00:00:00
welcome everyone to the Wharton serus XM podcast series on artificial intelligence sponsored by analytics at
00:00:06
Warton and artificial intelligence at Wharton I'm Eric bradow professor of marketing statistics and data science
00:00:12
here at the Wharton School and I'm also I guess for this role I'm also the vice
00:00:16
of analytics at Wharton and so I'm here interviewing thought leaders at the Wharton School on the impact of AI and
00:00:22
business and of course today's episode which is part of our multi-part series
00:00:26
is no exception today we're going to be talking about Ai and robotics or if you'd like Ai and automation we're going
00:00:32
to be I'm going to be talking to my colleague Lyn woo Lynn is an associate professor in our operations informations
00:00:37
and decisions Department here at the Wharton School uh she teaches everyone undergrads mbas and PhD classes about
00:00:44
the use and impact of emerging Technologies so Lynn Welcome to our podcast thank you thank you for having
00:00:49
me H it's great to be here with you um one of the things I even mentioned to
00:00:54
you briefly off air is that a lot of people might be listening to this podcast and serious XM version and
00:01:01
saying wait a second I thought AI was just just generative AI That's what I'm
00:01:06
hearing about today I'm hearing about chat GPT and B and open AI what the hell
00:01:11
does AI have to do with automation so I thought maybe for a few minutes it would
00:01:14
be good if you take our listeners through a history of artificial intelligence and what does it mean to
00:01:21
have artificial intelligence in Automation and then kind of how generative AI fits in into the broader
00:01:25
class of AI oh thank you that's a great question so AI as a field has been existing for decades like from the 1950s
00:01:33
and 60s that's when people started thinking about coin a term like artificial intelligence and like you
00:01:39
know you probably if you were around the 70s 80s you probably know about expert systems which is just bunch of rules if
00:01:46
them classes if you see a and then this B happens and they think that will Co cat out lots of you know medical
00:01:52
knowledge or other type of knowledge out there and that didn't work so well and
00:01:57
then come the neon networks and the neuro network was happened around the 1780s again and at the time it wasn't uh
00:02:05
performing very well because as you know neuron Network needs a lot of data and come the 90s actually we had AI winter
00:02:12
basically near wasn't working a lot of AI techniques weren't working so they
00:02:17
actually thought AI is not going to happen for a long time and then we had the internet explosion we have all the
00:02:22
digital Trac of our you know internet activities our search activities our social media our videos our photos that
00:02:30
data explosion ultimately fueled the current AI Revolution because now the neuro network has ton of data and then
00:02:39
you can build up very very big neuron networks which we call Deep learning okay and deep learning was was you know
00:02:46
was driving the AI Revolution from 2010 to about 2018 and then a special type of
00:02:53
AI techniques called Transformers which is basically you think about just make AI neuron networks
00:03:00
run really fast computationally very efficient and that again saves a ton of resources and that allows generative AI
00:03:09
so that's like although I'm really simplify things that that's like just
00:03:13
what it is so if you think of generative AI is only a tip of the very end and although it's transformative I I gu I
00:03:20
mean I absolutely agree with you with if you see chbt and Dolly is amazing but is
00:03:25
really this extension of existing technology in the Ro in deep learning yes just in this case predicting it to
00:03:32
language so these are large language models but there's as you're as we're
00:03:36
going to talk about today they're applications of AI in many many different areas and also I think it's
00:03:41
also valuable for our listeners to know you know a lot of us well a lot of us trained back in the old days as
00:03:47
statisticians but even all of us are familiar with regression which is obviously a form of linear prediction
00:03:53
models and neural Nets obviously and deep learning just allow for much more complicated interactions between
00:04:00
variables and that's what I think I maybe you could just comment briefly on this that's why these Simple Rules tend
00:04:05
not to work well because the way real life Works isn't just simple if thens it's not like a linear regression more
00:04:13
is necessarily Better or Worse there are these complicated relationships that's
00:04:18
exactly right so often we cannot describe a relationship using a simple representation or a linear model what
00:04:25
neur net word does is it you can have link very variety of different functional form in a way that we cannot
00:04:32
describe any at all but machine knows how to transform input output through many different layers of Transformations
00:04:41
well let's actually talk about the main topic and first of all thanks Lynn for
00:04:44
bringing that up because again a big part of this series A lot of people are talking about generative Ai and you've
00:04:50
just pointed out it is just a special case and a specific application area so let's talk about this a matter of fact
00:04:56
one of the things I present to my MBA students which is interesting that it's
00:04:59
something that you wrote down in advance as a question is that previous Studies have indicated that between 40 to 70% of
