0:00

It's the New Year

0:02

and for some that means resolutions to

0:04

eat healthy food and get in shape. You've probably

0:06

heard the standard advice to lose weight

0:08

eat fewer calories than you burn, but

0:10

we at GastroPods wanted to know, is

0:13

a calorie a calorie no matter what

0:15

food it comes from? And is one calorie

0:17

for you, the same as one for me, to

0:19

find out we visit the rooms where calories

0:21

are measured and the labs where scientists

0:23

are proving that the numbers on our food labels

0:25

are off sometimes by quite a bit.

0:28

So is the calorie broken? Find

0:30

gastric pod and subscribe wherever you get

0:32

your podcasts.

0:36

James Wan produced the hit horror film

0:38

Meghan. He made the saw movies.

0:41

He made the conjuring

0:42

and he can see horror everywhere.

0:45

At nighttime, if I hear

0:47

something? Oh, I go somewhere and I

0:49

think that they may be someone out there.

0:52

I actually go, hello?

0:54

Is someone there?

0:56

I got she found myself to have done

0:58

that. James Wan on why

1:00

we always want to be scared. This

1:03

week, on Intuit, Vulture's

1:05

pop culture podcast.

1:15

So, Sengita, I did something for

1:17

the podcast. Tell me about it.

1:20

I started swiping again, but

1:22

not on a real dating app. This was

1:24

a simulated dating game, monster

1:26

match. I thought it might help me understand

1:28

how real dating apps work.

1:32

Yeah. I don't think I've ever had a soundtrack

1:34

to my swiping session. I

1:36

guess that's because they're gonna deliver some

1:39

unpleasant news and they need me to have

1:41

music. Monster

1:45

Match works like a Bizarro dating app.

1:48

You create a profile, I get to

1:50

choose my body. Wouldn't

1:53

that be an interesting option to have in life?

1:56

I'm gonna pick kind of like a snake

1:58

like It's kind of interesting to think

2:00

about, you know, reptiles mating

2:04

and dating. I ended up

2:06

with an iron man robot mask,

2:08

big hair with highlights, and

2:10

a necklace. Very hugged.

2:13

Then I got to choose a background. I

2:15

picked a city. Then it was

2:17

time to start swiping. First,

2:19

I saw the profile of count Daniel.

2:22

For work, heat has a

2:24

startup that makes farm to table

2:26

one hundred percent vegan blood for

2:28

vampires. Good

2:30

for you, I swiped right.

2:33

Daniel has already started talking to me.

2:36

I'm gonna say you seem like

2:39

bloody fun. You seem

2:41

like bloody fun. That's

2:43

a great line. I

2:45

mean, you have to work with what people give you.

2:47

I'm glad I still got some chat game.

2:49

Then there was little Johnny Chestface.

2:52

Larry, he was a person, cringe

2:55

Chris, who is a self identify

2:57

afide nightmare date, Zoloth,

3:01

the great demon. He says in

3:03

the zombie apocalypse, I'd be the one

3:05

dot dot

3:05

dot, starting it all.

3:08

Monster Medtruly lives up to its

3:10

name. Yep. And every

3:12

so often, the game would reveal a little bit

3:14

about how my swipes were affecting my

3:16

recommendations. Basically, the profiles

3:18

the app would show me next. And

3:20

that is the point of the game, which is

3:22

modeled on how real life dating apps work.

3:26

Berman is the guy behind monster match.

3:28

The game was inspired by a friend of his

3:31

who'd moved across the country to San Francisco

3:33

and was single.

3:34

Charming, handsome, friendly,

3:38

nine months of swiping

3:40

led to one atrocious date

3:43

after another. And

3:46

I felt like either something is

3:48

seriously wrong with him. Something

3:51

is seriously wrong with all the

3:53

women in San Francisco or

3:55

something was wrong with the software.

4:00

Software. That's something

4:02

Berman, a game developer, could figure

4:04

out. But what was it

4:06

specifically? What element of

4:08

the tech was making all of this feel

4:10

so bad? It turns out to be

4:13

the algorithm that recommends Who

4:15

you see?

