00:00:00
AI is going to be like electricity or
00:00:03
like the steam engine or like computers
00:00:06
meaning the kinds of technology that
00:00:09
changed the world forever that changed
00:00:11
Humanity forever
00:00:12
welcome to the ripple effect the podcast
00:00:14
that takes you on a journey through the
00:00:17
minds of work and faculty I'm your host
00:00:19
Dan Loney and in each episode we'll be
00:00:21
diving deep into the inspiration behind
00:00:24
the groundbreaking research that Wharton
00:00:26
professors have conducted and exploring
00:00:28
how their findings resonate with the
00:00:30
world today we'll be covering a diverse
00:00:32
range of topics bringing you the latest
00:00:34
insights and knowledge that you can
00:00:37
apply to your life into work so get
00:00:39
ready to dive into new ideas with the
00:00:42
ripple effect
00:00:43
well we know there's been a lot of talk
00:00:46
about artificial intelligence uh
00:00:48
especially in the immediacy kartik and
00:00:51
this is something that you have written
00:00:53
about and talked about a lot what was it
00:00:56
that kind of got your juices flowing and
00:00:58
interest you about artificial
00:01:00
intelligence in the first place
00:01:02
well you know my undergraduate degree
00:01:06
was in electronics and and there was a
00:01:09
masters in computer science so I'd
00:01:10
studied computer programming but this
00:01:13
was back in the 90s at a time when AI
00:01:17
wasn't what it is today and in fact he
00:01:19
had mostly failed to deliver on its
00:01:22
early Promise by then so the interest in
00:01:25
AI was diminishing we had very limited
00:01:27
coursework in AI
00:01:29
and so that's the context in which I was
00:01:32
first to introduce to AI but what really
00:01:34
piqued my interest were actually a
00:01:36
couple things when I was in grad school
00:01:38
doing my PhD at Carnegie Mellon I took a
00:01:41
course from
00:01:43
you know one of these Geniuses of modern
00:01:46
times his name is Herb Simon
00:01:48
he is I think probably the only person I
00:01:51
know who's won the highest award in
00:01:53
economics
00:01:55
the Nobel Prize and the highest award in
00:01:57
computer science the Turing award and
00:01:59
the highest award in psychology
00:02:01
so he won all of these three and he was
00:02:03
on campus and he was teaching a course
00:02:05
and you just try and register for the
00:02:07
course if you can
00:02:09
and I did without having any interest in
00:02:11
the subject
00:02:13
and I remember when I was in that class
00:02:17
and he would talk about these things
00:02:18
which are like a mix of
00:02:20
you know psychology how the human mind
00:02:22
works computer science how can we take
00:02:25
those ideas into uh the world of AI and
00:02:29
computers and then economics as well in
00:02:31
terms of what this means for for the
00:02:33
world you know that was the first time
00:02:35
my interest in this started to
00:02:38
um
00:02:38
get picked nonetheless my work still
00:02:42
wasn't yet AI for the next few years I
00:02:45
was working on e-commerce Internet
00:02:47
advertising and so on and my first
00:02:50
genuine interest in this topic came in
00:02:52
when you know you started to see like
00:02:54
personalized recommendations on Amazon
00:02:56
and Netflix
00:02:58
um in all these places and a student of
00:03:00
mine
00:03:01
uh brought up this question of you know
00:03:04
what is it doing to the kinds of
00:03:06
products we consume and kinds of media
00:03:08
we consume and how is it changing it and
00:03:11
then I got really interested in this
00:03:12
idea that algorithms are influencing
00:03:15
decisions we make
00:03:16
and that was my first entry into this
00:03:19
subject of algorithms broadly but then
00:03:21
within that AI as well so I find it
00:03:24
interesting because
00:03:26
um there's so much conversation going on
00:03:28
right now about Ai and how it's going to
00:03:31
impact business and realistically Ai and
00:03:34
business are are not new to each other
00:03:36
they've been connected for some time but
00:03:38
it feels like the conversation has taken
00:03:41
a different level how do you view that
00:03:43
combination of business and Ai and how
00:03:46
those two will work in the future
00:03:49
yeah so you know I think it's really
