00:00:00
Dan Loney: Well, of course,
artificial
00:00:01
intelligence has been one of the
00:00:02
big stories in 2024. Companies
are investing millions of
00:00:06
dollars right now to figure out
how it can help their operations
00:00:10
improve. And then there is the
public at large. They are seeing
00:00:14
how it can change and improve
their day to day lives. Pleasure
00:00:18
to talk AI in 2024, about what we
can expect next year and beyond,
00:00:22
with Lynn Wu, who's an Associate
Professor of Operations,
00:00:25
Information and Decisions here
at the Wharton School. Hi Lynn.
00:00:28
Great to see you again.
- Same here.
00:00:30
Great to have you with us.
00:00:32
Look, I think my first line in
there kind of sets the stage for
00:00:35
this. We've had artificial
intelligence around, but it
00:00:39
feels like this year— maybe to a
degree a little bit in 2023, but
00:00:43
this year, specifically— it has
just been this explosion. How
00:00:46
have you viewed what we've seen
play out around AI this year?
00:00:50
Well,
00:00:50
I think this year we have seen a
tremendous amount of progress on
00:00:55
both technology and also just
general adoption by wider
00:01:01
part of the society. Like, not
just the tech firms, but small,
00:01:05
medium sized businesses are
starting using AI. Big firms are
00:01:08
having— developing AI strategies
around the operations, marketing
00:01:12
and a variety of work functions.
So we see definitely a wider use
00:01:16
adoption. Also just general use.
00:01:18
That component of both large and
small businesses using AI
00:01:21
I think is one of the unique
aspects of it, because a lot of
00:01:25
people would probably say, okay,
larger business has the
00:01:28
resources to be able to make the
investment, but there are
00:01:31
avenues that small businesses
can really dig deep into and
00:01:35
really benefit from.
00:01:37
Oh yeah, absolutely. I like to
think of the example just like, you
00:01:41
know, like search ads back in
like, 20 years ago. Remember, it
00:01:45
kind of democratized advertising to
some extent. - Sure, yeah.
00:01:47
- Because I think things like Gen AI
democratize some of the
00:01:50
knowledge, certain knowledge.
You know.
00:01:52
Or knowledge search. And also
00:01:54
general use of it for writing, for
coding, for development projects,
00:01:58
for players that wouldn't—
weren't able to do this in a
00:02:01
large scale before.
00:02:02
One of the areas that
00:02:04
was talked
about for a while, and I think
00:02:05
we're seeing it, you know, play
out maybe in a different path,
00:02:08
was how it was going to impact
labor. I mean, the discussion
00:02:11
point was, "Well, it's going to
take jobs away. It's, you know,
00:02:14
we're going to see new dynamics
at play."
00:02:15
It feels like that
00:02:18
it's a little bit different.
That there is a complementary
00:02:21
aspect to AI that's maybe taking
00:02:24
a bigger role, at least
right now.
00:02:26
Well, I think the complementarity is
going to be key in the future
00:02:29
and also the present. There are
lots of things Gen AI can do,
00:02:33
and there are lots of things it
does terribly and does it wrong.
00:02:35
And you've seen the
hallucination examples, right,
00:02:39
and the things they're not very
good at. And I think finding
00:02:41
that complementarity, figuring
out what humans are really good at
00:02:45
and what machines are really
good at, and how the machines
00:02:48
and the humans can work together
on a variety of different tasks in a
00:02:52
variety of different industries,
that's going to be key. And that
00:02:55
is not just about figuring out
complementarities. Once you
00:02:58
figured out that working
relationship, you also need to
00:03:01
redesign the workforce, the work
processes around it, to support
00:03:05
that complementarity.
00:03:06
I think the belief by a lot of
people is, or at least the
00:03:09
expectation, is that AI will be
perfect. We know that's not the
00:03:14
case, because it's still humans
that are doing a lot of the
00:03:17
behind the scenes work. So what
should our expectations be
00:03:21
around AI and its usage and its
capabilities,
00:03:24
maybe going forward?
00:03:25
I think one good rule of thumb
is that if you already good at
00:03:29
this, AI is going to help even
more. But if you're not good at
00:03:33
this already, or have a very,
very rudimentary knowledge about
00:03:36
something, I would use AI
cautiously. Because it does
00:03:39
hallucinate. It will tell you
things. Sounds very true, but in
00:03:43
fact, it's not true. In fact, they
confidently say the wrong thing.
00:03:47
And that is where it's difficult
for somebody who do not have in-
00:03:50
depth knowledge about the field
to disambiguate.
00:03:54
Is this right
or is it wrong?
00:03:55
So that becomes an important
dynamic for businesses to have
00:03:58
that understanding and to kind
of factor that in to the
00:04:01
operational piece of
understanding that AI is going
00:04:04
to be great, but it won't be
100% you know, proof, in terms of
00:04:08
what it can bring to the
table.
00:04:10
And it'll never be, with the
state of art today, right? It
00:04:13
will never be for the next couple
years. So I think it's really
00:04:16
important for all business to
mitigate the risks of these type
00:04:20
of using these type of tools, in
a variety of settings.
