0:00

Time for a quick break to talk about

0:02

McDonald's. Mornings are for mixing and matching at

0:04

McDonald's. For just $3, mix and match two

0:07

of your favorite breakfast items, including a

0:09

sausage McMuffin. Welcome

0:33

to the You Are Not So Smart podcast. Episode

0:53

285. Are

1:11

you familiar with the internet debate over

1:14

whether or not a hot dog is a sandwich? I'm

1:17

familiar. That

1:24

is the voice of famed

1:26

psychologist Celeste Kidd. She

1:28

is a cognitive scientist and developmental

1:31

psychologist who is well known for

1:33

her research on human curiosity and

1:35

human certainty, how brains

1:38

develop knowledge. My

1:40

name is Celeste Kidd. I am

1:42

an assistant professor in psychology at

1:44

UC Berkeley. And my

1:46

lab is the Kidd Lab. We study human belief

1:49

formation. Kidd

1:54

studies how we acquire and

1:56

conceptualize information, how we form

1:59

beliefs around those concepts and

2:01

in general makes sense of the

2:03

torrent of information blasting its way

2:05

into our nervous systems via all

2:07

the senses and how that affects

2:09

our development how the development of

2:11

the mind proceeds from childhood all

2:14

the way to the day we

2:16

find ourselves in an argument about whether

2:18

a hot dog is a

2:23

sandwich so

2:27

what do you think is a hot dog

2:29

a sandwich this is one

2:32

of those debates that became the central

2:34

topic of discussion in our marketplace of

2:36

ideas a few years back one

2:38

of those viral moments that spread

2:40

all across the internet and then

2:42

into homes and house parties if

2:45

you missed out on this you

2:47

really missed out on a mid

2:49

2010s cultural milestone because everyone chimed

2:52

in on this from Meryl Streep

2:54

to Matt Damon to better homes

2:56

and gardens to readers digest to

2:58

the National Hot Dog Council and

3:01

yes there is a National Hot Dog Council the

3:03

Today Show ran a piece about it and

3:05

so did the verge and USA Today and

3:07

parade and the Guardian even

3:11

Supreme Court Justice Ruth

3:13

Bader Ginsburg chimed in on this

3:16

saying quote you tell me

3:18

what a sandwiches and then

3:20

I'll tell you if a hot dog is

3:22

a sandwich Stephen Colbert

3:24

who was interviewing her replied

3:27

to that definition by suggesting the definition

3:29

of a sandwich was two pieces of

3:31

bread with any filling in between as

3:34

long as the filling was not also

3:36

bread Ginsburg then

3:38

asked if that included a

3:40

roll cut openly but not

3:42

completely he said yes since

3:45

many subway sandwiches fit that definition

3:47

and the two agreed that yes hot

3:50

dogs were sandwiches and soon

3:52

the Miriam Webster dictionary

3:54

came down definitively on

3:57

this topic stating that yes a

3:59

hot dog is is a sandwich because,

4:01

quote, the definition of

4:03

sandwich is two or more slices

4:05

of bread or a split roll

4:08

having a filling in between. And

4:10

they didn't mention if bread could be that filling, but

4:13

we have to assume maybe so. However,

4:16

many others disagreed with all

4:18

of these sources. Anthony

4:20

Bourdain said the bread in a

4:22

hot dog was just a ballistic

4:24

delivery system for the meat and

4:26

its toppings. And therefore, it was

4:29

not a sandwich in his view. Eric

4:31

Minton, tall, the president

4:33

of the National Hot Dog

4:35

and Sausage Council concurred, telling

4:38

all recipes.com, quote, a

4:41

hot dog is not a sandwich. If

4:43

you go to a hot dog vendor and you say, give me

4:46

a sandwich, they're going to look at you like you're crazy. It's

4:48

just culturally not the same as

4:50

a sandwich. And he

4:53

added, quote, in essence, it

4:55

boils down to a hot

4:57

dog is its own unique

5:00

item that exceeds

5:03

the sandwich category. It

5:05

breaks itself free of

5:08

the sandwich category. You

5:10

saw that debate. What did

5:13

you think of

5:16

it going like, first of all,

5:21

what is your opinion? And then what do you think of the

5:24

public discussing the area of your

5:26

expertise? What is my opinion?

5:29

It's a hot dog, a

5:32

sandwich. I don't

5:34

have a strong opinion on that one.

5:36

But I love

5:38

that that's a question. I love hearing

5:40

people talk about it. I love seeing

5:43

how confident people are that whatever

5:45

their intuitive judgment is, is the

5:48

right thing. And I love seeing people argue

5:51

with each other and be sure they're going to convince

5:53

the other person when I like, I've seen this these

5:55

kinds of debates before. And I know it's minds

5:58

are very rarely changed. for this one.

6:04

I love that debate in

6:07

part because it demonstrates not only

6:09

are people's concepts not aligned, people

6:12

are not good at

6:14

representing that variability. People are

6:16

generally expecting that whatever

6:19

concept they have in mind the other person will

6:21

share. If you think that a hot

6:23

dog is a sandwich, you are sure that

6:25

everybody else should think a hot dog is a sandwich

6:27

and you are offended. If they do not, it

6:29

feels wrong. It feels wrong

6:32

and seeing that play

6:34

out in this debate is

6:36

the most fun part of that debate for

6:38

me. As Kid just mentioned, our

6:41

concepts don't always align and that's

6:43

what we're going to talk about

6:45

in this episode with psychologist Celeste

6:48

Kid, semantic disagreements.

6:50

The notion of talking past

6:53

each other. Kid has a new

6:55

paper which details her research

6:57

into just how misaligned we

6:59

are conceptually speaking, but

7:01

also how unaware of how

7:03

misaligned we are, which results

7:06

in an unsupported confidence in

7:08

the notion that when you

7:10

are discussing just about anything

7:12

with just about anyone, the

7:14

odds that you share the same subjective

7:17

idea, notion, concept, mental

7:20

model, schema of what you are discussing

7:22

is very, very low. For

7:25

instance, in the paper they found that

7:27

when one person says the word penguin

7:30

out loud in a conversation, the

7:32

odds that the other person, the

7:34

person listening, is imagining the same

7:36

concept, the same penguin as the

7:39

speaker, the odds are around 12%.