00:03:51
interesting if you look at Ai and
00:03:53
business of course AI is the big
00:03:55
buzzword in business and so I find it
00:03:59
often in the business world gets
00:04:00
divisive and a bit polarized in the
00:04:03
sense that there are the Believers
00:04:05
who talk about look AI is going to be a
00:04:08
game changer and then there are people
00:04:09
who feel like oh okay this is the next
00:04:11
nft or the next uh I don't know wearable
00:04:15
computers or Google Glasses or whatever
00:04:16
pick your example where there's a
00:04:19
technology with a lot of hype that goes
00:04:21
nowhere so I'm going to make a big board
00:04:23
plane here which is that I think AI is
00:04:26
going to be like electricity or like the
00:04:30
steam engine or like computers meaning
00:04:33
the kinds of technology that changed the
00:04:36
world forever that changed Humanity
00:04:37
forever there's the
00:04:39
you know human lives before electricity
00:04:41
and there's human lives after
00:04:42
electricity it's going to be like that
00:04:44
where they are
00:04:45
and this is not just a statement I'm
00:04:48
making based on my gut feel which by the
00:04:51
way there is gut feel in that statement
00:04:53
but it's based on real evidence so
00:04:56
people economists and other researchers
00:04:59
have studied these kinds of technologies
00:05:01
that we refer to as general purpose
00:05:03
Technologies these are Technologies like
00:05:05
electricity computers that
00:05:08
are different than other Technologies in
00:05:10
a few ways one is that
00:05:12
at a macro level they stimulate a lot of
00:05:16
innovation and a huge amount of economic
00:05:18
growth at a micro level meaning
00:05:21
individual firms they end up changing
00:05:24
winners and losers of individual markets
00:05:27
because of how companies adopt the
00:05:29
technology like take Internet for
00:05:30
example
00:05:31
well the largest retailer is in Walmart
00:05:34
it's Amazon
00:05:36
are Kmart one of the largest retailers
00:05:38
before the internet doesn't exist today
00:05:40
things like that right it changes
00:05:42
competitive Dynamics fundamentally and
00:05:45
researchers have looked at what are the
00:05:48
properties of technologies that go on to
00:05:50
become general purpose Technologies and
00:05:53
all the early data suggests that AI
00:05:56
looks like a general purpose
00:05:57
Technologies if you look at
00:05:59
hiring patterns related related to AI if
00:06:03
you look at patent filings related to AI
00:06:06
you look at a number of other things on
00:06:10
in fact there was a recent study by my
00:06:12
colleague Dan rock where he looked at
00:06:15
specifically large language models like
00:06:17
chat GPD and even his study finds even
00:06:19
those models have some of the properties
00:06:22
of general purpose technology so
00:06:25
if you you started by asking what is the
00:06:28
connection to business and I think my
00:06:30
answer is it is going to be
00:06:32
fundamentally transformative for
00:06:34
business
00:06:36
then you're talking about basically kind
00:06:38
of like a pivot moment uh you know we
00:06:41
use the term pivot a lot over the last
00:06:43
three or four years because of the
00:06:44
pandemic and how businesses had to make
00:06:47
pivots in order to be able to survive
00:06:49
this is a pivot but on a much larger
00:06:51
scale of where we are going uh as a
00:06:55
society
00:06:56
absolutely I mean you've just brought up
00:06:58
the pandemic imagine
00:07:00
with the pandemic
00:07:02
without the internet what that pandemic
00:07:05
would be like you know we were able to
00:07:07
navigate the pandemic because of the
00:07:09
internet we were able to continue to
00:07:10
work because of zoom and other things so
00:07:12
the internet was really a general
00:07:16
purpose technology that has changed our
00:07:18
lives and it had a huge impact over the
00:07:21
last 20 years and certainly the last two
00:07:22
three years AI will be similar as well I
00:07:25
mean we're just starting to see the
00:07:27
early
00:07:28
you know things like chat GPD but this
00:07:31
is just a start I mean it's going to
00:07:32
change everything and companies that
00:07:34
don't wake up to that reality that want
00:07:37
to follow rather than lead that want to
00:07:41
say look