4:22

This is Land of the Giants. Dating

4:25

apps use algorithms to determine

4:27

who users see and what kinds of

4:29

matches they have access

4:30

to. Some apps brag

4:32

about their algorithms, Some won't

4:34

talk about them at all, but none

4:36

reveal exactly how they work, which

4:38

means users are at the mercy

4:40

of technology that few people

4:42

understand. But is

4:44

all powerful when it comes to matching them

4:46

with potential partners. So

4:49

how good are these algorithms? And

4:52

when users feel like they aren't working,

4:54

can they hack the system? Asking

5:00

a dating exec how their matchmaking

5:02

algorithm works is like asking

5:04

Coca Cola for its top secret Coke

5:06

formula. You never get a straight

5:08

answer. But I asked who

5:10

I could, what I could, when

5:12

I could, like Jonathan Badin,

5:15

cofounder of Tinder. There's

5:18

like an urban legend around the Tinder

5:20

algorithm. Did you guys ever rank

5:23

people based on attractiveness or desirability

5:25

like hot or

5:26

not? It was an algorithm that was essentially

5:28

based upon the number of

5:30

swipe right swipes that people would

5:32

get, and that would eventually give you

5:35

some sort of score, and you could then

5:37

show people others that had more

5:39

similar types of scores to

5:41

themselves.

5:42

But Ian says this concept is called

5:44

an Elo score, which is also

5:46

how chess players are ranked in competition.

5:49

If you beat a high ranked player,

5:51

your ranking goes up a

5:53

lot. On Tinder, if a

5:55

popular person swipes right on you,

5:57

then you will see more attractive people on

5:59

the app. That sounds like high

6:01

school BS at scale. That sounds like

6:03

a popularity contest. Doesn't

6:05

it? Nadine's no longer at

6:08

Tinder and the company says Elo

6:10

scores are gone too. Tinder

6:14

says that today, the app has a recommendation

6:16

algorithm that factors in how active

6:18

you are on the app. Your profile details

6:20

in photos, your swipe history,

6:23

and how often your profile is

6:25

liked, which still seems like a popularity

6:27

contest. When we asked

6:30

Bumble about its algorithm, we got

6:32

a somewhat opaque answer.

6:34

A spokesperson told us that the algorithm

6:36

learns from your history to serve up

6:38

match That sounds so vague.

6:40

It could describe a lot of things.

6:43

Totally. There are apps that

6:45

we'll share a tiny bit more about

6:47

their algorithms. Because the algorithm

6:49

is a big part of their marketing and

6:51

their pitch to users. Oh, that

6:53

sounds like hinge. Exactly.

6:56

I asked former design and product

6:58

VP, Tim Matt Guggen, about it. The

7:00

most compatible feature on

7:02

hinge uses an algorithm

7:05

that pairs people together and it uses

7:07

something inspired by close to the

7:09

Gail shapely algorithm.

7:10

You'll actually see this come up in Hinge's

7:12

marketing. Its special proprietary

7:15

algorithm was modeled on the Gail Shapley

7:17

algorithm. A complex matching

7:19

formula, which won its creators, a

7:21

couple of mathematicians, a Nobel

7:23

Prize. So that's cool.

7:26

But what does that mean? How does

7:28

it work? I'm

7:28

not gonna tell you how it works. The

7:31

first rule of talking about your

7:33

algorithm is that you don't

7:35

talk about your algorithm. Macguin

7:37

left Hinge in twenty twenty two.

7:40

We followed up with Hinge, and

7:42

a spokesperson didn't have much

7:44

else to add to Macguin's explanation.

7:48

Matchmaking algorithms go back

7:50

to the o g's. Match dot

7:52

com and e Harmony how did their

7:54

recommendation systems to distinguish themselves

7:57

from free wheeling online chat rooms?

7:59

I'm wearing a Tom Bray, The

8:01

CEO of Match Group Americas told

8:03

me that these recommendations were rudimentary

8:05

at

8:06

first. Users came in and

8:09

they filled out this is who I am

8:11

and this is who I am looking for.

8:13

And then the algorithms like, okay,

8:15

you match each other around nine out of the ten

8:17

things you both say are looking

8:18

for, then that person became a

8:20

recommendation for you. That changed

8:22

in two thousand and eight. When he joined

8:25

Match, as vice president of

8:27

Strategy and

8:27

Analytics, and updated its

8:29

algorithm. We said, you know what?