7:41

Yes, there's a nearly 90% chance that

7:46

the last time you discussed penguins

7:48

with another person, you

7:51

weren't really discussing the

7:53

same idea. And if we aren't

7:56

sharing the same penguin in our

7:59

penguin discussion, questions, then

8:01

imagine what happens when

8:03

we discuss politics or religion

8:06

or philosophy or anything

8:09

even slightly more abstract

8:11

than penguins. But

8:14

before we get into that, Kidd had something

8:16

else to add to the sandwich hot

8:19

dog debate. Have you ever...there's work where

8:21

they ask questions that are even weirder than like

8:23

is a hot dog a sandwich. I feel like

8:25

it's a pretty reasonable question. You

8:27

can ask people weird questions like what's

8:30

a better hot dog? Is it a

8:32

snake or a table? And

8:35

that's a weird question to ask because it is

8:37

like a snake is not a hot dog, a table

8:39

is not a hot dog. But

8:41

you can ask that and people will answer...they

8:44

don't like answering it, but they'll answer it

8:46

and they'll all say, snake, a snake is

8:48

a better hot dog than a table. That

8:51

says something very interesting about our conceptual systems.

8:54

It says something about them being probabilistic. It

8:57

says something about the compositional nature of

8:59

them. So presumably a snake is a

9:01

better hot dog than a table,

9:04

maybe because it involves living

9:07

matter as a kind of gross. It's like meat,

9:09

it's also the same shape as a hot dog.

9:12

But again, it's weird that we

9:15

can answer that question and we can all

9:17

be surprisingly aligned on agreeing that a snake

9:19

is a better hot dog than a table. I

9:24

find these debates enjoyable

9:28

in part because of all of the

9:30

things they call attention to that we

9:32

don't understand about human cognition. I love

9:34

seeing people get very passionate. Again, that

9:36

says something about the

9:39

strengths of our intuitions that are not

9:41

right about when I use the word and

9:43

use the word, people really feel like

9:46

we should be activating the same concept. Even in

9:48

the face of evidence that we don't, as

9:50

people are split on that one, it's

9:52

really hard for us to wrap our head around

9:54

that. It feels like it's wrong. The

10:04

Last Kid is a professor at

10:07

Berkeley and heads up the Kid

10:09

Lab, which carries the torch of

10:11

psychologists like Jean Piaget, Maria Montessori,

10:14

and Lev Vygotsky.

10:17

In her lab, she and her

10:19

colleagues conduct all sorts of experiments

10:21

to better understand how our minds

10:23

engage in learning and create knowledge

10:25

and conjure abstractions and interpret ideas

10:27

and conceive and make sense of

10:29

concepts and schemas and models of reality. They

10:32

use really cool stuff like eye

10:34

tracking and they develop apps and they measure

10:36

all manner of human behavior and communication. It's

10:38

an incredible world. I

10:42

was very excited to get a chance

10:44

to spend some time with Kid because

10:46

her lab recently published a paper that

10:48

is exactly the thing I was looking

10:51

for right now at this moment because

10:53

I'm in the middle of researching,

10:55

I'm in the middle stages of researching

10:57

my new book about what the word

10:59

genius really means, which has me exploring

11:02

linguistics and cognition and intelligence and creativity

11:04

and so on. In

11:06

particular, I'm fascinated with how we

11:08

articulate the ineffable and how we

11:10

come to agree or disagree on

11:12

concepts like what is a

11:15

genius, what does that word mean.

11:17

But also, Kid's newspaper is a great

11:20

topic to discuss as a sort of

11:22

epilogue to the project we just finished

11:24

here on the podcast, the three-part series

11:27

where I interviewed authors of recent books

11:30

about how to have better conversations and

11:32

better disagreements with people who see the

11:34

world differently, which is the topic of

11:36

my most recent book, How Minds Change.

11:39

Those were the three episodes before this one and as

11:41

it turns out, Kid's work has

11:44

a lot to say about that topic from

11:46

a cognitive science perspective and

11:49

Kid's newest paper puts in my

11:51

mind the final nail in the

11:53

coffin of something called the information

11:55

deficit model, one of those old

11:58

ideas about scientific communication. that

12:00

ironically has not stood up to the scientific

12:03

analysis applied to it. In

12:05

that model, there's this concept called

12:08

knowledge deficit, which posits

12:10

that many of the

12:13

problems faced by civilizations arise

12:15

from a lack of access

12:17

to scientific information. And if more

12:19

people had more access to scientific

12:21

evidence, if they just knew the

12:24

facts, then they'd stop being

12:26

so unscientific and superstitious and cult-like

12:28

and extremist and so on, we

12:30

would become less polarized

12:32

and more engaged politically. It's

12:35

a great idea. It's a wonderful dream. But

12:38

the last 100 years of psychological

12:40

research on this topic has revealed

12:42

that the more education you

12:44

gain on any topic, the

12:46

better you become at rationalizing and justifying

12:48

your existing beliefs and attitudes, regardless

12:51

of their accuracy or harmfulness. And

12:53

the better you become at working

12:55

towards your goals, however

12:58

motivated they may be, however

13:00

motivated your reasoning may

13:03

become. And that's what we're going

13:05

to talk about in this episode. All

13:07

that and more after

13:09

this commercial break. Welcome

13:21

back to Money Mania, where we're

13:24

talking tax season. And what to

13:26

do with that refund? If

13:29

you really want your tax refund to have

13:31

a positive impact, I'm bullish on Ohio's

13:33

529 plan. I always

13:36

say invest in something you love.

