00:00:00
Dan Loney: Well, as we head towards 2025,
the role that AI is playing in
00:00:03
our lives feels like it's
increasing significantly each
00:00:07
and every day. And to that end,
understanding how to lead in a
00:00:11
landscape like this is going to
be very important as well. A new
00:00:15
Wharton executive education
program will be focusing on that
00:00:18
element specifically. It is
titled, "Leading in an AI Powered
00:00:22
Future." It's going to be six
weekly sessions focusing on a
00:00:26
variety of topics, starting on
January 27. Ethan Mollick is co-
00:00:29
director of the Generative AI
Lab here at Wharton, as well as
00:00:33
an Associate Professor of
Management. Great to see you
00:00:36
again. How are you?
Ethan Mollick: I'm wonderful. Glad to be here.
00:00:38
- It is kind of interesting for me, and
I'm sure it is for you, to see
00:00:41
how everything in the last year
and a half, and AI has been
00:00:45
around longer than that, but how
it has become really the
00:00:49
main focus of so many people
in so many areas of business,
00:00:53
our culture, our lives, whatever.
- We
00:00:55
have a general purpose
technology that is, by
00:00:57
definition, a technology that
touches everything that we do in
00:01:01
different ways that aren't
predictable in advance. So it's
00:01:04
going to affect every field. I
mean, AI is a real thing. It's
00:01:07
really here to stay. We have
enough data coming in that it
00:01:10
actually has real impacts on
performance and jobs. I mean, I
00:01:13
think there's a reason why it's
the big topic. - So
00:01:15
as we turn the calendar for 2025,
what do you think are some of
00:01:19
the most significant changes
that AI might bring to a
00:01:22
business in the near
future? - So
00:01:25
there are -- there's sort of two
sets of things happening.
00:01:27
There's a set of technology
advancements that are going to
00:01:30
change things, and then there's
a set of organizational and
00:01:33
leadership changes happening. So on
the organizational leadership
00:01:36
side, I think that lags
technology. So the technology
00:01:39
has been amazing out of the
gate. What's been happening is
00:01:41
individuals have gotten access
to these tools, and they're
00:01:44
getting individual performance
improvements. Those aren't being
00:01:47
translated to the organizational
level. That's because
00:01:49
individuals are hiding their
use. They're hiding it because
00:01:52
they're either afraid that
they've been told they'll be
00:01:53
fired if they use AI the wrong
way, so they're not showing
00:01:55
people. They're viewed as
geniuses right now. So they
00:01:58
don't want to show people that
they're not geniuses and the AIS
00:02:00
being the genius. They're
worried that efficiency gains
00:02:02
translate to being fired or
being assigned more work or they
00:02:05
won't get credit for it. So one
of the big organizational
00:02:08
challenges, how do we move this from
the individual level to think
00:02:10
about how organizations can
benefit, and that's the
00:02:13
big challenge for the next year,
organizationally. - So
00:02:15
how do you think then leaders
of companies, what areas do they
00:02:19
focus on in terms of
incorporating AI into kind of
00:02:23
the organizational structure?
00:02:25
- So we don't have easy answers. I
mean, one of my biggest fears
00:02:28
about AI is that organizations
have, by and large, given up
00:02:32
what they used to spend a lot of
effort doing, which was thinking
00:02:35
about organizational redesign,
right? You look at how Ford was
00:02:38
successful, right? Or even how
Salesforce was successful. These
00:02:41
companies were successful
because they thought about how
00:02:43
to redesign organizations, like
organizational structure. Now,
00:02:46
most organizations don't do a
lot of R&D in their own
00:02:48
organizational structure. They
count on outside consultants
00:02:51
like McKinsey coming and helping
them, or they count on, you
00:02:53
know, building, buying software
products that kind of impose
00:02:55
their organizational structure
on them. You learn from, you
00:02:58
know, Workday teaches you how to
use HR in the Workday kind of
00:03:01
way. And as a result, they're
not innovating. Because nobody
00:03:03
has answers right now. We have
to figure this out. So I think
00:03:06
it has to start with
leaders embracing using AI
00:03:09
personally, getting their 10
hours of AI use in, to find out
00:03:12
what it's good or bad for
them. And then really starting
00:03:14
to think at a fundamental level,
how are we doing R&D in our
00:03:17
organization around this?
00:03:18
- And so then that has the
opportunity to impact the, quote
00:03:21
unquote, culture in and around
the organization, in terms of
00:03:26
people really buying in and
understanding how valuable AI
00:03:30
can be in various aspects of
their work.
