00:00:00
Dan Loney: We obviously have
seen a rise in the use of things
00:00:03
like ChatGPT in the last couple
of years, but one of the
00:00:08
questions now being asked is,
can you use AI to help with the
00:00:12
brainstorming of ideas? Gideon
Nave was part of a research
00:00:17
project done on the topic. He is
an Associate Professor of
00:00:21
Marketing here at the Wharton
School. Gideon, great to talk to
00:00:24
you again. How are you, sir?
00:00:25
I'm well, Dan. It's really hot
outside, but we are in a nice
00:00:29
fresh air-con here at Wharton. So.
00:00:31
Absolutely, absolutely air
conditioning is nice at this
00:00:35
time of the year. Hey, let's
start, I guess, with the genesis
00:00:39
of the idea. And is part of
this kind of the transformation
00:00:44
we're all going through in
understanding just how much we
00:00:48
can use things like ChatGPT,
and in what areas, and maybe
00:00:52
what areas we can't use them?
00:00:54
Well, you know, the work we've
been doing recently was about
00:01:00
creative tasks. Our creative
task, I would say the goal of
00:01:04
them is to come up with
something that is novel, unlike
00:01:07
more routine tasks of, you
know, just maybe translation,
00:01:12
right? Translation, you have a
sentence, you will translate. It
00:01:15
doesn't have to be surprising in
any way. It doesn't have to be novel.
00:01:20
You know, finding
new ways to do translation is a
00:01:24
creative task. And when we talk
about creativity, we talk about
00:01:27
things that are novel and
useful. That's— you know, people
00:01:32
don't seem to agree on the
definition, but when people see
00:01:34
something that is creative, they
do seem to agree that it is
00:01:37
creative. I've been teaching
creativity at Wharton for a
00:01:41
while. Even during the time when
ChatGPT came out, I was in the
00:01:45
middle of class, and I even got
to do like a live demo of it at
00:01:49
the time. And actually what I
remember in this class, I
00:01:52
actually demonstrated to people,
to my students, how you can use
00:01:57
GPT to just write an email to
your professor. And I wrote an
00:02:01
email where you write an excuse
to the professor, like, why did
00:02:04
you— why did you not submit your
homework in time. And one of
00:02:08
the students said, "Hey, ask it
to write a creative excuse." And
00:02:13
GPT did it well. It came out
with the idea that, you know, my
00:02:17
project turned out into
something that seemed like a
00:02:19
product, and I was working
very hard to do it, to get it
00:02:25
into market, and I missed the
deadline, right? That's very,
00:02:28
very creative. And that's the
old GPT, the first one that
00:02:30
came out. So— and we know that
these models have a tendency to
00:02:35
hallucinate. Meaning, to come up
with new stuff when— and pretend
00:02:38
that they know it. So for this,
for creative tasks, it's
00:02:41
actually pretty useful and cool.
And these models are trained on
00:02:45
a lot of human thoughts,
expressions, emotions and
00:02:50
whatnot. So we definitely see a
great potential here. And some
00:02:55
companies have been doing it
too. Coca Cola, for example, used
00:02:59
the AI to come up with, you
know, a new flavor that is like
00:03:03
the year 3000 and so on. The
thing is that as scientists, we—
00:03:06
what we want to do is go beyond
these anecdotes and try to
00:03:09
systematically understand what's
going to happen if all of us are
00:03:12
going to start using AI for our
creative processes.
00:03:15
And realistically, when you're
talking about the link between
00:03:17
creativity and what companies
look for, in many cases, that's
00:03:22
a core element of their
operation. They have to be
00:03:25
creative in order to be able to
take that next step in their
00:03:28
success level.
- When
00:03:30
you're in a competitive market,
obviously, like, creativity
00:03:33
is key. You want to do
something that your opponent
00:03:36
cannot do, and you want to— you
want to beat your opponent,
00:03:40
whether it is as a company that
tries to compete with others,
00:03:43
whether it is an employer that
competes for promotion, and so
00:03:47
on. And in this sense,
creativity is an ever-moving
00:03:50
goalpost, right? If I tell you a
joke now that you've already
00:03:53
heard, that's not going to be
funny. If I tell you a joke that
00:03:56
has the same principle that you
may have already heard about,
00:03:59
maybe, maybe it has a chance.
And in this sense, you know,
00:04:03
maybe getting the principles of
a joke like, sometimes you kind
00:04:06
of build something, and then you
break a pattern, that will be—
00:04:08
maybe a joke like this
could work.
