Optimal Stopping is the first chapter of Brian Christian and Tom Griffiths' book Algorithms to Live By, which does a good job describing the problem in laymen terms.
For martingales (stochastic processes whose expected value given the current value is the current value) there is no strategy that does better (on average) than the (expected) initial value (Doob's optional stopping theorem).
I can't find it at the moment but I remember reading about this in an article about how the infinite pool of options in online dating broke the optimal stopping model (which previously had to do with optimal stopping relative to our Dunbar number) and made all matches unsatisfactory.
Seems like folk science, but a variation of it could be applied to job satisfaction in tech, where the greater number of perceived options makes all choices seem to be in the less green field.
Absolutely. I get the sense that we're approaching a cultural "oh, shit" moment when we realize we've just been building massive global mimetic conflict machines. Oops!
I don't think that would break it because you have a limited amount of time. You base it on 1/e of the number of people you expect to date not the number of potential options.
The problem in dating (and arguably also in the secretary problem) is that the criteria for selecting a person are not so clear to most of the people and may not be so easy to assess.
My not optimal algorithm has been to have at least 2 lists of attributes you want in your partner. I had 3. One is the "MUST HAVE" (or deal breakers), another one for the IMPORTANT ones, the last one was the "NICE TO HAVE".
I stopped when I found someone with all MUST HAVE and most of the IMPORTANT ones.
I don't see how unclear criteria would be an issue in the secretary problem.
For the secretary problem you only need to be able to answer the question "if I had to hire one of person 1, 2, 3, ..., k would it be person k?". Most people can do that without having a clear idea of what their actual criteria are.
> I have been married for 10+ years =)
That's sensible. Assume you want to be married for 40+ years. Applying reasoning similar to that of the secretary problem to determining how long you should stay married to your current partner before divorcing and marrying someone else, you've got another 5 years in the baseline establishment stage. :-)
Most of those apps have a (totally gameable) ratings system based on other people’s action of liking you or not
I think the apps should allow users to be opt out of that (short of being completely regulated away) because most users dont think their match/not match choice is affecting that profile’s visibility to others
Even when you are aware of it, and game it to increase your own visibility, its a very toxic way to treat other people
Oh yeah, many profiles you liked are never seeing your profile to make a decision, in case you didn’t know
You basically have to act as discerning as actually visually ideal people will be acting.
Matching everyone means matching already skipped people, which lowers your algorithm desirability to their level. More desired people wont even see your profile after that point. You wont appear in their feed and they aren't paying to see the whole queue of their matches.
So it is more important to imagine your desirability score than it is to amplify reach by matching more hoping for a bite.
As far as actually refining your profile, you’ll have to figure that out either way. A/B test, do whatever, thats its own separate art. But making sure it stays visible means being discerning. Takes several weeks for the algorithms to set you in place, and several weeks to change. So best to keep yourself distracted with other things, otherwise it'll just hurt your self esteem until more consensus-attractive people start matching you.
So let me rephrase this to ascertain I understood you correctly:
Your algorithm is basically "swipe 'NO!' upon all the non-very-beautiful people to get yourself put into the 'very beautiful people' bin by Tinder and then hope that it eventually shows you to people in that bin", right?
If yes, that's funny, because that's what I would do anyway without even wanting to game the system:
I think it is a rude waste of people's time if you date them even though you don't think they're attractive :) Prevents them from meeting someone who actually does.
The frequency of matches can be so low that people begin to swipe YES on almost all matches, hoping to see who bites. On an app like Tinder, it inadvertently or intentionally makes that easy to rationalize.
This is A strategy, but it practically ensures being in the less attractive bin and never even shown to people in the other bins.
Regarding rude waste of time, people adjust their standards, just like in the physical world.
Considered using this algorithm for simple robotics.
Specifically, a "dark seeker" robot that prefers to hide in dark places.
Sure, wire a photocell to a simple turtle-like robot and send them on a random walk to wander and measure how dark their environment is. But how to decide what is dark?
It's dark under the table, but even darker under the chair. Should the robot stop under the chair to "hide"? Perhaps the robot might have wandered a bit more and found that it was even darker still under the couch. But is it darker still somewhere else?