00:05:05
jobs could be automated so let's start with like which jobs do you think could
00:05:10
be automated and two what would prevent that from happening or there is kind of no way back at this point we're talking
00:05:18
about generative AI specifically like the TP like this could be or it could be for example now Automation and and
00:05:27
artificial intelligence is going to allow Rob OTS to do lots of physical tasks that we couldn't do before so it
00:05:33
could be Vision based types of jobs and opportunities it could be I don't know
00:05:38
robots doing surgery it could be robots replacing people in Plants it could be in lots of different ways okay great so
00:05:44
let me be more concrete I talk about robotics okay so how how physical robots can change labor composition right
00:05:54
so contrary to the popular notion that 40 60% of employees are actually getting laid off when they have no jobs so in
00:06:01
our longitudal research over about 20 years using Canadian data where we capture every single robots being being
00:06:08
used in the in the firm in establishment and look at the you know what kind of employees are working with the robots
00:06:15
who are there and who are not there anymore and the the revenues their firm practices in terms of HR Management in
00:06:23
terms of uh how uh people are being rewarded so we found it uh interestingly that when firms adopt robots they
00:06:32
actually hired more people they did not lay off more people very interesting why so why is that it
00:06:39
actually turns out um it's a robot adopters who are becoming much more productive and much more efficient so
00:06:48
they grow their pie bigger and hire more people so it turns out it's not a robot
00:06:55
displacing people directly is a robot nonadopters are no longer competitive and they are laying off people okay so
00:07:03
this is actually as a statistician this is an interesting what I'll call self-
00:07:07
selection problem so the fact that someone has adopted robotics is indicative of the likely growth pattern
00:07:15
of the firms and faster growing firms tend to hire more people is that the is is from a selection bias story is that
00:07:25
the selection argument that's being made uh that's a very good uh point so we we
00:07:30
actually looked at the type of people that are being laid off that was going to be my next question right so I think
00:07:35
I can talk about statistics the methods we did but I think it's really probably
00:07:38
easier intuitive to understand but let me just understand if I hire a robot if I hire if I build a robot or utilize a
00:07:45
robot to do a specific type of task are you suggesting that the firm hires more people to do that type of task or the
00:07:53
firm hires more people in general to do other tasks now this task is done more efficiently this firm can produce more
00:07:59
units and now I need more people in other jobs it's ladder your it's really
00:08:03
ladder the ladder okay but no one is doing every everything like robots are not doing everything a human can do yet
00:08:11
so you are robots are replacing part of a Tas a human worker used to do but increase the demand of the other task
00:08:18
that other that person used to do so if you are a manager and you mostly looking
00:08:24
at quality assurance issues if that robot's taking care of that you can do more leadership you can do more people
00:08:31
management like you do other work that robots you know can do yet I see now your paper finds that the case for
00:08:40
employment this this this to me I I love it when people find things that are counterintuitive your paper finds is
00:08:46
what I've written down here that the case for employment is reversed for managers that is more robots can equal
00:08:53
fewer bosses so tell tell us about that cuz that must have been one of the to me
00:08:58
intriguing finding I I think most people would say no it's the I'll call it lower
00:09:02
level laborer that's going to lose their job but managers someone has to make
00:09:06
decisions you're fine yeah that's also a very intriguing finding we we thought we
00:09:11
got a WR at first so we really double checked triple by dble checked and then we find out why um it's are two effects
00:09:17
one direct effect right I mentioned the monitoring technology can't figure out
00:09:23
what your employees are doing right you think about a laptop well you can't even
00:09:27
figure out the every key employee can type if you if you want to right and uh uh you know in a warehouse situation
00:09:34
you're clocked in each box coming in they know exactly what you what you've
00:09:38
done so you don't need a foreman or a you know person who will see you're
00:09:42
actually doing the work because there's all automatically captured and then your
00:09:46
performance reports automatically generated so that part of the manager TI is being directly substituted so you see
00:09:54
the very beginning within zero one year of adoption you see a slight dip in man managerial work but then after 2 three
00:10:02
year you see a really big downturn a very sharp downturn so why is that right that's not no longer a direct
00:10:08
substitution story anymore okay what happened is that the composition of Labor the employees at the firm changed
00:10:16
right so if you think about what robots can do it turns out robots are really good at doing the middle skilled work
00:10:23
part so they're they're not very good at you know doing thought leadership you
00:10:27
know thinking about what to do that's what humans are doing we need to figure out our goals we need to you know