00:04:12
So you mentioned Coca Cola,
Gideon, about how they use it.
00:04:16
How much expectation should we
have that companies are already
00:04:19
using AI to enhance the
creativity that they have?
00:04:23
You know, many companies
are using it already. And what
00:04:26
we as scientists try to do is to
try to go beyond these anecdotes
00:04:30
and get into real science and
see, what is going to happen if
00:04:34
everybody is going to use the
same model, such as ChatGPT? It's
00:04:37
very possible that the new AI is
going to give you an edge. But a
00:04:41
non-specific AI is kind of the
very start. And that's where we
00:04:45
come to the study that we
recently published. So a study
00:04:49
published in <i>Nature Human
Behavior</i> about four or five
00:04:52
months ago did an experiment
where they gave people various
00:04:57
creative tasks, such as coming
up with a novel use for a
00:05:01
garden hose and a rocket, or
coming up with— with ideas to
00:05:07
buy for an 80-year-old birthday.
And they did— they randomized
00:05:12
people into groups that either
could use Google searches or
00:05:15
ChatGPT, and they just compare the
creativity of the people who used
00:05:20
the former versus the latter.
And they found that the GPT
00:05:24
increased the creativity, overall,
of every person. What we did,
00:05:29
though, is, once we reanalyzed
their data, we found that GPT
00:05:33
indeed increased the creativity
of every individual, but it made
00:05:36
also everyone more similar to
one another. And if this is the
00:05:40
case, we actually end up with
less new ideas as a group. Okay?
00:05:45
So it's kind of like a
prisoner's dilemma, where it's
00:05:47
the benefit of everyone to use
the GPT. They, on average, become
00:05:51
more creative. But then, as a
global company, as a society,
00:05:57
we're going to become more
similar to one another, and then
00:05:59
we're going to actually smother
creativity.
00:06:02
And do you potentially end up
losing, you know, the individual
00:06:08
creativity? Because how one
person is creative compared to
00:06:11
another is most likely
different. And you're coming now
00:06:15
with a more kind of steady path,
straightforward path, of the
00:06:19
same types of creativity if
you're relying on ChatGPT.
00:06:22
Exactly. On the long run, I
think that once all of the ideas
00:06:25
are available to everyone, I'm
pretty sure that, like— we call
00:06:30
it the sprinkler effect. Like,
the one use that GPT came up as—
00:06:35
for garden hose and rocket ball,
is to make sprinklers from it.
00:06:39
It's an idea that is great. It's
useful, it's novel, and it's
00:06:42
rated highly. But if I'm a
manager, I'm going to get 50
00:06:46
different proposals of what to
do with that. And 20 of them are
00:06:50
sprinklers. I'm probably not
going to judge the people that
00:06:53
came up with the sprinkler as more
creative, even though, you know,
00:06:56
if I get one sprinkler it
seems like a great creative
00:07:00
idea. So I think we need to be
careful here.
00:07:02
So this is very much like a lot
of things around AI right now.
00:07:06
We're kind of testing where it's
taking us, but we still have to
00:07:09
be wary as to where it's going
to take us down the road.
00:07:11
I think that one—
00:07:13
you know, initial promise of
studying AI was to look at the
00:07:16
individual and how it makes the
individual more X, Y or Z.
00:07:20
When it comes to— when it comes to
productivity in maybe routine
00:07:24
tasks, it makes sense. When it
comes to creativity— but
00:07:27
creativity is always measured
against the prior art, against
00:07:30
what everyone else is doing.
Using GPT in a conventional way
00:07:35
is going to make you look
conventional, because everybody
00:07:37
is going to use it the same.
Even though, you know, if you're
00:07:41
the only one using it, it will
make you look like a superstar
00:07:44
creative person. But we're
talking here about technologies
00:07:47
that is available to everyone.
So there are risks here. And one
00:07:51
of the emphasis in my class— my
creativity class, I fully
00:07:54
endorse AI. So I wouldn't say
that, you know, don't use AI.
00:07:58
But there are ways to use it,
and there are smarter ways to
00:08:01
use it than just asking it, "Hey,
come up with a creative solution
00:08:05
to me, for my problem."
00:08:06
Gideon, great to talk to you
again. Thanks very much.
00:08:08
Always, Dan. Thank you.
00:08:10
Thank you.
Gideon Nave, Associate Professor
00:08:12
of Marketing here at the
Wharton School.