8:31

We just gotta put that aside,

8:33

what people say they want. And we're gonna look at

8:35

what people are doing and then build recommendation

8:38

engines that entirely were based on user

8:39

actions. It also helped that

8:42

these sites had professionals to

8:44

legitimize their systems and build trust

8:46

in the

8:46

algorithm. Eharmony had

8:49

doctor Neil Clark Warren, Match dot com

8:51

had doctor

8:51

Phil. So you both have these PhD

8:55

psychologists, old white

8:57

dudes as I call them out on

8:59

Oprah and the talk

9:00

shows. Sam Yiggin with CEO of Match Group

9:03

when I went public in twenty fifteen. But

9:05

he launched OkCupid in two thousand

9:07

four. He'd already been thinking

9:09

about the online matchmaking space for more

9:11

than a decade. They had

9:13

this resounding message

9:16

which is usually talking to middle aged

9:18

women, you don't know what you're looking

9:20

for. We

9:22

know what the right match for

9:24

you is. And there's

9:26

something very, to me, creepy

9:28

about that. But there's also

9:30

something very lucrative about that,

9:32

which is if you can convince people

9:35

that the reason you're single is very simple. You

9:37

just don't know what you're looking for and we have

9:39

the answer. We know

9:41

what a match looks like and we know who your soul

9:43

mate is gonna be, that's a great business

9:45

proposition. In other words,

9:48

trust the algorithms.

9:52

This same message has persisted

9:54

through today. Sure.

9:56

An IRL meet cute sounds

9:58

great, but it relies so much on chance.

10:01

Plus, are you meeting enough people for

10:03

that? And do you really want to talk to

10:05

a stranger? Enter

10:07

the apps. They filter thousands

10:10

of people who are ready to mingle.

10:12

They use quantitative analysis

10:14

to create your dating pool.

10:16

It's

10:16

math. It's science. It

10:19

seems kinda magic to us. We're not really sure why I

10:21

work so well. We don't want anyone to see

10:23

how how dumb it is. Cathy

10:26

O'Neil is a mathematician, data

10:28

scientist, and author of The Book:

10:30

Weapons of Math Destruction. It's

10:33

all about algorithms and their impact on

10:35

society. She used to work in big

10:37

tech. O'Neil says there's a reason

10:39

why the apps are so KJ about

10:41

their algorithms. She thinks that

10:43

if we really knew how they

10:44

worked, we might not want to put so much blind

10:47

faith into them. You know, this is framed

10:49

as really hard because they want you to sort

10:51

of trust it and slash be intimidated by

10:53

it. But it's actually

10:55

not hard I asked O'Neil to

10:57

break it down for me. She says

10:59

that dating apps use predictive algorithms,

11:02

and predictive algorithms are

11:04

about digital pattern

11:05

making. You basically are saying, if it worked

11:07

in the past, it'll work in the future.

11:09

Predictive algorithms just take

11:11

historical data. They look for patterns

11:13

of success or failure with respect

11:15

to a specific definition of

11:17

success, and then they project the

11:19

future from the

11:20

past. Extrapolate. That's

11:23

pretty simple. An app isn't smart.

11:25

It doesn't really know what users

11:27

want. It can't tell whether two

11:29

people will have fun together, share

11:32

passions, or be sexually compatible.

11:35

But it knows if you've looked at someone's

11:37

profile, it can count the number of

11:39

words used in a chat.

11:41

Or how many numbers were exchanged.

11:43

This is the data that is

11:46

being used to determine

11:48

your success or failure as a

11:49

mate. So

11:50

what do you do if you feel

11:51

like you're failing more than you're

11:54

succeeding? Theaters

11:56

don't really know how the algorithms

11:59

work. But they do know what it

12:01

feels like to go on a bunch of terrible

12:03

dates. So in the

12:05

dark, they come up with strategies in

12:07

an attempt to mitigate all of

12:09

those bad

12:09

matches. Sengita, I

12:12

wanna tell you about Sara Satteroff.

12:15

She's thirty seven and lives in

12:17

Toronto.