00:10:32
figure out what to do and execute it and write plans are execute it and then but
00:10:37
robots are still not very good at some of our like what our hands can do like top tail SS are like you know like the
00:10:43
residual task somebody still need to you know pick up a ball and put in a box like you know in various shapes and form
00:10:51
our hands can handle infinite shapes where a robot you got to train a lot of time just hand handle one type of shape
00:10:57
so that versatility although people are still are are working on making a lot of Anis on this
00:11:02
area but the versatility of the robots is not there yet you can train a very narrow task but you cannot generalize
00:11:10
very quickly to lots of tasks and do you think the you know if you've like the
00:11:15
explosion of Transformer models so the ability to handle richer and bigger and larger data sets in real time is even
00:11:24
going to threaten that that in some sense robots are going to be better at very let's call them fine motor tasks
00:11:31
and complex tasks and it's just a matter of time even before let's call it the pi
00:11:36
expands to even more fine motor types of jobs people are definitely working on that but I do want to caution you that
00:11:43
for robotics area it's the data generation it just not there as a AI because AI you can simulate lots of
00:11:51
different kind of datas right like you were pictures your videos automatically there on the internet you can just grab
00:11:56
it right but if you want to train robots but you got to train them in realistic settings when humans are involved so why
00:12:04
can't you have let's as a form of by the way the term you might call it a form of
00:12:08
supervised learning why can't you have a bunch of humans doing a task I actually
00:12:13
have video cameras with extraordinary maybe the person has sensors all over her his their body and now all of a
00:12:19
sudden I've got this extraordinarily Rich data set of movements and actions and stuff and matter fact I'm going to
00:12:24
use AI to train the model to start using robotics to replace those tasks is that
00:12:31
because companies are not doing these types of I'll call it measurement experiments in large scale or why can't
00:12:36
that be done I think you can do that but again it's only if you're a very narrow
00:12:41
set of tasks unless that task has extremely high value to the firm that makes sense right and also most of the
00:12:49
robots you can't just plug in a robot in a firm and then expect it to to be like
00:12:53
autonomously working on its own so there's always a human monitoring or you know working with robots and many lot
00:13:00
the cobots like a lot of the new development and using uh you know more advanced technique are cobots and these
00:13:05
cobots is usually a human machine interaction in those scenarios right you actually have to generate human robot
00:13:13
interactive data and that's something you cannot just simulate you can't you
00:13:17
actually have to capture right you you can't you can put a sensor all over me
00:13:20
or robots but you got to catch me every everywhere and that's just hard to capture right like you know you can't
00:13:26
just put in a you know car in a road right I capture all this stuff but when the hum involved is expensive right
00:13:32
capturing that data is really expensive so in terms that data like I would say robotics at least 10 years behind in AI
00:13:39
in generating the techniques that can do what gener AI can do I see we're here on
00:13:43
the Wharton serus XM podcast series on artificial intelligence sponsored by analytics at Wharton and AI at Wharton
00:13:49
I'm here talking to associate professor of operations information decisions ly
00:13:53
woo uh we're talking about uh Ai and Robotics so let me ask you um how do firms how should firms think about their
00:14:02
employees now or how should employees think about the skill sets they need to develop to kind of continue to thrive in
00:14:10
a world where I think we agree it's unlikely robots are going to be doing less overtime probably more overtime
00:14:17
what Can employees do and how should firms think about managing them I think one of the biggest thing human tend to
00:14:24
underestimate is that you we have a lot of tacit knowledge about how work is being done right so robots can only
00:14:32
observe you know a small part a lot of our knowledge not codified for machine to consume so the more knowledge you
00:14:40
have that's not codified the better off you are because there's no way robots
00:14:44
can use that knowledge to replace you but then again that's very abstract right but if you've been if you have
00:14:49
been in a firm in the industry for a long time you have a deep expertise knowledge by the industry or by that
00:14:55
firm is always going to be good because that know knowledge cannot be easily transferable outside has anyone I don't
00:15:01
know I'm sure someone has studied this are employees strategic like for example
00:15:07
maybe I you know it's the classic I teach my student but I don't teach them
00:15:11
everything I know I teach my student you know strategically enough but if I give
00:15:17
away everything then in theory maybe I could be replaced by a robot is there any studies that have been done on the
00:15:22
Strategic nature of employees I am sure it has been done I think more will be will have to be done because as you know
00:15:30
lot of PE lot of firm these days are using human the TR robots I am sure those kind of strategic Behavior will
00:15:36