13:39

Well, if there's nothing we love

13:41

more than our kids, so invest

13:43

in their future and let your

13:45

tax refund help. For as little

13:47

as $25, you can get started

13:49

today at collegeadvantage.com. and

14:00

on delivering your product or your services, the

14:03

more margin you have and the more

14:05

money you keep. Okay, that seems pretty

14:08

obvious, but with higher expenses on materials

14:10

and employees and distribution and borrowing and

14:12

everything else, even the price of bananas

14:14

and cookies and tires for your car,

14:18

you wanna reduce costs and you

14:20

wanna reduce headaches. Everything

14:22

costs more and smart businesses know

14:24

this and that's why they're graduating

14:26

to NetSuite by Oracle. NetSuite

14:28

is the number one cloud financial

14:30

system. It brings accounting and financial

14:33

management and inventory and HR, all

14:35

of this into one platform, which

14:37

gives you one source of truth.

14:40

With NetSuite, you reduce your IT costs

14:42

because NetSuite lives in the cloud with

14:45

no hardware required, so you can access

14:47

it from anywhere. It

14:49

cut the cost of maintaining multiple

14:51

systems because you've got one unified

14:53

business management suite and you're improving

14:56

your efficiency by bringing all of

14:58

your major business processes into

15:00

one platform, slashing manual

15:02

tasks and errors. More

15:05

than 37,000 companies have already done this. They've

15:07

already made the move to this. So

15:09

you can do the math here. You'll

15:11

see how you will profit with

15:14

NetSuite and get this, now through

15:16

April 15th, NetSuite is

15:18

offering a one of a kind

15:20

flexible financing program. To take

15:22

advantage of this through April 15th, go

15:26

to netsuite.com/

15:29

not so smart. Do that

15:32

right now. It's

15:34

NetSuite, N-E-T-S-U-I-T-E, netsuite.com/

15:37

not so smart. One more time, netsuite.com

15:41

slash not so smart. And

15:49

now we return to our program. Hey,

16:06

One of the things that I love

16:08

most about psychology and cognitive science in

16:11

general is that much of the work

16:13

up until now has been taking concepts

16:15

from philosophy and putting them to the

16:17

test, quantifying them,

16:19

seeing if they hold up to scientific

16:21

scrutiny. In the realm

16:24

of epistemology, the philosophical discipline that

16:26

asks, how do we know stuff?

16:29

How do we know anything? And what does

16:31

it even mean to know things? What

16:33

is the nature of knowledge itself?

16:36

Is the question of how

16:38

people's concepts align? That

16:41

is, when I debate politics with you, are

16:44

my concepts of democracy and taxation

16:46

and national defense and so on,

16:48

are they the same concepts as

16:51

yours? But it goes much deeper

16:53

than that. Your concepts of bananas

16:55

and biscuits and beach volleyball, are they

16:57

the same as mine? Are

16:59

your concepts of love and justice and truth,

17:02

are they the same as mine? Here's

17:05

Celeste Kidd. This has been studied

17:08

in various ways over a very

17:10

long period of time. As long as people

17:12

have been trying to communicate with other people,

17:14

they've wondered why is this so hard? Why

17:16

is this so hard? See, there's a

17:19

chance that when we meet to discuss

17:21

an issue, we're just going to

17:23

talk past each other while assuming

17:25

we're not doing that. That we're

17:27

truly connecting and that when we can't get

17:29

on the same page, when we disagree, we

17:33

assume all sorts of other things about the nature

17:35

of our disagreement when it could be as simple

17:38

as our concepts are not aligning. This

17:40

whole idea, it's old, it's ancient. So

17:43

understanding and overcoming all of this could

17:45

be a true breakthrough. But in psychology,

17:47

with the scientific method applied to the

17:49

matter, much of our

17:52

past attempts have, in Celeste Kidd's

17:54

words, stumbled because we don't really

17:56

understand the compositional nature of most

18:00

concepts. In other words, we can't

18:02

yet open the black box of the

18:04

skull and peer into the

18:06

subjective experience of consciousness to measure

18:08

and quantify and map out the

18:11

differences between one person's mental models

18:13

and another's. We know there

18:15

must be variation in our personal

18:18

concepts, in our subjective representations, in

18:20

our interpretations, but all of that

18:23

must be measured indirectly. Until

18:25

now, there hasn't really been a great

18:28

way to do that. There have been various

18:30

attempts to look at to what degree we

18:32

are aligned with one another. One of my

18:35

favorite examples of an attempt to do this

18:37

was Bill LeBeouf. Bill LeBeouf was

18:39

interested in the ways in

18:42

which people's concepts aligned or

18:44

they didn't. He picked

18:46

the domain of cups and bowls because this is

18:49

a domain in which actually you can have pretty

18:52

good intuitions about which dimensions might matter

18:54

for where you draw this category boundary.

18:57

If you think about what makes a cup a cup or

18:59

what makes a bowl a bowl, it's not

19:01

really about the absolute size. It's more about

19:03

the relationship between the height and the width,

19:05

the shorter and spottier and the wider it

19:08

is, the more people tend to say that

19:10

that's a bowl. The taller and skinnier it

19:12

is, the more they tend to say that

19:14

that's a cup. Then there's

19:16

a couple other features that you can add

19:18

that will bias people who think it's a

19:20

cup or a bowl. If you had a

19:22

handle, that's more likely to be assigned the

19:24

cup category than the bowl category. Then this

19:26

is really cool. He also manipulated the context

19:28

in which you show people pictures of possible

19:30

cups and possible bowls. If

19:33

you add a food context, people will accept things

19:35

as a bowl for longer than they will if

19:38

there's not food present. Bill

19:41

LeBeouf is a great example of someone

19:44

who was thinking about conceptual variability and

19:46

was coming at it from a compositional

19:48

perspective. He was trying to figure

19:51

out how do we represent these concepts? How

19:53

can you explore alignments if you don't know

19:55

how the concept is represented? You

19:57

can't really. That's basically the impediment that that

20:00

people came into when they were thinking

20:02

about these questions. So for Cups and Bowls,

20:04

what his work showed is that there is

20:06

quite a bit of alignment at the center

20:08

of the category. There's more disagreement at the

20:10

edges and you can

20:12

manipulate the category boundary. It's not just based

20:15

on the height and the width of the

20:17

bowl. It's also the context in which it's

20:19

occurring in. So he studied this, but

20:23

he was not able to,

20:25

he wasn't really focused on trying to

20:28

quantify how much variability there is. That

20:31

was a question people had asked,

20:33

but had difficulty making progress

20:35

on because of all that

20:37

we don't know about the compositional nature

20:39

of concepts. So like Cups and Bowls,

20:41

heights and widths, irrelevance, how

20:44

do you represent a dog? How do you represent

20:46

the concept of love? How do you

20:48

represent even a table, even

20:51

for concrete objects, it's not clear how

20:54

we're doing that. There's various proposals

20:56

on the table. There's a lot

20:58

of work trying to look specifically

21:00

at that part of the problem.