00:07:43
you know this could be just a next
00:07:45
buzzword we will play it safe
00:07:49
or the companies that say
00:07:51
you know the moment they see an early
00:07:53
failure that backtrack and say there's
00:07:55
no Roi on this like the companies that
00:07:58
did do that when the.com bust happened
00:08:01
companies that play these kinds of moves
00:08:03
will pay a big price and I think it's
00:08:05
the companies that truly Embrace its
00:08:07
potential and play the long game they're
00:08:09
going to be the big winners from from
00:08:11
this trend so then what do you say about
00:08:13
some of the recent calls to maybe slow
00:08:16
down the development process and maybe
00:08:19
take a little bit more time and and
00:08:21
really think this out because it seems
00:08:24
like there are obviously with some of
00:08:26
the people that have talked about this
00:08:27
they have some concerns about how fast
00:08:30
things are moving
00:08:32
I think first of all the concerns are
00:08:34
legitimate
00:08:36
it is moving very fast this is a
00:08:39
technology that is unlike other
00:08:40
Technologies we've seen in terms of the
00:08:43
rate of change and the rate of progress
00:08:46
and especially given its implications
00:08:51
for
00:08:52
simple things like employment
00:08:55
employability all the way to
00:08:59
you know
00:09:01
things like use of AI in Warfare
00:09:05
or AI going out of control there's a
00:09:08
range of concerns here so I think the
00:09:10
concerns are real
00:09:11
now what is the right solution to those
00:09:14
I'm not yet sold on whether a six-month
00:09:17
pause in AI work is going to change
00:09:21
anything
00:09:24
if first of all I don't even think it's
00:09:26
feasible but let's say it's feasible and
00:09:28
you're able to stop all people working
00:09:30
on these kinds of AI models and say stop
00:09:33
for six months what's going to happen in
00:09:35
six months
00:09:36
nothing
00:09:39
um because it's not like you'll find the
00:09:41
magic
00:09:42
uh
00:09:43
solution in fact what needs to happen
00:09:47
is
00:09:49
you know
00:09:50
investments in education
00:09:53
at school levels where people are
00:09:56
trained to understand AI they're trained
00:09:58
to understand things like deep fakes
00:10:00
they're trained to understand
00:10:02
issues around ethics when Building
00:10:04
Technology this is not something you
00:10:06
solved in six months this is something
00:10:08
you solve over 10 years
00:10:10
and and change curriculum you need to
00:10:13
retrain Engineers you need to retrain
00:10:16
managers you need to also retrain
00:10:19
your
00:10:20
congressmen and senators and all of the
00:10:24
politicians and lawmakers
00:10:26
you can none of that so what what are
00:10:29
you going to change in six months
00:10:30
nothing and so I think what it requires
00:10:32
is like really a focused effort
00:10:36
where you're changing things over a
00:10:38
20-year period and you are fast to react
00:10:41
to problems that you've noticed with AI
00:10:46
I was gonna say because I I think
00:10:48
there's also another issue to bring up
00:10:50
here as well and and I'll use chat GPT
00:10:53
as the example because seemingly that is
00:10:55
the one that everybody is talking about
00:10:57
right now and everybody wants to
00:10:58
incorporate in their operations whether
00:11:01
it's Microsoft Google companies
00:11:03
Etc uh how are companies going to be
00:11:08
able to use this technology and say be
00:11:10
better than their competition if they're
00:11:13
all using the same type of product
00:11:16
yeah
00:11:17
great question by the way I actually
00:11:20
think that
00:11:22
a lot of the companies that will use
00:11:26
off-the-shelf tools like say chat GPD
00:11:29
and others
00:11:30
will create amazing efficiencies that
00:11:34
will be copied by a lot of their
00:11:35
competitors which will bring costs down
00:11:38
and all of the value I think will accrue
00:11:40
to the eventual customers and users
00:11:43
because it'll bring prices down the
00:11:46
second thing that's going to happen is
00:11:48
because they bring prices down it will
00:11:50
help expand markets because it'll bring
00:11:53
in new customers into various markets
00:11:56
and the expansion of markets will mean
00:11:59
there's value created for all of those