become more and more uh prevalent there are some literature in in the back in the 6070s but I think that needs to be
00:15:41
updated I see so do firms have a choice like is there any choice firms really have or in some sense like you know I do
00:15:51
a lot of work in sports do I really need some former 30y old sorry 50 60y old baseball player throwing batting
00:16:00
practice why don't I just get a robot to do that you know I study things in high-tech do I really need some sort of
00:16:07
H human supervision anymore where I can just have a robot do it are there any specific like how should confirms really
00:16:15
compete in scale nowadays without adopting Robotics are there's any or another way to frame it is is there any
00:16:21
industry you think that is essentially immune to it based on my study in Canadian data I think he's not you
00:16:29
because adopters are really just killing the competition and unless nobody in your industry collect to decide not to
00:16:34
adopt robot I think you're then but you you worry about new entrance coming in
00:16:39
with a robot so I think you have to be forward looking at least and think about how robots is going to be affect your
00:16:45
firm and productivity but the most important thing is not like oh my God my competitor do robots I need to do
00:16:51
immediately and I I I would caution that because the value of adopting robots is
00:16:58
is really to help you understand your existing firm processes and how to help you understand which part can be
00:17:04
automated and which part can be strengthened to complement that automation but in fact I would say nine
00:17:11
out of 10 90% of value come from studying that process understanding where it can be automated where it
00:17:19
cannot be automated and how to strengthen that relationship between human a machine that collaboration so
00:17:24
this is not a a do robot buy some expensive hardware and get some consultant and you're done right this is
00:17:30
required people in your firm been doing that process for a long time and and from them they can tell oh you know this
00:17:37
part of process makes sense actually yeah sure the robots can do it but I don't think it's a good idea because
00:17:42
this is a very risky operation if something happen to the you know it's like a something happen to the person
00:17:48
it's very dangerous or high like Risk situation it should not be should not be
00:17:52
automated even if you it can be done so it's really understanding that process
00:17:56
deciding human decision year understand understanding where it can be automated where it shouldn't even if it can that
00:18:02
is where the value come from so let me ask a a question about it's really a two-part question but it's really the
00:18:08
same question which is so how big an effect are we talking about here so a firm hires a robot and then some measure
00:18:16
and this is the second part but it's really the same question like first of all what are the in our language as
00:18:21
academics what are the dependent variables that people worry about so let's let's imagine you have a model
00:18:26
that says if I adopt robots to a certain degree to do a certain task my let's say
00:18:32
I care about sales or market share or employee retention or employee satisfaction what depend what outcome
00:18:39
variables do people tend to study when they think about the adoption of robots that's number one and second you know
00:18:46
how big an effect size are we talking about here like if all the ways that a firm could improve efficiency and
00:18:51
profitability is this in the top five and how big an effect are we talking about that's a great question so
00:18:56
obviously as a as you know a Economist adti we capture any variable we have in our you know Arsenal right Revenue
00:19:03
employee satisfaction individual productivity performance right and that's a great thing about the Canadian
00:19:08
data set is that we actually have a lot of that data we look at employee satisfaction there are their incentive
00:19:15
pay system so we were able to capture a lot of this um so yeah it it is a very important study in terms of what firm
00:19:22
should like what the effecta size is u i want what did you find like how big an effect are we talking about what happens
00:19:29
I'm just saying let's say you know the way we always tend to think about it is
00:19:32
there's always three populations at least the way I think about it is what happens to firm let's even just say
00:19:39
we're thinking about profits or Surplus what happens to the firm what happens to
00:19:43
customers and what happens to society yeah let me just go by the revenue like proct profit first right so if you think
00:19:50
about a profit maximizing firm right you should you should like so you know I adop one robot I got $10 and if my robot
00:19:58
is only costing $1 right that means I should just buy more robot until I got I got I reached exactly 10 robots and then
00:20:06
I got the same same same bio and some right I put in $10 robots I get $10 of output out right right that's should be
00:20:12
the problem math firm right so the fact what we found is that robot is 10 times that factor share so that's 10 times was
00:20:21
value so I you put in one you get 10 back that's what we found but there has
00:20:25
to be some sort of diminishing marginal returns to robots it's like the second
00:20:28
robot's got to be worth maybe a little bit less of course my cost might go down
00:20:32
as well yeah exactly so there's that but I what I what I what I want to emphasize
00:20:37
This Not That robots responsible the the whole 10 here nine of them come from process Improvement I got it the study