21:03

Eleanor Roche famously proposed prototype theory by

21:05

which she suggested that how you're representing

21:07

these things is you take all of

21:09

the particular instance of a cup or

21:11

a bowl or a table, you integrate

21:14

those, but what you're storing and what

21:16

you're using in order to judge whether

21:18

something is cup-like or bowl-like is a

21:20

prototypical example of a cup or a

21:22

bowl. So there've been a lot of proposals

21:25

over the years for how we represent concepts,

21:27

for what the compositional nature is.

21:29

If we don't have that solved, it

21:32

was hard to see how you could measure

21:34

to what degree concepts align when we don't even

21:36

know how to imagine how

21:39

they're represented. So Celeste Kidd and

21:41

her colleagues in their new paper

21:44

titled, Latent Diversity in Human Concepts,

21:46

that's what they set out to do

21:48

with their research to solve some of

21:51

these problems, to understand how these became

21:53

great cocktail party questions. Hot dogs, are

21:55

they sandwiches? Is a hot dog more

21:57

a snake or is it more

21:59

a cat? more a table. And in so

22:01

doing, they discovered some really

22:04

unexpected things, one of which

22:06

involves how surprisingly misaligned we

22:09

all are when it comes

22:11

to the subject of penguins.

22:14

Tell me if I'm wrong here. The probability

22:16

that two people selected at random

22:18

will share the same concept about

22:20

penguins is around 12%.

22:25

That sounds about right. It's about that. More often

22:27

than not, when you think of penguins, when I

22:29

think of penguin, we are not activating the exact

22:31

same concept. When you think about penguins,

22:34

and when I think about penguins, and dear listener,

22:36

whoever you are, whatever

22:38

you're thinking about when it comes to penguins, there's

22:41

an assumption that if we all sat down and sort

22:43

of talking about how much we like penguins or how

22:46

cute they are, that we all have the same thing

22:48

happening in our brains. And your

22:50

research suggests that's

22:52

about a 12% chance that that's actually

22:55

happening. Very likely, it's super certain

22:58

that we are not

23:00

actually having the same subjective

23:03

experience of feeling, thinking, imagining,

23:05

and remembering penguins. That

23:07

is correct. There's further

23:09

evidence of exactly why

23:12

in the paper. So we

23:14

did a second study

23:17

where we looked to see, given that we

23:19

can now quantify that your penguin and my

23:21

penguin are probably not the same penguin concept,

23:24

why are our penguin concepts different?

23:27

So what we are not talking about when we talk

23:29

about differences in the concept is differences

23:31

in the context. So we're not talking about like,

23:34

maybe you're thinking about a cartoon penguin, and maybe

23:36

I'm thinking about a penguin I saw at the

23:38

zoo, controlling for the context, your

23:40

overall concept of penguin is fundamentally

23:42

different from mine most of the

23:44

time. And when we

23:46

go back and then try to figure

23:49

out why, we can do that by

23:51

looking at agreement or disagreement across people

23:53

in terms of the features. So

23:56

I can tell you that there are some

23:58

aspects of penguin concepts along the way. which

24:00

people are very aligned. People

24:02

agree that penguins are pointless.

24:04

They agree that penguins are

24:07

cute. They agree

24:10

that they are birds.

24:12

They agree that they

24:14

are not furry. Where

24:17

people differ in terms

24:19

of their concepts are along features like

24:21

the weight of a penguin. So people

24:23

disagree about whether a penguin is

24:26

heavy or light and people are pretty

24:28

much split on that one, which is

24:30

very interesting because it's clear

24:33

in that case that that variability is

24:35

in part coming from a lack

24:38

of experience getting to pick up a penguin.

24:40

If you pick up a penguin, you would

24:42

know something about the weight of a penguin.

24:44

Really to have a good

24:47

sense of how heavy your light penguins are,

24:49

you want to pick up a bunch of

24:51

different penguins with a bunch of different rays.

24:53

People when you ask them are split on

24:55

penguins weight because some people have a very

24:57

strong intuition that they should be heavy because

24:59

they kind of look like they're when they're

25:01

wobbling they're not flying. Maybe they're not

25:03

flying because they're heavy. Other people have

25:05

a strong intuition that penguins should be light

25:07

because they are birds and they have bird

25:09

bones. So this is another thing

25:12

you can bring out of the party when you're trying to

25:14

divide people. I want to go to your cocktail parties where

25:16

you're asking people contentious questions about hot dogs and whether they're

25:18

sandwiches. I was like, do you want to further divide your

25:20

cocktail party? You can ask them whether they think penguins are

25:22

heavy or light. You're invited. You're invited. It

25:28

cannot be understated just how much

25:30

semantic disagreement factors into our

25:33

social, legal, and personal conflicts.