Never mind that the value it reads varies depending upon the time of day and which environment you drop it into.
I think you can see how I imagined the Secretary Problem algorithm to step in. Let it wander for some fixed amount of time simply to "sample" the environment. Then after that sample period the robot stops as soon as it stumbles into an area darker than the darkest it had just sampled.
Seems like a reasonable algorithm for a robot with fewer than two neurons to rub together.
22 comments
[ 3.6 ms ] story [ 30.5 ms ] threadhttps://en.wikipedia.org/wiki/Optional_stopping_theorem
Seems like folk science, but a variation of it could be applied to job satisfaction in tech, where the greater number of perceived options makes all choices seem to be in the less green field.
My not optimal algorithm has been to have at least 2 lists of attributes you want in your partner. I had 3. One is the "MUST HAVE" (or deal breakers), another one for the IMPORTANT ones, the last one was the "NICE TO HAVE".
I stopped when I found someone with all MUST HAVE and most of the IMPORTANT ones.
I have been married for 10+ years =)
For the secretary problem you only need to be able to answer the question "if I had to hire one of person 1, 2, 3, ..., k would it be person k?". Most people can do that without having a clear idea of what their actual criteria are.
> I have been married for 10+ years =)
That's sensible. Assume you want to be married for 40+ years. Applying reasoning similar to that of the secretary problem to determining how long you should stay married to your current partner before divorcing and marrying someone else, you've got another 5 years in the baseline establishment stage. :-)
I think the apps should allow users to be opt out of that (short of being completely regulated away) because most users dont think their match/not match choice is affecting that profile’s visibility to others
Even when you are aware of it, and game it to increase your own visibility, its a very toxic way to treat other people
Oh yeah, many profiles you liked are never seeing your profile to make a decision, in case you didn’t know
How does gaming it work if it depends on other people's actions?
You basically have to act as discerning as actually visually ideal people will be acting.
Matching everyone means matching already skipped people, which lowers your algorithm desirability to their level. More desired people wont even see your profile after that point. You wont appear in their feed and they aren't paying to see the whole queue of their matches.
So it is more important to imagine your desirability score than it is to amplify reach by matching more hoping for a bite.
As far as actually refining your profile, you’ll have to figure that out either way. A/B test, do whatever, thats its own separate art. But making sure it stays visible means being discerning. Takes several weeks for the algorithms to set you in place, and several weeks to change. So best to keep yourself distracted with other things, otherwise it'll just hurt your self esteem until more consensus-attractive people start matching you.
So let me rephrase this to ascertain I understood you correctly:
Your algorithm is basically "swipe 'NO!' upon all the non-very-beautiful people to get yourself put into the 'very beautiful people' bin by Tinder and then hope that it eventually shows you to people in that bin", right?
If yes, that's funny, because that's what I would do anyway without even wanting to game the system:
I think it is a rude waste of people's time if you date them even though you don't think they're attractive :) Prevents them from meeting someone who actually does.
This is A strategy, but it practically ensures being in the less attractive bin and never even shown to people in the other bins.
Regarding rude waste of time, people adjust their standards, just like in the physical world.
They don't? Isn't this the fundamental mechanic of tinder?
Something I think is more surprising is that your choices effect your visibility to others. Swiping right too many times is bad.
“do I want to potentially have attention from this person or not”
[1] https://en.wikipedia.org/wiki/Ski_rental_problem
Specifically, a "dark seeker" robot that prefers to hide in dark places.
Sure, wire a photocell to a simple turtle-like robot and send them on a random walk to wander and measure how dark their environment is. But how to decide what is dark?
It's dark under the table, but even darker under the chair. Should the robot stop under the chair to "hide"? Perhaps the robot might have wandered a bit more and found that it was even darker still under the couch. But is it darker still somewhere else?
Never mind that the value it reads varies depending upon the time of day and which environment you drop it into.
I think you can see how I imagined the Secretary Problem algorithm to step in. Let it wander for some fixed amount of time simply to "sample" the environment. Then after that sample period the robot stops as soon as it stumbles into an area darker than the darkest it had just sampled.
Seems like a reasonable algorithm for a robot with fewer than two neurons to rub together.