00:20:46
that the stuff you do to make robot work so the robots one because that's makes
00:20:51
sense you possibly maximize it one but the nine is what you do with it I see and what about for employees like um so
00:20:58
you could imagine one of two things one is the employees that are still remaining now that the firm's making
00:21:03
more money you could argue that the employees are in better shape than before I don't know what do you find in
00:21:08
terms of employees that's great uh what we found is that is precisely when human
00:21:12
are working with robots like robots make different kind of Errors than humans right so ifite you and I work together
00:21:19
we could collude in a sense let's Shir today right it'll be say our machine
00:21:23
died so let's just have a coffee right I'm not saying it's going to happen but
00:21:26
we could right and a manager sitting far away where cannot see you or know a headquarter couldn't figure out okay
00:21:34
they cannot directly observe observe you oh okay I guess the machine wasn't working I I repair it right but when you
00:21:41
are working with a robot you can't do that anymore all right because robots are doing these things these kind of
00:21:46
Errors won't is less likely to occur but because human or robot have different
00:21:51
errors that means when you're working with a robot I can tell your performance
00:21:56
I can capture your performance more accurately than you work with the human so what happened is that firm adopting
00:22:04
robots also changed the performance pay practices I see so they actually become like the they're rewarding the high
00:22:10
performers more so than before like you are actually good performers is not because you were lucky or some other
00:22:16
stuff so good performers in some ways should be happier with robots cuz in some sense there's less measurement
00:22:23
error in their performance y that's exactly what we found there's there less
00:22:28
team based uh promotions uh incentive pay but mostly more individual based performance pay to what extent is going
00:22:35
to happen in the future I don't know because in we we see a lot of problems in the individual performance pay too
00:22:40
but at least right now the the the swing the pendulum is more to individually based incentive pay and how about you
00:22:46
know policy makers so how should like should this be a regulated market or is this just you know I think most people
00:22:55
on the surface would say this is bad for society because it's replacing paying
00:23:00
jobs for humans although on the other hand there could be greater efficiencies how should policy makers and Society
00:23:07
think about it I think um I think the genie is out a bottle you can't put it
00:23:12
back anymore you can't just say let's not use AI because the US doesn't use AI
00:23:16
the other country would and so as we we have to be competitive the national front as well in terms of Regulation I
00:23:24
think is really important we we actually study the phenomena to Greater detail we
00:23:28
now we see gener AI people getting worried but we really have to see evidence does it actually make consumer
00:23:37
worse off right does it actually hurt people because in my because you know remember in that my robot study is
00:23:43
premis by saying robots will kill 40 to 50% of employees employment but we found
00:23:49
opposite we actually found employment to have gone up quite a bit so we really have to have large imperical evidence
00:23:56
first before we make any significant policy changes because we don't want to
00:24:01
you know we don't want to throttle all this process we want to AI to grow before we kill it completely just
00:24:10
without having evidence that it's going to actually have significant I'm not
00:24:14
saying it will have it won't have harm it it will but we need to watch and monitor closely before we make any
00:24:19
decisions so maybe in the last minute or so that we have could you tell me if we
00:24:24
were sitting here 10 years from now and maybe we will be what so as we look back
00:24:29
on those 10 years what are we saying this has been the big advance in Ai and Robotics oh I think generative AI would
00:24:35
is for sure the most prominent evance in NLP it pretty much wiped out all the existing natural language processing
00:24:45
techniques out there this is de facto replaced it so I think uh this entire field needs to be revolutionalized uh so
00:24:53
in terms of in terms of I think the next 10 years we're going to see may not be
00:24:58
necessarily a technology advancement in that area per se is really about application how is that P marketing how
00:25:05
does it apply to schools and education how does it apply to you know writing how does it apply to reporting like as
00:25:12
Industries are starting to find novel use as technology then that's where we're going to see how amazing that just
00:25:19
like think of internet internet was invented 1970s but we didn't really see made how the potential of it 192000 with
00:25:26
right with with the internet with e-commerce so I think uh generi may be faster but it will still be a decade or
00:25:34
two until we see the real impact well I'd like to thank my colleague uh associate professor of operations
00:25:40
information decisions Lin moo for joining me today on the Wharton sirusxm podcast series on artificial
00:25:46
intelligence today we've been talking about Ai and automation uh we have plenty of episodes to go but Lynn thank
00:25:51
you for joining me today thank you for having me this is so much fun