25:35

So understanding scientifically the origins

25:37

of those disagreements is vital

25:40

and important work. Semantic cognition

25:42

is understanding the meaning of

25:45

things. I guess simply put,

25:47

there is substantial variability

25:49

in terms of how people are representing

25:51

different concepts, what they understand them to

25:53

mean. When I say a word and

25:56

you hear the word, when you say the word

25:58

and I hear the word, what the

26:01

paper is showing is that we have

26:03

a tendency to overestimate the degree to

26:05

which our concepts align. The

26:07

paper presents a new method for

26:09

quantifying the degree to which our

26:12

concepts do not align and

26:14

it turns out even for very common concrete

26:17

objects. Yeah, our concepts are misaligned more often

26:20

than they are aligned. In kids

26:23

most recent research, she and her

26:25

team found at least 10 to

26:27

30 quantifiably different

26:30

variations of even

26:32

our most common nouns. And I

26:34

love this because I've been

26:36

interviewing all sorts of experts on this topic

26:39

on how difficult it is to agree

26:42

on definitions and how difficult

26:44

it is to define some

26:46

words in particular, especially the

26:48

abstract ones, what linguists sometimes

26:50

call conceptual shells. Words like

26:52

genius and morality and intelligence

26:54

and curiosity and life in

26:56

the biological sense. And I've

26:59

learned that for each one of those words

27:01

and for many others, a huge part of

27:03

what goes on in any scientific discipline is

27:06

participation on every conceivable level

27:09

in a debate over just what those words

27:12

mean exactly. I've

27:14

been told that one of the quickest ways

27:16

to derail a conference or a meeting

27:18

or any gathering of academics is to

27:21

write a word on a whiteboard or

27:23

put up on a slide, something

27:25

like genes or species or

27:27

measurement, and then suggest

27:30

before we continue, let's

27:32

agree on what that word means exactly.

27:36

This rarely goes well and can send

27:39

people into an hour-long debate until they

27:41

finally go, what are we doing? Psychologist

27:43

Andy Luttrell told me that much of

27:46

psychology, especially in the early decades, was

27:48

mostly just a debate around definitions

27:51

and he pointed me toward a

27:53

book by Kurt Danziger titled, Naming

27:55

the Mind, How Psychology Found Its

27:57

Language, which is about this very

27:59

thing. So, I asked

28:02

the kid about how all of that might

28:04

have influenced her decision to get into

28:07

this line of work. I

28:10

think it started for

28:12

me with wanting

28:15

to understand what was true. My

28:17

background was actually not in

28:20

science. I didn't grow up thinking I was going to be a

28:22

scientist. I thought I was going to be an investigative reporter. I

28:25

had Associated Press reporters in my family, and

28:27

I thought the highest calling was going out,

28:30

finding out what's true in the

28:33

world, especially discovering truths that are

28:35

reflective of people

28:37

that don't usually have a voice. You

28:40

find what's true, you tell people

28:42

what's true, and the world changes,

28:44

was my view of things. I

28:48

went to college. My first majors were computer science

28:50

and journalism. I thought I was going to do

28:52

that. It doesn't take much

28:55

interacting with people about

28:57

a topic that you care about

29:00

before you realize it's really hard

29:02

to align in terms of values,

29:04

in terms of concepts. I

29:07

think a lot of the inspiration for my

29:09

work was arguments I had with editors over

29:12

the years about, like, no, this is important.

29:14

Also, you've edited my piece, and this is

29:16

not what I meant. You

29:18

still have the right idea. I'm

29:21

very interested in what

29:23

are the psychological limits to

29:26

how close we can get to objective

29:28

truth in the world. How

29:30

can we get better access to what

29:32

is true through coordination with one another?

29:35

That necessarily involves having conversations with people

29:38

and trying to align in terms of

29:40

what conceptual representations you're thinking of. You're

29:42

trying to, when you talk to somebody

29:45

and coordinate with them, get

29:47

the same image and concepts and

29:49

ideas in their minds as what you were

29:51

thinking. That's really hard. I

29:54

was interested in this piece

29:57

in exploring what... uh

30:02

exactly how hard of a problem

30:04

is that like it feels like it's very hard

30:06

uh how much are people online um uh this

30:08

is an age-old question and uh yeah

30:10

we wanted to we wanted to see you had

30:12

this beautiful chart i'm looking at it right

30:15

now of like and i love like visualizing

30:17

this like there's this sort

30:19

of nucleus of things

30:21

we would all kind of say

30:23

are penguin-ish penguin-like but

30:26

as you move away from

30:28

that nucleus into the the electron cloud

30:31

of penguin-ness it's very very like people

30:33

will not agree with all the other

30:35

concepts that we may have about this

30:37

thing and i will i dig this

30:40

because you show this for

30:42

every concept like whales salmon barack

30:45

obama uh people

30:47

have sort of a small

30:50

number of things that we would

30:52

agree upon as the features

30:55

of this concept and

30:58

then we have a much larger range of things we

31:00

would not agree about and

31:03

if i'm thinking correctly

31:05

that that leads to an assumption that

31:07

my gigantic word cloud association

31:09

super network whatever metaphors we need to

31:12

make sense of this i

31:14

have this intuition that mine must be pretty much

31:16

more similar to yours than it actually is and

31:19

i'm wondering the

31:24

two people who like work in if somebody

31:26

who works in a zoo in san diego

31:28

and somebody who works in a zoo in

31:30

atlanta georgia who both uh deal

31:32

with penguins every day um and

31:35

a veterinarian who treats penguins specifically

31:37

if that is such a person

31:40

like i would imagine their overlap is much

31:42

greater right

31:45

yeah that is a prediction you would expect that

31:47

is not something that we test in the paper

31:50

but there's a lot of other

31:53

evidence that the beliefs

31:55

that we form are a byproduct

31:57

of the experiences that we have

32:01

This is a really simple

32:03

but really profound aspect of

32:05

human psychology. The

32:08

world is super big and you're forming

32:10

beliefs about the big wide

32:12

world necessarily on a very

32:15

sparse sample. That's

32:17

a subset of it, a really small subset.

32:19

So because all people are living very different

32:21

lives and having very different sets of experiences,

32:23

they're taking very different samples and you'd expect

32:26

that that is the origin point

32:28

for why there is variability. I

32:30

would add though that while

32:33

you would expect two people who

32:35

work at two different zoos to have

32:37

borderline concepts on penguins, two people, even

32:39

if they're raised in the exact same

32:41

environment, you would not expect for, I

32:43

would not expect for them to have

32:45

exactly the same concepts. I sometimes use

32:47

the very dramatic example

32:49

of conjoined twins. Two

32:52

conjoined twins, even though they

32:54

are physically in the same space all

32:56

of the time, they have two different minds, two

32:58

different sets of eyes. They are

33:00

going to be sampling differently even though

33:02

they're in the same physical environment.

