voices be more musical, and provoke each other to talk about love and sex. They should tell stories about their lives and friendships and try to avoid mundane details.

A man who wants to signal flirtation to a woman should ask lots of questions that provoke the woman to talk about herself at length. Note that the "appreciations" and "backchannels" that are negatively correlated with women's flirtation are responses that women tend to give to men who keep going on about themselves. This is the old standard advice.

A woman who wants to signal flirtation to a man should maybe find topics they can complain about together - hopefully in a lighthearted way. She could also talk about her life in such a way that it provokes him to be curious and ask questions about her or observe connections between himself and her.

11 comments

Comments sorted by top scores.

comment by ChristianKl · 2020-09-18T15:24:12.452Z · LW(p) · GW(p)

This sounds to me like it goodharts on the wrong thing. When on a date your core concern isn't to signal to the other person that you are flirting but that you are a desireable mate.

comment by AllAmericanBreakfast · 2020-09-18T15:51:11.346Z · LW(p) · GW(p)

My take is that we’re trying to do both in equal measure, and that a big part of showing you’re a desirable mate is showing that you know how to flirt.

In fact, my gut feeling is that signaling that you’re interested/are flirting is more important. When meeting a stranger, there’s plenty of time to suss out their desirable qualities as time goes on. But if you fail to signal interest, you miss out on the opportunity to discover those desirable qualities entirely.

There are plenty of people who’d make excellent mates who fail to find relationships because they don’t know how to tell when somebody is interested in them. Likewise, there are people who are terrible mates but have no trouble finding relationships because they know how to tell when somebody’s interested in them.

comment by ChristianKl · 2020-09-19T09:16:50.653Z · LW(p) · GW(p)

From a male perspective a core way of expressing interest is to ask for a way to follow up and follow up. 

From a female perspective signaling interest is more important, but engaging in actions believed by guys to be a show of interest is something different then engaging in actions from which a well-trained machine learning model can infer interest.

comment by magfrump · 2020-09-19T01:08:49.605Z · LW(p) · GW(p)

My primary use case for this would be at parties where whether or not someone is flirting is the core question.

comment by ChristianKl · 2020-09-19T09:29:15.586Z · LW(p) · GW(p)

That's not how the OP proposes this information to be used. All three paragraphs at the end that start bolded are recommending other ways to use the information.

comment by AllAmericanBreakfast · 2020-09-19T15:58:46.659Z · LW(p) · GW(p)

It actually is how I was proposing this information could be used:

“This suggests that if you can get your partner to engage in their own natural flirting style, and get good at detecting it, then you can guess their intentions with much more confidence than the average person is capable of.“

comment by ChristianKl · 2020-09-19T16:11:33.578Z · LW(p) · GW(p)

If you get someone to interact in style X that's normally a signal for Y it stops being a good signal for Y. That's the basic Goodhard's law principle. 

comment by AllAmericanBreakfast · 2020-09-19T18:50:38.749Z · LW(p) · GW(p)

It's hard to be charming, rather than just doing a clumsy imitation of charm. If one person is overly practiced at being charming, they might be able to influence their partner's behavior so that it's no longer a reliable signal of their level of attraction. This fits with the Goodhart's law interpretation.

Then again, it might be that in a romantic context, like speed dating, people are being careful to flirt only if they genuinely want to signal attraction. Flirting might even make the participants feel attracted to each other. This would work against Goodhart's law.

My intuition is that the latter factors are more important than the former, though I do think it's very possible for people to fall into clumsy imitations of charm, especially at first. But I have to assume that charm is a learned skill like just about everything else, so the clumsy attempts might be just an awkward phase.

comment by ChristianKl · 2020-09-19T19:18:35.692Z · LW(p) · GW(p)

The study doesn't measure people being charmed in the sense that they are perceived to be charming by other humans. 

comment by Rudi C (rudi-c) · 2020-09-19T08:46:42.152Z · LW(p) · GW(p)

Causation != correlation seems to me to be a huge hurdle in extracting any practical advice from the research.

comment by AllAmericanBreakfast · 2020-09-19T16:19:03.074Z · LW(p) · GW(p)

Hmm. I see this paper as being less “intervention X works to improve people’s flirtation detection rate” and more “here’s an outside view on people’s ability to detect flirtation, vs. an ML algorithm using clear standards to classify speech patterns.”

I think it’s actionable in two ways.

  1. You can speculate about ways to imitate the ML algorithm and get better at detecting flirtation yourself, or creating a context for healthy flirtation.
  2. You can give up on trying to detect flirtation since people tend to be so bad at it, and just ask out everybody you’re interested in if you think that would be socially appropriate.

For someone who feels anxious about flirtation and wants guidance, even having a rule or strategy for going about it that’s a non-toxic and plausible interpretation of these empirical findings might help. Certainly a controlled study might help, though it seems hard to blind it or to execute a flirtation strategy reliably.