12:19

Sodaroff had burned out

12:21

on the apps at some point. The

12:23

thought of having to go back onto them

12:26

was almost worse

12:28

than the breakup. It was an almost worse

12:30

feeling of failure

12:32

and having to, like, walk

12:34

back into this

12:36

cohort of

12:39

rejects as I felt it to

12:41

be. Honestly, I think I was paralyzed with

12:43

doing it for

12:44

weeks. She was over the apps but

12:46

not over the idea of finding a partner.

12:49

So she settled on a new game

12:50

plan. This time, she'd say yes

12:52

to everyone. I was very mission

12:54

driven and I was very active.

12:57

I think you're told a lot dating

12:59

is a numbers game. You gotta get out there. You

13:01

gotta get your reps in. You've gotta

13:03

be giving yourself the

13:05

maximum ability to meet with them and match

13:07

with the most amount of

13:08

people. Her last cycle through the apps, she

13:10

thought maybe she'd been a bit too picky.

13:13

Only rightswiping on select

13:15

profiles and interacting with

13:17

certain kinds of

13:18

guys. But then she felt like she wasn't getting the

13:20

matches she'd hoped for. I tried to open that up

13:22

to say you never know if you're gonna have a connection with

13:24

somebody. On him, I would say

13:26

yes to anybody. Who had messaged me or I

13:28

would match with anybody who had messaged me because that's the

13:30

option that they give you and then have those

13:32

conversations. But Sudaroff's

13:34

new approach bombed. She

13:37

could feel the algorithm adjusting to

13:39

her new preferences, but not in

13:41

a good way. I definitely

13:44

can sense that they bucket you into

13:46

different thresholds based on people who

13:48

match with you and like you. After

13:50

languishing on the apps for like a couple months, you get

13:52

bucketed into kind of like a more general

13:54

population, which is a lot of riffraff and

13:56

a lot of people you probably don't wanna match

13:58

with. She was getting further away

14:00

from her goal. Time for

14:02

a new strategy. Hind

14:09

says its recommendation algorithm

14:11

suggests compatible users who like

14:13

each other back. It was a

14:15

glimmer of

14:15

direction, so Sodarov grabbed

14:18

onto it, and leaned in.

14:20

When you first start

14:22

matching, swiping, and

14:24

liking, it will say we're

14:27

getting a feel for the kinds of person you

14:29

like. And so I think you think, oh,

14:31

the app is reading my behavior

14:33

I think it might make the user experience more

14:35

enjoyable because you feel as they're getting people who

14:37

are actually better tailored to

14:38

you. She reset her criteria,

14:41

made her own filters, Try to

14:43

give Hinge's algorithm enough information

14:46

to go on. Try to

14:48

be a model Hinge citizen in hopes

14:50

of getting better

14:50

matches. You get

14:53

higher quality matches if you

14:55

tend to interact with a prompt. But

14:57

I think it leaves you with some level of comfort. You're

14:59

like, oh, the app is working for me. The app is

15:01

looking out for me. Which is

15:03

a fantastic behavioral trick.

15:06

But this comfort didn't last

15:08

long. I'm often offended by

15:10

the people they think I am most compatible

15:12

with. I'd love to learn from them what that

15:14

compatibility is actually based on because you'll

15:16

see that and say, I'm not physically

15:18

compatible with this person. We don't have the same interest.

15:20

So where does this algorithmically match

15:23

me as a compatible person

15:24

there? So that wasn't

15:26

working either. I

15:27

think there is a certain sense of like, let

15:29

go and let God. The algorithm is gonna

15:31

present to me people who are good matches

15:34

me. Maybe I'm a bit scorned, but I

15:36

don't think it's working for me. I

15:38

don't think the matches that I

15:40

have been given are of

15:42

any

15:42

interest, I would say

15:45

to me, that is just a big

15:47

gray hole. When

15:49

Sotirof didn't give the app enough to go

15:51

on when she matched with everyone, she

15:53

got irrelevant matches. When

15:55

she tried to feed the algorithm,

15:57

the experience was still trash. And

16:00

now Sideroff is wondering if there's

16:02

any strategy at all that could get

16:04

her closer to finding a partner.

16:06

Sounds like probably not?

16:10

Okay, luxury. Now, I have

16:12

a story for you about someone

16:14

who tried to make the algorithm work

16:16

for him. That's in a minute.

