00:00:01
Speaker: This podcast is brought to you
by Knowledge@Wharton.
00:00:13
Angie Basiouny: Welcome to Knowledge@Wharton.
I'm Angie Basiouny. I'm here
00:00:15
with Wharton Marketing Professor
Jonah Berger, and we're going to
00:00:18
talk about his latest research
paper. It's titled "Topography of
00:00:21
Thought," co-authored with Olivier
Toubia, who's a professor at
00:00:25
Columbia Business School. Now
this research is about how
00:00:28
someone's writing style can be
indicative of their future
00:00:32
success. Jonah, thanks for
coming on.
00:00:35
- Jonah Berger: Thanks
so much for having me back.
00:00:36
- So I really was hoping you would
give us an overview of this
00:00:39
paper by way of explaining your
title, which is great. What is
00:00:43
typography of thought?
00:00:45
- You know, I don't have to tell
you that we all use language all
00:00:48
the time, right? We write
emails, we make presentations,
00:00:54
we write college essays when we're
younger, we write job
00:00:57
applications when we're older.
We use language all the time.
00:01:00
And people have often thought
that language in some sense is a
00:01:03
fingerprint. It reveals or
reflects something about the
00:01:06
people that produce it. So
people that use a certain type
00:01:09
of language, for example, it
might indicate something about
00:01:11
them. There's research, for
example, that shows we can guess
00:01:14
how much someone is an
extrovert, for example, about
00:01:16
the language that they use when
they -- when they talk or write.
00:01:20
But one thing we wondered is,
okay, beyond the individual
00:01:22
words people use, whether they
use a word, for example, that
00:01:25
indicates they do extroverted
things, might the power of ideas
00:01:29
they put out there, the way the
ideas they share organize
00:01:32
beyond the words, but the way
they organize ideas, tell us
00:01:36
something interesting about
them and their likely future
00:01:38
success. And so I think it's
important to talk a little about
00:01:41
what I mean about the pattern of
ideas, right? When someone talks
00:01:44
about something, they can cover
a small amount of ground, or a
00:01:48
large amount of ground. If you
ask someone about their work
00:01:51
history, for example, they can
talk about a variety of
00:01:54
different things they've done,
or a smaller set of things
00:01:57
they've done. So they can cover a
lot of ground or a little bit of
00:02:01
ground. If you want to use an
analogy here, you can almost
00:02:03
think about someone going for a
run, for example. You know,
00:02:06
someone can go for a run and go
all the way around the city,
00:02:09
or they can take the same
length of run, they can go for
00:02:11
the same number of miles, but
just go around the block a
00:02:14
number of times. And so in both
cases, they did the same
00:02:18
distance of things, right? But
they covered more ground in one
00:02:21
than the other. And so one way
we express ideas is the amount
00:02:24
of ground we cover. We cover
more ground with our ideas or
00:02:27
less. We can talk about sort of
more ideas, more topics, more
00:02:30
themes, more things that are
disparate from one another, or
00:02:33
things that are related to one
another. But it's not just that.
00:02:36
It's also the speed with which
we move between adjoining ideas.
00:02:40
So imagine -- think about a movie,
for example. A movie could cover
00:02:43
more or less ground across the
whole course of the movie. But
00:02:46
it can also move faster or slower
between ideas, right? If a movie
00:02:49
has one scene, for example,
that's in a city, let's say, and
00:02:53
it's the beginning of a wedding,
if the next scene is the later
00:02:56
part in a wedding, that's related
to the first part, right?
00:02:59
Different things may happen, but
it's -- it's pretty closely
00:03:01
related. Whereas if you jump
from a wedding, for example, to
00:03:04
a completely different city and
an action scene, that would be
00:03:08
really different sorts of ideas.
They're not very related.
00:03:10
They're moving further across
those two points. And so we
00:03:14
wonder, are these two ideas, the --
the ground cover, or sort of
00:03:17
the volume of of content someone
produces, does it cover a lot of
00:03:21
ground or not a lot of ground?
And how quickly it moves. Does
00:03:25
it move very quickly between
ideas that aren't so related, or
00:03:29
does it move more slowly
between ideas that are more
00:03:31
related? Could that tell us
something about their likelihood
00:03:34
of being successful in the
future?
00:03:36
- So I just want to be clear,
you're not talking about
00:03:39
like the length of the piece of
writing, you're not talking
00:03:41
about whether it's a 500 word
essay, or a 1,000 word essay.
00:03:44
It's really about how they use
that space to move through their
00:03:48
ideas, correct? - Really --
00:03:50
yeah, really great points. It's
not about the length,
00:03:52
it's about the ground. And so I
think that ground sort of makes
00:03:55
sense, right? Are they covering
a lot of ground in their 100 or
00:03:58
1,000 words or whatever it is, or are
they covering less ground in
00:04:01
their 100 or 1,000 words, right?