00:03:32
- So I think people are already
learning that. I think the
00:03:34
question is, how do we capture
that, right? What does the
00:03:37
organization that has AI in it
look like? It can't look the
00:03:40
same, right? With the
organizational structure we have
00:03:43
today, it is the exact same one we
had in 1855 when the New York
00:03:46
Railroad invented the org chart, right?
00:03:49
And it was designed to solve a
bunch of problems that the New
00:03:51
York Railroad had, which was, how
do we coordinate work across a
00:03:54
vast distance in real time? We
invented a whole bunch of new
00:03:57
techniques in the 1910s around
these, around assembly
00:04:01
lines, in the the early 2000s, run
agile development, but all of
00:04:03
them were limited by people. It
all depended on people being the
00:04:06
unit of intelligence and the
unit of action that things
00:04:09
happened in. That's not
what's going to happen in the
00:04:12
future.
00:04:12
- So then, how do you see AI
impacting different business
00:04:15
models as we move forward?
00:04:17
- So I think that's where the
technology starts to come in,
00:04:19
right? So one of the big things
that we're going to look for in
00:04:22
2025 is the rise of agents,
which are already a thing. So
00:04:25
agents are AI systems that are
goal directed and have agency to
00:04:29
pursue those goals on their own.
And that means they can use a
00:04:32
computer mouse, they can do work
for you. That starts to look a
00:04:35
lot more like a human doing work
than it does a chat bot that you
00:04:38
ask a question to.
- Yeah.
00:04:40
It has to be a challenge,
though, for a lot of companies,
00:04:42
when you realize that your
competition is probably, you're
00:04:45
going to be using these types of
technology as well, and you have
00:04:50
to implement them and get them
working up to optimum so that
00:04:54
you can remain competitive in
whatever field it is.
00:04:56
- It's actually much larger than
that, right? Because, first of
00:04:58
all, there's no company out
there that is listening to this
00:05:01
podcast that's going to build
their own LLM. There's really --
00:05:04
it's too expensive to build
large LLMs, beat small LLMs, and
00:05:07
there's six or seven companies
in the US that could do it, few
00:05:09
more around the world. That's
it. So you're not going to build
00:05:11
your own. So that means you're
going to be using a -- you know,
00:05:14
whether you're using a closed
source solution, like an Open
00:05:16
AI, or whether you're using
something with Llama, you're
00:05:18
still going to be using someone
else's large language model.
00:05:20
And that large language
model is ubiquitously available
00:05:24
around the world. Every kid in
Mozambique has access to the
00:05:26
exact same tool that your top
analysts do. So it's about
00:05:28
thinking about structure and
approach and how to leverage
00:05:31
what you already do to be more
successful. You're not going to
00:05:34
win by technology play alone.
00:05:36
- How then do you look at not even
just next year, but the next
00:05:40
several years, in terms of how AI
could potentially have an impact
00:05:45
on business? - So
00:05:46
I think that we're going to see --
you know, there's this -- there's
00:05:49
Amara's Law, which says that
people overestimate change in
00:05:53
the short term and underestimate
change in the long term. I think
00:05:56
that, you know, you're not
going to see an instantaneous
00:05:58
50 percent improvement to ROI, but
you're going to start seeing
00:06:01
that in the next few years, not
the next year or two. So the next
00:06:04
year or two is still
experimentation, but that's
00:06:06
going to set the groundwork for
whether or not you succeed in
00:06:08
the long term. Every piece of
evidence we have is that this is
00:06:11
indeed the big one. Like, all the
control experiments, a lot
00:06:13
of my colleagues at Wharton and I
have been doing research on
00:06:16
this. The impacts are very, very
00:06:17
large. - You mentioned a
00:06:18
moment ago about the leaders,
the managers within the company
00:06:22
getting their 10 hours, getting
that experience. How invaluable,
00:06:26
though, will that component be
as we move forward and we try
00:06:29
and expand AI in throughout our
organizations, of the leaders
00:06:33
having the understanding and the
background around AI? - So
00:06:37
you don't need a lot of
understanding to make this work.
00:06:39
You just need to use it. I mean,
I talk to executive audiences all
00:06:42
the time. Maybe 10 to 20 percent of the
audience has spent 10 hours with
00:06:46
a Frontier model trying to get it
to do real work. If you don't do
00:06:49
that, you have no idea what
these systems can do. You cannot
00:06:50
have your direct reports tell
you what to do, because these
00:06:53
things are already operating at
a high enough level that it is
00:06:55
executive in a lot of ways,
right? We have some research.