16:27

Music pop quiz time. Can you identify

16:30

these three sounds? That's

16:32

an 808. It's made by pitching and

16:34

just jorting the sound of a kick

16:35

drum, and it's now replaced bass guitar in

16:38

countless pop records from Drake to Kim

16:40

Petras. Okay. Number

16:41

two. That's a

16:43

vocal chop, a short sample

16:45

of a singer, cut up, drenched in

16:47

reverb, and made totally ubiquitous

16:49

through the hands of EDM producers like

16:51

SquirrelX. Alright. Last one.

16:54

That's a Reese'space. It's a sludgy

16:57

synthesized baseline created by

16:59

producer Kevin Reese's Sonderson in

17:01

nineteen eighty eight. You can hear it in a

17:03

genre like Jungle, Drummond Base, UK

17:05

Garage, and it's even featured

17:07

all over Taylor Swift's

17:09

latest record. Sounds of popular music

17:11

are always evolving. If you

17:13

wanna be able to know all of it sounds,

17:15

the switched on pop podcast will

17:17

break it down for you I'm

17:19

Charlie Harding, the cohosta switched on

17:21

pop. And if you can name this

17:23

sound, honestly, you should come on the

17:25

show and tell us about it. Switched on pop comes

17:27

out every Tuesday wherever you

17:29

get podcasts. Jeremy

17:37

is thirty and lives in Philadelphia.

17:40

He's on dating apps and has a very

17:42

specific understanding of

17:44

them. an app designer, I

17:46

pay a lot of attention to

17:48

the patterns and behaviors and, I

17:50

guess, just methodologies behind

17:53

the apps that I use on a daily basis? Jeremy

17:55

asked us not to use his last name

17:57

because he works in tech and wanted to

17:59

speak frankly about his

18:01

experience. These days,

18:02

he does all right on the apps. My

18:05

friends call me, like, generically attractive,

18:07

and I think that that's something that

18:10

I, like, flight or, like, at least into on dating apps

18:12

where I'm, like, okay. Well, I'm,

18:14

like, a thirty something white male

18:16

decently not

18:17

ugly. I end up with, like, quite a few

18:20

matches on dating apps. But there

18:22

was a time when that wasn't the case.

18:24

When despite his efforts, he

18:26

didn't have an abundance of

18:28

matches. He used

18:31

to make his profile as true to his

18:33

personality as possible. He

18:35

talked about his actual interests, like

18:37

going to museums and concerts. He

18:39

gushed about his love for experimental

18:41

film. But it all backfired. I realized, like, oh,

18:44

I'm just not gonna care about constantly

18:46

scrutinizing every detail and updating my

18:48

dating profile because of

18:50

you

18:50

know, it just never

18:53

really seemed to get anywhere. So

18:55

he

18:55

went back to basics literally.

18:58

When I started getting a lot more matches on my profile,

19:00

it was because I had just

19:03

like the things that were meaningless it's

19:06

definitely discouraging to know

19:07

that, like, what's generic and

19:09

watered down kind of works best. What

19:12

is generic Jeremy's profile look

19:14

like. Can you give me, like, a sample of

19:16

one of your prompts?

19:17

And, like, what what are the photos?

19:20

I think my current about me like, my interests

19:22

include, like, getting lost and listening, and

19:24

that one seems to, like, convert the

19:25

best. Getting

19:26

lost, what the fuck does that

19:29

mean?

19:29

Don't know. It's just something I thought that I thought was, like,

19:31

really stupid and then it kinda worked.

19:35

For Jeremy, volume was

19:37

a success. The more

19:39

matches, the better. And that meant

19:41

engaging in a bit of Jeremy

19:43

Erasure. He walked me through

19:45

his photos, One with his

19:47

mom, check. Another with a

19:49

dog, check. If

19:51

you look up tips to improve your dating

19:53

out pro file, moms and

19:55

pets are huge. This

19:57

one is

19:57

a new addition. Just a photo of me

19:59

blowing bubble gum. I think it

20:02

works fine. Maybe it makes me look funny

20:04

or something. It's still very generic.

20:06

And then

20:07

there

20:07

are the props. This

20:09

is a good one that says, we'll get along if you can

20:12

laugh. That's

20:12

it. That's a killer. You know, that's a

20:14

real that's a real charmer there.