00:04:23
I saw a paper that you had
talked about, about how AI was
00:04:28
going to impact innovation. Take
us into your thoughts about
00:04:33
where AI will play a role in
that area specifically.
00:04:37
Yeah. I think AI is going to really
change how products are being—
00:04:41
being developed. I think one key
thing we've discovered in our
00:04:45
recent research is that AI
really turbo charges prototyping.
00:04:50
Right? Before, you can— you have to
make an ad yourself. Right?
00:04:54
Im— figure out image,
00:04:56
which actor to use. Now these
are all just generated
00:04:59
automatically using
generative AI, so you can just
00:05:02
decide to have 10 different
copies. - Right.
00:05:04
- Right? And with—
00:05:06
with which— whatever
actor you want,
00:05:09
and then play it out. So that
00:05:11
rapid prototyping is going to be
huge, right? And that's— that's—
00:05:15
that's not just for, you know,
digital products. - Right.
00:05:18
We also see in physical products too.
00:05:21
AI guides prototypes.
Like, AI gave
00:05:23
you ideas on where you actually
should build a prototype. - Right.
00:05:27
And how you should release in the
wild, and collect data, creating
00:05:30
this positive feedback loop
where AI guides you to develop a
00:05:33
product, and a product— when the
consumer using the product,
00:05:38
those data are being fed back to
the machine and continuously
00:05:41
improve your product. So that
really expedited prototyping,
00:05:44
and that can dramatically change
the way products even develop in
00:05:48
both small firms and big firms.
00:05:50
What do you say to the people that—
obviously there are some
00:05:53
concerns out there about bad
actors and what they can bring
00:05:58
forward by using AI. And I think
the question of how we
00:06:05
mitigate some of the negatives
out there becomes an important
00:06:08
component to this as well.
00:06:10
I think this is— this is huge,
and for a variety of reasons.
00:06:14
You mentioned bad
actors. I think that's where
00:06:16
regulation could potentially be
helpful. You know, just like, you
00:06:20
know, we— you can have a —
00:06:23
you can have a saw, right?
And a saw can
00:06:26
be used to kill somebody. A saw
can also be used to—
00:06:29
an amazing tool to help you,
00:06:31
you know, cut down a
tree, build furniture, right?
00:06:34
AI is just a tool. But you need to
have regular— you need to have
00:06:38
uses on where it should be used
and where it shouldn't be used.
00:06:43
And also, I want to also caution
that as firms are widely using
00:06:49
these generative AI tools, these
tools are actually very, very
00:06:53
expensive. - Right. - Right?
00:06:55
And it won't become
cheaper anytime soon.
00:06:58
You would generally
consider the exponential growth
00:07:01
like, you know, Moore's law, we'll
continuously be making technology
00:07:05
cheaper and cheaper. - Right.
- But we also see generative AI is
00:07:09
a highly concentrated industry,
with a few players each level of
00:07:14
the AI stack. So it's never going to
be that cheap because of the,
00:07:18
you know, because of
concentration industry.
00:07:20
What about the— what about the
component of AI with the general
00:07:23
public and how it impacts their
day to day operations of doing
00:07:28
projects? And I know part of the
conversation for me started here
00:07:33
at Wharton, with professors
talking about how they were
00:07:35
going to deal with generative AI
use in their classes and with
00:07:39
testing and such. How do you see
that playing out?
00:07:42
Oh, you mean AI
00:07:43
in education, or AI—
- Well, AI
00:07:45
in just use with the public, I
think in general.
00:07:48
In general. Well, I think AI is a general
purpose technology.
00:07:53
Just think of the
internet, right?
00:07:54
Internet can
be used for a variety of things,
00:07:57
variety of reasons, right? And I
think for a general public, it's
00:08:00
also same thing, right? You
should, as individuals, should
00:08:04
try to do everything you can—
you're doing,
00:08:07
and see if AI can do it.
00:08:09
Right? And then see where
things that AI is really good
00:08:12
at, like, that can help you
expedite the work you do. And
00:08:16
also see where is it— it can
go astronomically wrong, right?
00:08:20
Then, that's the area you should
focus on putting more time
00:08:23
and effort into. I think this is
very individual for— for people,
00:08:27
and also individual for firms to
figure out what the best use
00:08:30
cases for you, specifically, it's
going to be. And that is going to
00:08:34
be key to unlock the benefit of
AI for you.
00:08:38
Well, part of that conversation is
almost generational, like
00:08:41
younger generations versus older
generations, and how they accept
00:08:45
and use AI. It feels like, right
now, and you can shed light on this
00:08:50
that a lot of generations are
trying to, you know, get their
00:08:53
feet wet and understand the
impacts of of generative AI.
00:08:57
Younger generations may be able
to just run with this as we go
00:09:01
in future years and be able to
use it in even, you know, even
00:09:05
greater uses in the years ahead.
00:09:08
I actually think older people
can use generative AI pretty
00:09:12
well. Maybe my sample is small.