You could imagine if someone
00:04:04
tells you about their vacation,
they could cover a lot of ground
00:04:07
talk about a variety of
different things, and use 1,000
00:04:10
words to do it, or they could
talk at length about a smaller
00:04:13
set of things, right? They could
only talk about the restaurants
00:04:15
they went to. It might be different
restaurants, but they'd be more
00:04:18
closely related than talking
about the restaurant they went
00:04:21
to, and the hotel they went to, an
amusement park they did, or a day
00:04:24
they had at the ocean. Those more
disparate, less related things
00:04:28
would involve covering more
ground even using the same
00:04:30
amount of
00:04:31
words to do it. - So how did you go
about studying this?
00:04:34
- Yeah, so we looked at this in a really
consequential setting, which is
00:04:39
college application essays. And
to be clear, we're not
00:04:41
particularly interested in
college application essays. The
00:04:44
same ideas could be applied to
job application essays, they
00:04:47
could be applied to in other
work we're doing at the moment,
00:04:50
for example, apply to things
like online reviews. In the
00:04:54
past, we've looked at things
like movies, books, and
00:04:57
television shows, and the content
of those things as well. So, same
00:05:01
ideas can be applied to any type
of content. But here we looked
00:05:03
at college application essays. And so
we took 40,000 college
00:05:07
application essays from a
variety of different folks that
00:05:09
were applying to school. And we
looked at what they wrote, and
00:05:13
the topography of thought of
what they wrote, sort of how much
00:05:15
ground they covered in that
essay. Again, similar length,
00:05:18
but how much ground they
covered. And how quickly they
00:05:21
move between ideas. And we
looked at whether that related
00:05:24
to their future success. In a
sense, how well did they do in
00:05:27
school? What was their GPA once
they got there? And we were
00:05:30
interested in seeing whether,
hey, does how they write, and
00:05:33
again, not just the individual
words they use, but the way they
00:05:36
express ideas, might that reveal
something or predict how likely
00:05:40
they are to be successful in
00:05:41
the future? - That is a really
interesting dataset. What did
00:05:43
you find?
00:05:45
- Yeah, so we found two -- two very
important things. So first,
00:05:49
covering more ground. So again,
that idea of sort of doing --
00:05:52
covering a broad array of
things in the same amount of
00:05:55
length was linked to greater
success. But importantly, doing
00:05:59
so while moving rather slowly
was also important. So you could
00:06:03
think about covering a lot of
ground, right? Almost think about
00:06:06
the points on a clock, think
about the numbers arrayed on a
00:06:08
circular clock, right? You could
cover a lot of ground, either by
00:06:12
moving in a circle, right? Going
sort of from twelve to one to two
00:06:15
to three to four, around that
circle, which wouldn't require
00:06:18
much movement at each step, right?
At each step it would just be a
00:06:20
little. Or you could traverse
that same aground going from
00:06:23
like twelve to six to one to seven
to three to eleven. And again, you'd
00:06:29
cover the same amount of ground,
you'd have covered the surface
00:06:31
of that regular sort of circular
clock, but you'd have taken much
00:06:35
more of a longer route between
each individual point. And so what we
00:06:39
found is that folks that are
successful in school are able to
00:06:42
blend these two things that
might seem mutually exclusive,
00:06:45
right? It might seem like
covering a lot of ground
00:06:47
requires moving really quickly
between points to get to a lot
00:06:50
of different points. But folks
doing well in school figured out
00:06:53
a way to cover that ground
really efficiently. They're able
00:06:56
to do so by moving slowly
between these points so they
00:06:59
don't have to take a lot of big
jumps along the way.
00:07:03
- I think that's a really
important note for anybody who
00:07:04
does writing. As an editor, I
always tell folks, write until
00:07:08
the car runs out of gas, don't
get out and push. So it's this
00:07:12
idea that your thoughts are more
concise, and you move from one
00:07:15
topic to the other, you cover
that ground efficiently. Tell me
00:07:19
a little bit about, you in this
paper, in this experiment, you
00:07:22
controlled for some socioeconomic
factors. Can you touch on that a
00:07:25
little bit? I think that's
important.
00:07:28
- Yeah, so someone could wonder,
okay, so you found that people
00:07:31
did well in school, and you
found that they wrote a certain
00:07:34
way, but is that indicating
they're thinking the way they're
00:07:36
thinking, right? You're
using writing as a way to
00:07:38
indicate sort of how they think?