33:05

We know that the way in

33:07

which people sample from the world follows

33:10

from the samples that they've accumulated

33:12

previously and especially to people that

33:14

are spending a lot of time

33:16

together again, conjoined twins is a

33:18

dramatic example, they tend to specialize.

33:20

So if you live with somebody,

33:22

people report acquiring very specialized knowledge

33:24

so that one person will know

33:26

what trash day is. The

33:29

other person will know something else about how to

33:31

clean the gutters and when one person

33:33

is out of town, the other person might be

33:35

surprised that they don't have some set of knowledge

33:37

and it's because it doesn't make sense to have

33:39

redundant knowledge. People specialize in that way. Even

33:42

two people that were physically in the same

33:44

space all of the time should be taking

33:46

slightly different samples from the world and I would not expect

33:48

we have exactly the same concepts as a result of that.

33:58

Let me briefly run through the paper. talk

34:00

about it. All right, so you

34:02

got about 3,000 people, nearly 3,000 people, and you divided

34:06

them in half. I'm looking at my notes.

34:08

Yeah. And then you, what did you do with

34:10

those people? Yeah, so

34:13

we got around the problem that other

34:15

people had had in trying to quantify

34:17

how much concepts align. The primary

34:21

impediment to doing that in

34:23

the obvious way you think to do it first

34:25

is like, well, let me understand something about the

34:27

compositionality. That's hard. We can't do that. So

34:29

we skipped that step and

34:31

instead had people

34:34

make similarity judgments across

34:36

common concepts in two

34:39

domains. So the two domains we chose

34:41

were common animals. And then

34:43

we wanted something that was like common

34:46

animals. We want something that's animate, but

34:48

we want something about which people may

34:50

have different

34:52

types of judgments. We want something kind

34:54

of moralistic. So we also took US

34:56

politicians and had people make

34:59

similarity judgments in that domain too. Each

35:02

person was assigned to one target. So let's

35:05

say you have chicken. You make

35:07

similarity judgments like what's most similar to a chicken?

35:09

Is it a dolphin or is it a finch?

35:12

And we do that for

35:15

all, we do all of those pairwise comparisons

35:17

for each of the concepts that we're testing.

35:20

What you get out is a vector

35:24

of responses about what you rate to be

35:27

most similar to, for example, a

35:30

chicken. And the intuition

35:32

there is if it's the case that we

35:34

all have crudely pretty much the same

35:37

concept, all of your similarity

35:39

judgments should be the same also. So this allows

35:41

us to get at your conceptual

35:43

representation without knowing the details of what

35:45

the composition of your concept

35:48

is. So that's the hack that we

35:50

use in order to get at this. Once

35:53

we have those response vectors, we

35:55

can now perform clustering over them in order

35:57

to figure out what we're

35:59

testing. figure out how many latent versions

36:01

of each concept exist in the population.

36:04

And so we can see things like for

36:07

this concept, we have two distinct clusters. For

36:09

this concept, we have ten distinct clusters. And

36:13

what we found was that there's a substantial amount

36:15

of variation. There are quite a

36:17

number of different clusters for

36:19

both the more

36:22

concrete concepts and for the more abstract ones.

36:25

So you have these people, you've divided

36:27

them up, you've had them

36:29

talk about animals. You also have this

36:32

other thing where you had them list

36:34

ten adjectives and rate features.

36:37

I love the idea that seals are not feathered,

36:40

but they are slippery. This is something most people

36:42

think. But not everyone agrees whether

36:44

that seals are graceful. Some people think see

36:46

them as graceful and some do not. And

36:50

then with Donald Trump, you had like most

36:52

people agree that this is a man who

36:54

is not humble and that he is wealthy.

36:57

But they disagree quite a bit over whether or not he is

36:59

interesting, which is I love how nebulous the

37:02

term interesting is. What

37:04

did you what did you learn from this?

37:07

Did this these these two phases of the

37:09

research, what stuck out

37:11

to you? Yeah, so so

37:13

there's there's a lot that we learned

37:16

and a lot that was surprising. I'll

37:19

start with the first set of things. So

37:21

the like doing the clustering over people similarity

37:23

judgments showed us that there's a substantial amount

37:25

of variability in both concrete

37:28

and abstract concepts. There's

37:30

a second part of that first experiment, which

37:32

is we also ask people for their judgments

37:36

on how likely they thought somebody was to

37:38

share their concepts. And based on

37:40

that, we wanted to know

37:42

are people well calibrated to the likelihood

37:45

that somebody is aligned with them? And what we found out

37:47

from that is that they are not. When

37:50

you are activating a concept, when you are

37:52

saying a word, you generally expect

37:54

that someone else will share that concept, will

37:56

activate the same concept when you say the

37:58

word. you

38:01

think that's more common than it actually is.

38:03

Most of the time, as we talked about,

38:06

someone else is not activating the same concept, even for

38:08

the same word in the same context. So

38:11

that was useful. The second

38:13

analysis where we ask people for features

38:15

over the super set of words. So

38:17

we ask people to generate a bunch

38:19

of features for seal

38:22

and for penguin and for all of the

38:24

other words that

38:26

we have in our animal sets. Then we

38:29

ask a separate group of people to rate

38:32

whether or not each feature applied

38:34

to each concept. And that's

38:36

how we get those plots that look like kind of

38:38

word clouds. Something

38:40

interesting in those is that tells us something

38:43

about why we have distinct clusters. So some

38:45

things stand out like people disagreeing about whether

38:47

or not penguins are heavy or light. But

38:51

something else that was really interesting to us in

38:53

those plots is that for

38:55

the politicians, there's

38:57

not just information in terms of what

39:00

features are contentious and what features are

39:02

agreed upon. There's also information

39:04

in the distribution of features. So some

39:06

of the politicians have quite

39:09

a lot of features that are in the middle, where there's

39:11

a lot of disagreement about a lot of features. An

39:13

example of someone like that is Biden. People

39:16

really are not on the same page. Most

39:18

of the features are towards the middle. You

39:20

have very few features that everybody agrees do

39:23

or don't apply. For

39:25

Obama, Obama is often held up

39:28

as an example of like a

39:30

polarizing politician. There

39:33

are very few features in the middle, which

39:35

was surprising to me when I first thought.