I see my parents using it all
00:09:15
the time to say, "Hey, this is
the medical report. What does it
00:09:18
really say?" - Yeah.
- Right? I think it can
00:09:21
definitely, it's— because the usage
00:09:22
itself is pretty—
relative easy as a
00:09:24
conversational bot, in many—
that's where— how most people are
00:09:28
using it. So I actually think it
lowers the— the typical, you
00:09:32
know, technology requirement for
older people to use. - Right.
00:09:34
Than— than, you know,
00:09:36
some— some
other technologies out there.
00:09:37
Which will be great, then, for
all generations.
00:09:40
Exactly. I actually worry about
young people
00:09:42
using these tools too
much. - Why so?
00:09:44
Well, because you don't
know how to write anymore.
00:09:46
You don't know how to—
- Well, that's true too. Yeah.
00:09:48
You don't know
how to do the basic things, and
00:09:51
you use this tool to substitute it.
And I think that could be a problem.
00:09:54
What do you think, then, going
back to the component of
00:09:56
innovation, what are the
biggest challenges of
00:10:00
incorporating AI to enhance
innovation moving forward?
00:10:05
Oh, there's lots of challenges
out there.
00:10:07
So number one, one thing
00:10:09
you really have to be
aware is that all these AI,
00:10:13
guided prototyping, all these
AI general innovations, tend to
00:10:17
be recombinational and
incremental, right? These are
00:10:22
great innovations, right? Big—
77% economic growth come from
00:10:25
combining things in a new way,
improving existing things
00:10:29
incrementally, right? But the
13% or 19% is radically new
00:10:34
innovation, right? And those are
actually becoming more
00:10:38
important, because they
immediately get fed back, fed
00:10:42
into AI as they further enable AI
to recombine that into existing
00:10:49
technologies and improve it in some
new way. So I think for firms
00:10:53
that are doing recombination
incremental innovation, this is—
00:10:57
the— this is amazing, right? You
can really turbocharge it.
00:11:01
Well, and if you, if you have that
recombination, then from a
00:11:05
business aspect, you're— you're
kind of taking the next step,
00:11:08
but maybe making a different
level of investment than maybe
00:11:11
you would have if you're doing
just radical innovation.
00:11:13
Absolutely, absolutely.
00:11:15
But for radical innovation
people, right—so, GenAI
00:11:18
may or may not help you. It
may help you, but it's not
00:11:21
clear. It's not that clear how—
to the extent it can help you.
00:11:24
Then what do you do, right? So I've
produced this amazing
00:11:27
technology. If I release it to the
wild, I am a name, I create a
00:11:31
lot of positive spillover to
others who can use my technology
00:11:35
to combine something in a new
way or improve it
00:11:38
better than I could.
- Right.
00:11:39
But then what do I
get out of it? Right? So
00:11:41
there's— there's a bit of
incentive change, because one
00:11:44
part of innovation,
recombination and incremental,
00:11:47
got much, much cheaper. - Yeah.
- It elevates, actually, the value of
00:11:51
the radical innovation. And we
actually need to protect that radical
00:11:54
innovation more— more—
more than before.
00:11:55
What do you think, then, 2025
means for
00:11:59
the growth of AI?
00:12:00
Ooh, that's a really tough question.
00:12:03
I think it can go both ways, and
00:12:07
let me explain that to you. So
00:12:10
this pace of where AI technology
00:12:12
is improving is exponential.
And it's still riding an
00:12:16
exponential curve, okay? But
there's also limitation that can
00:12:21
constrain that exponential
growth.
00:12:25
Number one is that we
already ran out of
00:12:26
high fidelity data. Which is,
you know, the
00:12:30
human-written, you know, articles
and writings. Because remember,
00:12:34
OpenAI had made— and others— had
made exclusive access to <i>Wall</i>
00:12:39
<i>Street Journal</i> and others,
right? Because we need human to
00:12:43
write these these texts. - Sure.
- Right? And these data has mostly
00:12:48
trained on internet data
produced by humans. - Right.
00:12:51
But now this data is actually being
produced by machines now.
00:12:54
So it gradually degrades the
technology output, because
00:12:59
machine-generated data, when
they fed into AI algorithm
00:13:03
produce less good stuff than
human data. So there's a
00:13:07
limitation on data. - Right.
And also, I mentioned before,
00:13:11
these things are still going to be
very, very expensive. - Sure.
00:13:13
Right? So when—so we're going
to see exponential growth
00:13:17
and concentration, industry
concentration makes it much more
00:13:23
expensive. Then adoption will be
diminished. - Right.
00:13:27
So you're going to see the two forces
going on.
00:13:29
Like, which one, which
one's gonna win out?
00:13:31
It's going to be a big one.
So it's unclear. Yeah.
00:13:33
Lynn, great to see you again.
Thanks very much for your time.
00:13:37
Great to have you here. Lynn Wu,
Associate Professor
00:13:39
of Operations, Information and
Decisions
00:13:41
here at the Wharton School.