Or does it indicate something
00:07:41
else, right? Maybe it's just, hey,
people who do better on the SATs
00:07:45
also have higher grades, and
also tend to write a certain
00:07:47
way. Maybe that, hey, people who
have parents that are more
00:07:51
educated tend to write a
certain way, and also do better
00:07:54
in school. Or maybe people that
have parents that are more
00:07:57
educated can afford to pay for,
you know, an essay consult, who
00:08:01
helps them write a certain way,
and also helps them do better in
00:08:03
school. And so it's -- it's driven
by that. And so we controlled
00:08:06
for a variety of different
things. We controlled for what
00:08:09
they wrote about, right? So
maybe certain types of people
00:08:11
tend to write about certain
types of things rather than
00:08:13
other types of things. And maybe
it's not about, you know, how
00:08:16
they write, it's about what they
wrote about, the topics or themes
00:08:19
that they discussed. Nope, it
wasn't that. Maybe it's parent's
00:08:22
education. Nope, we controlled for
that. Maybe it's SAT score.
00:08:24
Nope, we controlled for
that. And so what this suggests
00:08:27
is that the topography of
thought goes beyond things
00:08:30
related to just socioeconomic
factors or family background.
00:08:33
It's not just that people who
might have had wealthier
00:08:36
families, for example, tend to
write a certain way, or have
00:08:39
application consultants and do
better in school because they
00:08:42
get tutoring as well. No, it
really suggests, hey, writing
00:08:46
reveals something about the way
we think, which can reveal or
00:08:49
predict our likelihood of being
successful in the future.
00:08:52
- Let's put this in the
business context. Because there
00:08:55
are other really critical forms
of writing that we do every day
00:08:57
in business, like cover letters,
resumes, anything, press release,
00:09:02
a communication to the c-suite.
How can you take this research
00:09:07
and translate it into that
context?
00:09:09
- You know, what I find
fascinating about these ideas is,
00:09:13
yes, we looked at the case of
college application essays in
00:09:16
this particular case. But it
doesn't have to be only about
00:09:19
application essays. These same
ideas should apply more broadly
00:09:22
to a variety of different sorts
of contexts. Whether it's a
00:09:27
cover letter that someone
writes, whether it's an online
00:09:30
review that someone puts
together, whether it's the
00:09:33
emails they write at the office.
All these things provide insight
00:09:37
into who people are, and what
they're likely to do in the
00:09:40
future. I think on a previous
episode when you had me on, I
00:09:42
talked about a paper I loved
recently where they can tell
00:09:45
whether someone's going to
default on a loan or not by the --
00:09:48
by the language they use in
their application. So sometimes
00:09:51
when applying for a loan, you
need to write -- particularly in these
00:09:54
online sort of marketplaces,
peer to peer loans, you have to
00:09:56
write a little bit about why
someone should give you money.
00:09:59
And you can predict -- beyond
controlling for all those
00:10:01
financial factors, control for
those, you can predict how
00:10:04
likely someone is to default
based on the language they use
00:10:07
in their application essay. Or
similarly, you can predict
00:10:10
whether someone's going to get
promoted, or fired, or leave
00:10:13
a job for better opportunity
elsewhere based on the language
00:10:16
they use in their email. And so
most of this work that I just
00:10:19
mentioned uses individual words.
But I think what our work
00:10:22
suggests is beyond the
individual words someone uses,
00:10:24
you can get insight into who
they are, how they think, and
00:10:28
how well they're going to do in
the future, based on the pattern
00:10:30
of ideas that they have, or
their topography of thought. - Yeah, the
00:10:33
manager gives you an indication
of how they're going to do their
00:10:36
work. You know, how they move
through their written work
00:10:39
is indicative of how they
might move through their
00:10:41
physical work or their knowledge
work. But --
00:10:43
- Yeah, one thing we're
looking at right now actually is
00:10:46
looking at, as people learn more,
does that change the way their
00:10:50
topography of thought works, right?
- I would hope so.
00:10:52
- So obviously, as we gain more knowledge
in a given domain, we may talk differently,
00:10:56
we may think about ideas
differently. One thought we have
00:10:59
is, well, hey, if if people who
are able to cover a lot of
00:11:02
ground really efficiently by
moving slowly between points,
00:11:05
right? Not moving quickly
between points. If that's true,
00:11:08
wait, how did they get there?
Are they naturally that way?
00:11:11
Probably not, right? Probably
they may have gained more
00:11:13
knowledge along the way that
allows them to represent their
00:11:16
ideas differently. So one thing
we're doing right now actually
00:11:19
is looking at online forums
where people write multiple
00:11:22
reviews over time. So someone,
for example, may write wine
00:11:26
reviews, hundreds of wine
reviews over the years from the
00:11:29
beginning, when they started
tasting wine, as they've learned
00:11:31
more about wine years later. And
so we're looking at, well, how do they
00:11:34
represent ideas differently as
they gain knowledge? And that
00:11:38
made -- that helped us understand why
people who represent ideas certain
00:11:41
ways end up doing better.