20:19

So basically, the more personality

20:21

you show, the worst for you.

20:23

Yeah.

20:23

That's what I've found for sure. Jeremy

20:26

says his strategy to lean

20:28

into his generically handsome white

20:31

maleness has gotten him more

20:33

matches. But have they been better

20:36

matches? I'm not like, wow. This

20:38

person is, you know enough.

20:40

I've never, like, wow, they're really

20:42

getting, like, who I'm going for. You know? It's it's,

20:44

like, almost opposite. It just feels, like,

20:47

an endless abyss

20:49

of

20:49

individuals. For Jeremy,

20:52

more matches was the goal. He'll venture

20:54

into the abyss and sort it out from

20:56

there. For Sara Sotteroff, quality

20:58

matches were the measure of success.

21:02

So two cases, two

21:04

people trying to game the apps

21:06

to get the results they want.

21:08

Each of them had a hypothesis.

21:10

They ran experiments and adjusted

21:13

parameters. They gathered data on

21:15

results and came to conclusions about

21:17

how the apps

21:18

work. But the truth is,

21:21

they don't know. They're just

21:23

trying anything to make all

21:25

of this feel a little bit

21:26

better. But

21:30

there are people who do

21:32

know. They don't need to experiment in

21:34

the dark, because they've got all the

21:36

variables laid out right in front of

21:38

them. The apps may not be willing to

21:40

demystify their algorithm, but

21:42

we found someone who would. Fanyan

21:46

Zhang. A leading dating

21:48

app for intentional daters

21:50

wanted to improve its algorithm and

21:52

partnered with Zheng and some of her fellow researchers

21:54

to help them figure out how to

21:56

do so. It wasn't Hinge, but

21:59

it is fair to say that this app functions a lot

22:01

like other leading apps. Zheng

22:03

is an assistant professor at

22:05

Columbia Business School. She

22:07

studies how data can improve business operations. For

22:10

this dating app, she was asked to figure out how

22:12

to increase the number of matches on the platform.

22:16

Zheng and team got access to all the

22:18

inner workings of this app. Its

22:20

user data, the algorithm, everything.

22:23

Beep pop boop. They pulled a few levers

22:26

and algorithm improved

22:28

the match rate by around the

22:30

thirty percent. That

22:34

seems significant. They

22:36

did overhaul some things.

22:38

One example, if there were tons

22:40

of great people who could be a great fit

22:43

for you, the new algorithm wouldn't flood you with

22:45

them. You're a catch and you

22:47

might be juggling a dozen conversations

22:49

on the app

22:50

already. They updated the

22:52

app to keep users a little bit thirsty.

22:54

What this means from the platform's

22:57

perspective is that when you

22:59

have some great potential profiles to

23:01

show to your users. You don't want to give

23:03

it to them or at once.

23:05

You want to space out these

23:07

attractive potential matches to

23:09

your users, and this increases the

23:12

chance of a successful match. They

23:14

found that when a user was overwhelmed with

23:16

options, they might miss a lot of

23:18

great potential matches. And Zheng

23:20

says a bit of withholding,

23:23

benefited the company

23:23

too. It may now be the

23:26

best thing for the platform if everyone

23:28

finds their right person

23:30

and never returned again.

23:33

Right? So there is a little bit of keeping

23:35

their users active on the

23:37

platform, at least for some

23:38

time. That sounds like a

23:41

conflict Many of the datas we've

23:43

spoken to who are looking for a long

23:45

term relationship, they would love to

23:47

download an app, find a partner, and

23:50

never have to open it

23:51

again. Yeah.

23:52

I know a lot people who would love that. Okay.

23:55

So the next thing

23:57

that Zheng looked at had to do with something

23:59

truly fundamental about how this

24:02

app and many others work.

24:04

It's about how the algorithms pick

24:06

matches in the first place. How

24:08

they figure out who might like whom,

24:11

the secret sauce, the recommendation engine

24:13

that shapes romantic prospects across

24:15

the apps. Here's how

24:18

it works, The algorithm needs data,

24:20

lots of it, to make an informed

24:22

recommendation. So you're lumped together with

24:25

others who seem similar to

24:27

you.

24:27

Then the data that you and thousands of others

24:29

give the app, who you swipe right and

24:31

left on, who you message, is

24:34

crunched. Those

24:36

millions of interactions help the

24:38

app make judgments and predict the future.