39:38

Biden, everybody agrees, is

39:42

honest that he's intelligent. He's

39:44

hardworking. People are on the same page about that.

39:47

They also agree on

39:49

the features that do not apply to him.

39:51

There's very little in the middle. So what

39:54

this means is that although

39:56

people feel very different ways

39:59

about Obama, that

40:02

disagreement about whether

40:04

or not you like Obama is not

40:06

originating from people disagreeing

40:08

about what Obama is or isn't. People

40:11

are generally on the same page and have a

40:14

more aligned in terms of their conceptual understanding of Obama.

40:17

That is super fascinating. Think

40:19

of the information deficit hypothesis, right?

40:21

That our disagreement lies in that

40:24

we don't have all the facts. You

40:27

just need some more of the facts and then

40:29

we both see this the same way. Yeah, I

40:31

think that's a great demonstration that it's not that

40:33

simple. One reason why we

40:35

might disagree is because we're activating different concepts.

40:38

We disagree about what's true, but

40:41

that's not the only reason we can disagree. We can

40:43

be on exactly the same page and still have disagreement

40:45

for other reasons. That is amazing. So, from

40:47

here you're correctly, in

40:49

your study, most people have a pretty

40:53

similar, as similar as you

40:55

can get considering all the things we discussed, concept

40:58

of... Well, I'll put

41:00

it this way. They

41:03

agree on what Obama is. Yes, it isn't.

41:07

But when you ask... But they don't agree on...

41:10

Considering this is what Obama is, they don't necessarily

41:12

agree on how to feel about that. Yeah, that's

41:14

right. And whereas

41:17

Biden is interestingly

41:21

inverted in a way, we don't all agree on what

41:23

Biden even is. Yeah, that's right. Yeah,

41:26

that's exactly right. We have not today

41:28

used this method to chart changes in

41:30

alignment or opinion over time, but this

41:32

is a method that you absolutely could

41:34

use to do something like that. You

41:36

could use this to see

41:38

if people's concepts become more or

41:40

less aligned over time. You

41:43

could also use this potentially to see

41:45

if some kind of intervention got people more

41:47

on the same page. It has

41:49

potential use beyond what we've used it

41:51

for in this paper. Your paper gives me

41:54

a new material for this

41:56

because I love the idea that, oh, we

41:58

can very much... agree

42:00

completely. We can both have PhD-level

42:02

understandings of an issue, but

42:05

completely disagree. We agree on what is,

42:07

but we don't agree on anything else.

42:09

I think that's incredible. I love it.

42:12

This feels like the haymaker to the

42:14

information deficit I bother. Yes.

42:18

Okay. Before we go into the final segments of

42:22

this episode, to sum up, kids were not

42:30

aware of the fact that the work

42:32

has supplied ample evidence to suggest that

42:34

there is tremendous variation in our internal

42:36

concepts of reality itself. We

42:38

may share words and we may share the

42:40

same dictionaries and textbooks, but thanks

42:43

to all the variations in our

42:45

experiences and differing levels of ignorance

42:47

and expertise and a vast array

42:49

of variable cultural influences, the

42:52

likelihood, even if we share the

42:55

same language, if we share the same country, the

42:57

same hometown, the

42:59

same household, the

43:02

likelihood that your concept of something

43:04

like, say, penguins is

43:06

identical to mine is

43:09

extremely low. Also,

43:11

her work suggests you can think of a

43:13

word and a concept it refers to as

43:16

a sort of atom-like

43:18

entity. There's a nucleus

43:20

of shared ideas surrounded by

43:22

an electron cloud of variations.

43:25

Some words have more variations than

43:27

others, but no word

43:29

is without a lot of variation.

43:31

So that means even for words

43:33

with broadly shared definitions, there's

43:36

a near zero chance that any one of us

43:38

shares an identical copy of that core

43:41

concept that the word

43:43

generates when conjured

43:45

up and contemplated by any

43:47

one particular brain. And here's

43:49

the most important aspect of this research for me.

43:52

The evidence suggests we

43:54

are all mostly unaware of this

43:57

variation and not just unaware. believe

44:01

the opposite is true. We

44:03

erroneously believe most other people

44:05

share our ideas, our

44:08

concepts and models and mental imagery, our

44:11

semantics. And as

44:13

the paper says, quote, this

44:15

points to one factor that likely

44:17

interferes with productive political

44:20

and social discourse. This

44:23

presents such an existential thing for me.

44:26

It's like to accept what you've got

44:28

here, your research and like your expertise

44:30

is to accept that we're living in

44:33

a grand illusion. It's a best

44:35

guess. Yeah, but like it's all

44:37

these separate subjective realities working

44:40

together. Right. And

44:43

there's this grand illusion that we're

44:45

all living in the same shared

44:48

space, but we are very divided

44:50

by our conceptual understandings. And

44:53

yet we still seem to manage somehow. How

44:56

people work together to create the technology for us

44:58

to be having this conversation. Somehow

45:02

they created the academic institution in which

45:04

you are employed. Yeah. What do

45:06

you what are your thoughts on like, if there's

45:09

so many concepts that you found just in

45:11

this one paper, one

45:13

one brain is interacting with another brain and

45:15

they do not have identical concepts about the

45:17

things they're discussing. How are we managing to

45:20

get anything done if we're all living in

45:22

these private universes? So that's

45:24

a really, really great question that

45:26

I love thinking about. And I

45:28

don't think we have a full answer.

45:31

We're talking, we're using a learned

45:35

symbolic system, which

45:37

is English, which is allowing us

45:39

to at least crudely align our

45:42

ideas with one another. We

45:45

don't know exactly to what

45:47

degree there may be variability, but

45:49

every time we get closer to

45:51

looking at it, like in this paper, we look at

45:53

just like concrete now in

45:55

our paper, we find there's way more variability

45:57

than we'd expect. So what follows?