00:12:00
companies equally meaning all of them
00:12:02
gain some
00:12:04
the companies that actually will be able
00:12:07
to use things like this to get a real
00:12:09
advantage over their competitors
00:12:12
are going to be companies that are able
00:12:15
to pair
00:12:16
off-the-shelf AI tools and capabilities
00:12:21
with something proprietary
00:12:23
and what is that proprietary
00:12:26
complementary asset they can bring to
00:12:28
the table is going to be the name of the
00:12:31
game for companies that are aggressively
00:12:32
investing so I'll give you a couple
00:12:34
examples of what is a proprietary thing
00:12:36
they can bring in
00:12:37
you could use an off-the-shelf large
00:12:39
language model like
00:12:41
gpd4 which is basically the underlying
00:12:45
model for uh you know the chat GPD which
00:12:50
was like GPD 3.5 but you can use an
00:12:52
off-the-shelf model like that but if
00:12:54
you've got a large proprietary data set
00:12:56
of say Healthcare information Healthcare
00:12:58
data set you can train or retrain those
00:13:01
models on your massive Healthcare data
00:13:06
set and now you've created a new AI that
00:13:08
is the best in class at answering
00:13:11
Healthcare questions
00:13:12
and you were able to do that because you
00:13:14
had the largest
00:13:16
proprietary Healthcare data set you
00:13:18
could do the same thing in finance in
00:13:20
other areas so that's one you bring in
00:13:22
something proprietary usually a very
00:13:24
large proprietary data set or
00:13:28
you change something
00:13:30
in terms of user experience
00:13:33
a good example of charge GPT itself open
00:13:35
AI the company be behind chat GPD has
00:13:38
had GPD one gpd2 gpd3 for a while and
00:13:42
developers have been using it the
00:13:44
capability in chat GPT is not
00:13:46
fundamentally new it was already there
00:13:48
and we've seen it I've used gpd3 for
00:13:51
over a year now
00:13:53
the difference is charge GPD provided
00:13:55
that in a very seamless easy to use sort
00:13:59
of UI
00:14:00
and that shows you the value of user
00:14:03
experience so somebody can take off the
00:14:05
shelf AI but integrate it into a great
00:14:10
user experience
00:14:12
that creates a winning combination
00:14:14
or somebody combines like for example
00:14:17
there are companies that are trying to
00:14:18
build image editing AI where you're
00:14:21
taking an image and you want to edit it
00:14:23
you want some things to be changed you
00:14:25
don't want to go to photoshop and do it
00:14:27
and you just want to give an instruction
00:14:28
to Ai and it does it for you great
00:14:30
there's lots of startups doing that
00:14:33
many will use the same kinds of AI they
00:14:35
look very similar
00:14:37
now but if an apple does it and
00:14:40
integrates it into an iPhone they can
00:14:42
give a seamless experience to the user
00:14:44
because you don't have to download an
00:14:45
app you can take a photo right there you
00:14:47
can make edits if Google does it and
00:14:49
integrates it into Android that again
00:14:52
gives a seamless experience on the phone
00:14:54
that gives them a leg up over anyone
00:14:57
else that's using the same kind of AI as
00:14:59
them so that so it's all about pairing
00:15:01
it with something called proprietary
00:15:03
that is also complementary
00:15:07
let me have you discuss the uh the legal
00:15:10
side of the advancements that we're
00:15:12
seeing around Ai and obviously there's
00:15:14
lots of discussion right now uh around
00:15:17
big Tech uh on the regulatory side at
00:15:19
the moment how then does AI
00:15:23
factor into the discussions on on legal
00:15:26
and IP issues
00:15:28
yeah I mean there's there's tons of
00:15:31
legal issues around AI I think I'll
00:15:34
mention a couple one
00:15:35
couple ones one is
00:15:38
is what happens when you use AI to make
00:15:42
decisions in socially consequential
00:15:45
settings and you do it at Large Scale so
00:15:48
for example there have been concerns
00:15:50
about using AI
00:15:52
in courtrooms for example
00:15:54
to predict the likelihood that a defense
00:15:57
a defendant will reopen
00:15:59
or what happens if you use AI to do
00:16:01
resume screening
00:16:04
or you use AI to do loan approvals and
00:16:07