24:42

That sounds romantic. Take

24:48

height. Looking at the data

24:50

in aggregate, the algorithm might

24:52

find

24:53

yes. People who are six four

24:56

prefer to match with people who are at

24:58

least five ten. So

24:59

then you have

25:00

Louise, who is six four.

25:03

And only sees people who are at least five ten on

25:05

the app.

25:06

But for this particular person,

25:10

the algorithm my realize that

25:12

sometimes when we recommend

25:14

someone who's five six, but with

25:16

all these other features, the

25:18

user could show particular interest.

25:20

Right? And then I can build that into

25:22

the algorithm. But Louise

25:25

also loves music and he also

25:27

matches with people who are really into

25:30

seventies Crock. Aaron fits this

25:32

bill. He wants to take a date to

25:34

see Jethro Tull. He could

25:36

be a match. Buddy's five

25:39

six. Will the app

25:41

show Aaron to Louise? Will

25:43

Louise be freed from the tyranny of the

25:45

tall people bubble? Oh

25:47

my god. You're killing me. What

25:49

happens? Well, their

25:51

fate is in the hands of the

25:53

algorithm. If it makes inferences

25:55

based on certain kinds of group behaviors,

25:58

like tall people or into other tall

26:00

people, it would actually keep

26:02

Louise and Aaron apart. To

26:04

avoid these kinds of missed

26:05

connections, Zheng decided to build more

26:08

granularity into the system.

26:10

We made the algorithm

26:13

more personalized really have the

26:15

algorithm learn more about the

26:17

users' preferences over

26:19

time, not to treating people

26:21

as groups or only look at

26:23

their average

26:23

behavior, but look at the individual

26:27

preferences. What

26:29

all

26:29

this tells us is that these algorithms

26:31

can leave the daters who depend

26:33

on them wanting.

26:35

If I'm Asian, do I only get to

26:38

see Asian profiles? And maybe that's

26:40

not what I'm interested in. But in the

26:42

end, this is the impact of the

26:44

algorithm. And I actually lead

26:46

to outcomes of matches and

26:48

how do we quantify the

26:50

laws of welfare for individuals like

26:52

that. Right? How do we say,

26:54

you know, is it just the preferences or is

26:56

it actually the result

26:58

of the algorithm? But

27:01

the companies have

27:02

sold us on the idea that math and data

27:04

are the key to romantic success.

27:07

Trust the

27:08

process, trust the algorithm.

27:10

Now that obviously makes sense

27:12

for the companies, but does

27:13

it make sense for you?

27:16

We trust the data that math and science

27:18

couldn't be wrong, that it's on us

27:20

if we're not finding the right people

27:22

in the apps. But

27:24

what if you don't fit into the world, the algorithm

27:26

thinks you should? That's why

27:28

I sometimes just say, they make lucky people

27:30

luckier and unlucky

27:31

people, unlucky. Mathematician, Kathie

27:34

O'Neil again. So if

27:37

you find yourself one of the unlucky

27:39

ones, do you question the

27:41

tech? Or do you question yourself?

27:43

We become more and more insecure and self

27:46

conscious about ourselves and

27:48

how we are reduced and flattened into

27:50

these little data points. The

27:52

overall effect is IKE and

27:54

it doesn't actually improve our

27:58

love experiences. Another

28:00

way to say that, our trust,

28:03

it's likely been misplaced.

28:17

There's another company that is trying to build

28:19

trust in online dating, not through

28:21

technology, but through culture, and

28:24

it's working. It has twenty two

28:26

percent of the market share in the US

28:28

and presents a serious challenge to

28:30

match group's

28:30

dominance. The story of

28:33

Bumble. Next time on land

28:35

of the

28:35

giants. Land

28:40

of the giants is a production of the cut,

28:42

the verge, and the Voxmedia podcast

28:44

network. Ella

28:46

Wakemi, Ella Dessui, is the show's producer.

28:49

Cynthia Batu Visa is our production assistant.

28:52

Charlotte Silver fact check this

28:54

episode. Jolie Myers

28:56

is our editor. Brandon McFerland is our

28:58

engineer and also composed the

29:00

show's theme