46:00

from that is that I should expect that we are not

46:02

as aligning as well as it feels like we're aligning

46:04

when we're talking. And

46:06

yet we're able to get stuff done. You're going

46:08

to record this podcast episode. We're going to leave

46:10

this conversation with new ideas that we didn't have

46:12

before that we got from one another. They

46:15

may not be the ones that the other one intended for us

46:17

to get, and that's an important thing to appreciate. It

46:20

could be the case that not being

46:22

perfectly aligned in terms of the

46:24

way that we think about the

46:26

world could benefit us in some

46:29

ways. One of the things

46:31

that I really, really love about

46:34

humans as a species is that we're

46:36

very weird. We have curiosity.

46:39

Other animals, other primates, other species

46:41

have curiosity. But human curiosity

46:45

is fundamentally different

46:47

in that we take it to extremes.

46:50

If you think about what we spend

46:52

doing in a day alive, it is

46:54

indulging our curiosity. We find out stuff

46:56

that we don't need, obviously, for

46:58

getting food or for reproducing. We spend a lot

47:00

of time doing that kind of thing. We're

47:03

willing to pay money for information sooner rather

47:06

than later. I'll pay $5 for an episode

47:08

of something to see how a story resolves

47:11

rather than waiting to see it free a month

47:13

later. We are very, very curious. Our

47:16

curiosity, the way that it works,

47:18

is we tend to want to

47:20

build on the ideas and concepts that

47:22

we already have some basis in. When

47:25

you're little, you don't know anything. Everything

47:27

is great. Anything you find on the carpet is very

47:30

entertaining for a baby. That's

47:32

because they don't have any base knowledge.

47:35

Once they start learning stuff about the world, the first

47:38

stuff they learn biases them to want

47:40

more information on those same topics. A

47:44

degree that is not true in

47:46

any species that I know of

47:48

specialize such that we

47:50

have people that know just

47:52

about the costuming

47:54

of Victorian

47:56

era blacksmiths. We

47:59

have people that specialize in blacksmiths. very, very specific things. I

48:01

mean, you have people that know all about

48:03

baseball statistics just from the 1950s from New

48:05

York. That is a byproduct

48:07

of the way our curiosity systems function.

48:10

And what that means

48:12

is two things. Number one,

48:14

individually, we're not very useful for

48:16

surviving in the world. One person

48:19

is not good at surviving in

48:21

the world by themselves. But two, as a

48:24

population of people, we

48:27

have way more breadth than other species in terms

48:29

of what we know as a group. So if

48:31

you have a bunch of different members of a

48:33

population that all specialize in different things, you

48:36

bring them together. Our strength is

48:38

in the variability in terms of our knowledge

48:40

and our concepts. We can do incredible

48:42

things. We can make spaceships. And we can shoot them

48:44

off to the moon. We can send a person

48:46

to the moon. We can build laptops. We

48:48

can do science. We can have this podcast.

48:51

Our strength as a species, I would argue, is because

48:53

of the variability in our knowledge. And

48:55

part of the variability in that knowledge is

48:57

also the variability in our concept. So it

49:00

may be a feature instead of a bug.

49:16

That is it for this

49:18

episode of the You Are

49:20

Not So Smart podcast. Celeste

49:22

Kidd's website is kidlab.com. That's

49:25

with two d's, k-i-d-d-l-a-b.com. Her

49:27

Twitter account is at Celeste Kidd,

49:30

C-E-L-E-S-T-E, K-I-D-D.

49:34

And over at

49:36

Berkeley, it's psychology.berkeley.edu/people

49:38

slash Celeste dash

49:40

K-I-D-D. For links

49:42

to everything that we talked about, including

49:44

the research paper, which is titled Latent

49:46

Diversity in Human Concepts, head

49:49

to youarenotsosmart.com. Or check

49:51

out the show notes inside your podcast player. There will

49:53

be links to stuff in there too. My

49:55

new book, How Minds Change. You can find links

49:58

to How Minds Change in the show notes. right

50:00

there in your podcast player. And

50:02

you can go to the homepage for how mine's changed. We

50:05

can find a round table video with a group of persuasion

50:07

experts featured in the book. It could

50:09

be a sample chapter, download a discussion guide, sign

50:11

up for the newsletter, read reviews. You

50:14

can also check out some of the podcasts that I've been on promoting

50:16

it. And I should

50:18

note that I did the audio

50:20

book for it and it was really fun and

50:22

I enjoyed doing that. I don't promote that enough.

50:25

For links to all the past

50:28

episodes of this podcast, go to

50:30

Stitcher's SoundCloud, Apple Podcasts, Amazon Music,

50:32

Audible, Google Podcasts, Spotify, or youarenotsosmart.com.

50:35

Follow me on Twitter at David

50:37

McRaney. Follow the show at Not

50:39

Smart Blog. If you'd like to

50:42

subscribe to the newsletter, it's over

50:44

on Substack. The title of the

50:46

newsletter is disambiguation. We're

50:49

also on Facebook slash YouAreNotSoSmart over there.

50:51

And if you'd like to support this

50:53

operation, help make it better, help pay

50:55

for transcription and other features, help pay

50:57

for me to fly to places and

50:59

interview people, go

51:02

to patreon.com/ youarenotsosmart.com.

51:05

Pitching in at any amount will get you the show

51:07

ad free, but the higher amounts

51:09

get you posters, T-shirts, sign books, and

51:11

other stuff. The opening

51:13

music that's Clash by Caravan Palace.

51:16

And the best way to support the show is just tell everybody

51:18

you know about it. If any episode

51:20

really landed for you, send

51:22

them a link to that. And check back

51:24

in about two weeks for a

51:26

fresh new episode. Time

51:39

for a quick break to talk about McDonald's.

51:41

Mornings are for mixing and matching at McDonald's.

51:43

For just $3, mix and

51:45

match two of your favorite breakfast items,

51:47

including a sausage McMuffin, sausage biscuit, sausage

51:49

burrito, and hash browns. Make it even

51:51

better with a delicious medium iced coffee.

51:53

With McDonald's mix and match, you can't

51:56

go wrong. Pricing and participation may vary,

51:58

cannot be combined with any other. offer

52:00

a combo meal. Single item at regular price.