00:11:43
- I would think that as a college
professor, you have a little bit
00:11:45
of insight into that, because
you've been able to watch the
00:11:47
evolution of the writing of many
of your students, I would hope,
00:11:51
over time. But listen --
00:11:53
- I wish I got this data over
a bigger window, but yes.
00:11:55
- So there's there's a theory
question here. Because ChatGPT,
00:11:59
artificial intelligence, has
entered the conversation, right?
00:12:02
So now people have access to
these free tools that can
00:12:05
help them perfect their cover
letters and their writing
00:12:08
assignments and their written
business communication. How do
00:12:11
we discern, as managers, as
editors? How does this change
00:12:16
things? We can't really tell how
good someone's topography of
00:12:19
thought is if they have an AI
assisted piece of writing. What
00:12:23
do we do? - Yeah, so
00:12:25
I'd say a couple things. So I
agree with much of what you
00:12:27
said except one word, which is I'm not
sure they allow you to perfect
00:12:31
your writing, right? What they do is
they allow you to write
00:12:34
something. They -- I should be
careful here. Because maybe you
00:12:37
are not writing. Something is --
they allow you to produce
00:12:39
content that's pretty good,
right? You didn't have to write
00:12:43
it. You give it a prompt and it
produces content that's pretty
00:12:46
interesting, right? It does a
pretty good job of doing
00:12:49
something that might have been
difficult for you to use. And to
00:12:52
the degree it is -- uses your own
content somehow, right? Like you
00:12:55
could say, take my CV and use it
to put together a cover letter
00:13:00
based on my past experiences. So
that is based on you, and
00:13:03
someone else's might look
different to the degree that their CV
00:13:06
is different. But I wouldn't
say it necessarily perfects
00:13:09
anything. I was recently
reviewing a paper of a colleague
00:13:13
where they look at AI generated
ads. So they take ads that are
00:13:16
generated by AI as well as
not. And the AI ads are
00:13:20
pretty good, right? If you look
at them on the surface, and you
00:13:23
say, hey, you know, is this a
pretty good ad, you look at one,
00:13:26
it makes sense, it's cogent.
It's not like the version two or
00:13:28
three years ago where it didn't
make any sense, right? Now, it
00:13:31
looks amazing, right? It looks
quite good. The copy can be
00:13:33
good, the images can be good,
and so on. But what they found
00:13:36
is, first of all, in general
it didn't seem like AI ads did
00:13:39
better. In fact, it looked like
they did worse. And part of the
00:13:43
reason why is they use language
differently than humans use
00:13:46
language. So they may use less
concrete language, for example,
00:13:49
or they may use other types of
language that humans tend not to
00:13:52
use. And so it's not that it's
not pretty good. It's -- it is
00:13:55
pretty good. But is it as good
as the best humans? I don't
00:13:58
think we're -- I don't think we're
there yet. And for these systems
00:14:02
to get as good as the best
humans, it depends on what
00:14:04
they're trained on, right? So if
they're trained on, hey, what do
00:14:06
cover letters look like in
general, then they're going to
00:14:08
do a good job of representing
what a cover letter looks like.
00:14:12
If you had a dataset of really
effective cover letters, and you
00:14:14
trained it only on the really
effective cover letters, well,
00:14:16
then maybe it would produce
really effective cover letters.
00:14:19
But I don't think we're there
yet. And so certainly, tools
00:14:22
like ChatGPT and others have
made the production of content
00:14:25
much easier. And I can certainly
imagine a time down the road
00:14:29
where we do use them for many,
many tasks rather than writing
00:14:32
ourselves. But I still think
there's a lot to be understood
00:14:35
about how language reflects the
people that produce it, and how
00:14:38
to write more effective content
based on understanding what
00:14:42
makes language impactful.
00:14:42
- Jonah, thanks for being here.
00:14:45
- Thanks so much for having me.
00:14:46
- The paper is called "Topography
of Thought." You can find it
00:14:48
online. I also encourage you to
check out Jonah's latest book on
00:14:51
persuasive communication. It's
called<i> Magic Words: What to Say</i>
00:14:55
<i>to Get Your Way.</i> Love that title.
And if you'd like to check out
00:14:58
more of our content, we invite
you to go to Knowledge@Wharton.
00:15:01
Just type ua into your search
bar. I'm Angie Basiouny. Thanks
00:15:04
for listening.
00:15:06
- For more insight from
Knowledge@Wharton, please visit
00:15:09
Knowledge.Wharton.UPenn.edu.