it turns out these AI have biases then a
00:16:10
company that uses them in these very
00:16:13
important settings exposes themselves to
00:16:17
uh you know litigation and those kinds
00:16:20
of issues
00:16:21
and so that's one type of issue and I
00:16:23
think at the end of the day my view on
00:16:25
this is look yes you can complain all
00:16:28
day you want about potential biases in
00:16:31
AI but before we do that let's talk
00:16:33
about what's the alternative the
00:16:35
alternative is humans fundamentally
00:16:37
flawed
00:16:39
human decision makers who have their own
00:16:42
biases so it's not like decisions in
00:16:45
courtroom today or in hiring today are
00:16:47
unbiased and you're switching to AI
00:16:49
That's more biased in fact the reality
00:16:51
is AI biases are probably easier to
00:16:55
detect than human biases and probably
00:16:57
easier to correct than AI biases and so
00:17:00
companies will have to make sure they
00:17:02
are taking sufficient safeguards
00:17:04
auditing their AI sufficiently
00:17:07
uh before they release Ai and these
00:17:09
socially consequential settings so
00:17:11
that's one set of concerns the other
00:17:13
related to generative AI
00:17:15
and we already saw this play out last
00:17:17
weekend when this track was released
00:17:20
That was supposed to be by Drake
00:17:22
uh it did really well took off and then
00:17:25
it turns out somebody created it with AI
00:17:27
and so I think the legal issues there
00:17:31
are going to be both on the input side
00:17:33
of generative Ai and the output side and
00:17:35
by input side I mean what kind of data
00:17:38
are used to train the AI so if you're
00:17:42
using data to train AI to create music
00:17:45
the question is where did the training
00:17:47
data set come from do you have the
00:17:49
consent of the musicians who's created
00:17:53
the music before you trained your system
00:17:55
on that are you giving them suitable
00:17:58
compensation
00:17:59
if money is made out of the resulting
00:18:02
product how are you tracking what is
00:18:04
each musician's contribution you create
00:18:06
a new brand new song how do you say two
00:18:09
percent of this song is inspired from
00:18:11
Jay-Z and three percent from Drake and
00:18:14
four percent from somebody else how do
00:18:16
you even determine that
00:18:18
that's on the input side and on the
00:18:19
output side if you create a new track
00:18:22
and in Drake's voice
00:18:25
uh or an Elvis's voice is that allowed
00:18:28
do you need permissions do you you know
00:18:30
so there's all these kinds of things and
00:18:32
by the way U.S copyright law doesn't
00:18:34
even cover
00:18:35
synthetic media so what is the copyright
00:18:39
law around creation of content by AI so
00:18:43
lots of issues that have to be tackled
00:18:45
in in the coming years and those are the
00:18:47
kinds of things where I think lawmakers
00:18:51
and lawyers in general will be
00:18:55
slow and not Progressive
00:18:57
and they will typically just resort to
00:19:00
lawsuits and we I think we'll see a lot
00:19:02
of lawsuits in the next two three years
00:19:04
can I have you finish up our
00:19:07
conversation around Ai and the labor
00:19:09
market because that I think is also just
00:19:12
the potential fascinating I mean
00:19:14
obviously there have been stories and
00:19:16
themes thrown out there for quite some
00:19:18
time about the potential impact but we
00:19:21
haven't gotten to the point I think yet
00:19:23
where we've really had the rubber meet
00:19:26
the road to a degree
00:19:28
yeah yeah true
00:19:30
uh look I think
00:19:32
as far as ai's impact on labor market
00:19:35
anyone who's
00:19:37
concerned about its potential impact on
00:19:39
the labor market is I think in the you
00:19:42
know is asking the right set of
00:19:43
questions
00:19:45
in the sense that
00:19:47
even though technology is in the past
00:19:51
in human history
00:19:53
have often had labor concerns associated
00:19:56
with those Technologies labor is always
00:19:58
for Technologies and in the past it's
00:20:01
almost always been the case that new
00:20:03
technologies have created more jobs than
00:20:06
they have destroyed
00:20:08
and so all of those concerns were
00:20:10
misplaced in the past
00:20:12
the real question is is AI like every