00:06:57
There's some research showing
that it gives us good advice as
00:06:59
strategy professors at top
business schools, you know, when
00:07:02
you ask the AI multiple times.
We know that, you know, already
00:07:05
high performing entrepreneurs in
Kenya who get advice from the AI,
00:07:08
only advice, not extra help, but
advice, have an 18 percent improvement
00:07:11
in profitability. Like, these are
real things. And so if you're
00:07:14
not using it, you're not going to
know what it can do. And it's
00:07:16
not a technology that works like
tech. It's bad software. AI
00:07:19
doesn't work like software.
Software shouldn't argue
00:07:21
with you or refuse to do work. AI
does all those things. It works
00:07:25
like a person. If you're a good
manager of people, you'll be
00:07:27
good with AI, even though it
isn't a person. - So
00:07:29
how important then do you think
bringing this type of a course
00:07:31
forward now is in kind of the
landscape of where we are with
00:07:34
AI, but truly diving deeper into
the leadership side of being
00:07:39
able to implement and build it
00:07:41
out? - So I think that we don't
have a perfect pathway to
00:07:44
getting to AI implementation,
right? So I'm not going to claim
00:07:47
that this course is going to be
something where it's like, okay,
00:07:49
now you learn how to implement
AI. We're actually -- and it's also
00:07:52
multiple people at Wharton,
because we have different
00:07:54
viewpoints, and outside of
Wharton, all kind of discussing
00:07:56
this, because we don't have
answers yet. Really good
00:07:58
research going on, but no
answers yet. But I think the
00:08:00
most important thing is to
demystify. I mean, it is the
00:08:03
weird and strange and mystical,
but to demystify using it at
00:08:07
least. To get people on board
with understanding the
00:08:09
implications of this, to see
that it's a very big deal, and
00:08:12
to get agency over their own
use. And that's really what I
00:08:15
hope this course will accomplish.
00:08:16
- And I think from, you know,
doing the show that we do on
00:08:19
Sirius XM every day, you hear
the conversation about, you
00:08:22
know, who all will really
benefit from AI as we move
00:08:27
forward, and it seems like there
are elements of this that are
00:08:31
working for multiple
generations. It's not just the
00:08:33
younger generation that's going
to come in and have the greatest
00:08:36
advantage. Older generations are
being able to dip their toe in
00:08:39
this water and see the
benefits as well. - Actually,
00:08:41
it's worse than that. We
actually have a study that I did with
00:08:44
my colleagues at Harvard, MIT, and
the University of Warwick, where we
00:08:46
went to Boston Consulting Group,
and we found massive
00:08:48
improvements in productivity,
but we also found that junior
00:08:50
employees were actually much
worse at explaining how this
00:08:52
technology should be used than
senior people, because they
00:08:55
don't understand how work works.
They don't understand how
00:08:56
management works. So they -- it's
not a technology problem. Like,
00:08:59
that's the most important thing
to get across to people. There
00:09:01
is a tech element to this, but
it's mostly a management
00:09:04
teaching and -- you know, and
structural problem. It's an
00:09:08
executive problem, not a IT
problem. And if you assign this
00:09:10
to your IT department as your
first go, this is not the right
00:09:13
way to approach AI.
00:09:14
- But then it will have a
significant benefit moving
00:09:18
forward in terms of how
leadership kind of runs an
00:09:21
operational perspective of a
business, because you'll have
00:09:24
the component of AI in the mix.
- Right, I
00:09:26
mean, it's not just a component.
Like, I've already done -- you
00:09:29
know, you can have AI watch a
scene and give you feedback on
00:09:32
what's going on. You can have --
it's good at advice. It's not like --
00:09:35
I think, again, there's a lot of --
another problem executives have
00:09:38
is they're anchored on AI prior
to 2022 with non-generative AI,
00:09:42
which was all about making
predictions about the future.
00:09:44
These OMS are not good about
predictions. What they're really
00:09:47
good at is about acting like
people and doing people-like
00:09:50
jobs. - Great
00:09:51
to see you again, Ethan. Thanks
very much for your time. - Thanks
00:09:54
for having me. - Ethan Mollick from
here at the Wharton School, who
00:09:56
is co-director of the Generative
AI Lab and also Associate
00:10:00
Professor of Management.