00:20:15
other technology in the past where it
00:20:16
will eventually create more jobs than it
00:20:18
destroys or is it going to ultimately
00:20:22
uh be a net
00:20:25
job
00:20:27
you know cannibalizer as opposed to
00:20:29
Creator and I think that's the big
00:20:31
question we don't know the answer if you
00:20:34
were to put a gun to my head and say
00:20:36
give me an answer right now Karthik I
00:20:38
would say well I think it's probably
00:20:40
going to have a net job loss rather than
00:20:43
that job creation however that is not
00:20:46
the full answer though
00:20:48
because AI will also augment jobs
00:20:53
and not just merely replace jobs so
00:20:55
there will be a lot of jobs where
00:20:57
there's a lot of routine things that we
00:20:59
do that we don't enjoy
00:21:01
that we do repetitively that are
00:21:05
you know soul-sucking in some ways we
00:21:08
will be able to Outsource that to Ai and
00:21:10
we'll free up time to do the more
00:21:12
interesting more creative pieces of the
00:21:14
job which I think will be great for all
00:21:16
of us so I think you know there's going
00:21:19
to be that as well and I also want to
00:21:21
distinguish between High skill and low
00:21:24
skill jobs so
00:21:26
the kind of AI that's been around for
00:21:29
the last 10 years we'll call it
00:21:30
predictive AI these are this is AI that
00:21:33
makes predictions and you plug it into
00:21:35
different tasks like predictive credit
00:21:37
card transaction is fraudulent or not
00:21:39
predictive an email is Spam or not or
00:21:41
things like that this kind of AI
00:21:45
you know
00:21:47
is one type and then there's generative
00:21:50
AI which is like chat GPD or uh you know
00:21:53
stable diffusion that's creating text
00:21:55
that's creating images and so on
00:21:57
and I want to talk about how these
00:21:59
impact jobs at different skill levels
00:22:02
but first thing I'm going to just say is
00:22:05
historically automation has affected
00:22:08
blue-collar jobs the most low skill jobs
00:22:10
that's what it's automated and affected
00:22:12
the most
00:22:14
question is is this true for the new
00:22:15
kinds of AI like chat GPD or image
00:22:18
creation and early research actually
00:22:20
suggests that it suggests two things one
00:22:23
is that these new kinds of AI increase
00:22:27
productivity
00:22:29
for workers so there's a test that's
00:22:31
being done on developers a research
00:22:32
study focused on developers using Code
00:22:35
generation AI
00:22:36
there's a research study that was
00:22:38
focused on chat gpdlex system to improve
00:22:41
writing and all of these show like
00:22:43
nearly two-fold
00:22:46
increase in productivity
00:22:49
with just these early forms of AI and
00:22:51
over time much greater productivity but
00:22:53
what these studies show is also that not
00:22:55
all workers benefit equally
00:22:59
the study with developers showed that
00:23:02
developers with the lowest skill levels
00:23:04
benefited much more than developers with
00:23:07
high skill levels
00:23:09
the study with writing showed that
00:23:11
writers who had the lowest writing
00:23:15
skills benefited more so than writers
00:23:18
with the higher skills
00:23:21
so one of the things it also shows is
00:23:24
that the new kinds of AI will affect
00:23:26
White Collar jobs for sure but they will
00:23:29
also empower
00:23:31
workers with lower skill levels and help
00:23:33
create an equal playing field and by the
00:23:36
way you know in in global Commerce today
00:23:38
English
00:23:40
just knowing English is
00:23:43
you know the paths to a job it's a path
00:23:45
to success and somebody who doesn't know
00:23:48
English it's you could have very high
00:23:50
intelligence very high skill but not
00:23:52
knowing English could itself be the
00:23:53
bottleneck you suddenly bring generative
00:23:55
AI
00:23:56
and you even the playing field for for
00:23:58
them and you can apply this for
00:24:00
developers you can apply this for many
00:24:01
other workers
00:24:03
thank you for listening to the ripple
00:24:05
effect we hope you found this episode
00:24:06
informative and engaging don't forget to
00:24:09
subscribe and leave us a review so that
00:24:11
we can continue to bring you the best
00:24:13
Insight from the warden School
