00:00:00
Well, as AI has become more of a growing
00:00:02
part of our professional lives, the
00:00:04
question has been asked about how much
00:00:06
it might impact a thing like the
00:00:08
workplace. And that question has
00:00:10
seemingly been buried by many members of
00:00:13
the seauite up until now. We're starting
00:00:16
to hear leaders of companies have
00:00:18
comments about just how many of their
00:00:20
employees might no longer have a job in
00:00:24
the years ahead. To more on that, we're
00:00:26
joined by Peter Capelli, who is a
00:00:28
management professor here at Wharton.
00:00:30
He's also director of the Center on
00:00:32
Human Resources. Peter, great to talk to
00:00:34
you again. How are you, sir?
00:00:36
>> Good. Thank you, Dan.
00:00:37
>> I guess it's probably not a surprise
00:00:39
that at some point we were going to have
00:00:41
to see leaders of companies address this
00:00:45
issue because it's just one that is
00:00:47
seemingly on the forefront of a lot of
00:00:49
people's minds.
00:00:51
>> Yeah, it's a good thing that they're
00:00:53
addressing it. I think as we'll see in a
00:00:56
couple of minutes in the conversation,
00:00:58
unfortunately they what they really
00:01:01
should be telling people is a lot of we
00:01:02
don't know yet. Uh and I think I'm not
00:01:06
sure what they're telling them is the
00:01:07
most useful thing yet. It's kind of
00:01:09
scaring people.
00:01:10
>> Right. Well, but the comments uh from
00:01:12
people like Jim Farley, the Ford CEO,
00:01:15
saying half of all white collar jobs
00:01:17
will be gone. Uh is that an overestimate
00:01:21
or is that to a degree somewhat on
00:01:23
point?
00:01:25
Well, uh, we've had those, uh, kind of
00:01:28
forecasts for about 20 years now and
00:01:31
they haven't worked out, right? So, I
00:01:34
would say if you were a betting person,
00:01:36
you would bet against that and nothing
00:01:38
I'm seeing right now suggests that
00:01:40
that's going to happen. Um, but, you
00:01:43
know, the effects might be very
00:01:45
different depending what kind of job
00:01:46
you've got.
00:01:47
>> Why do you think that is that maybe
00:01:49
those some of those estimates are maybe
00:01:52
a little bit high? What is it about, I
00:01:54
guess, the structure of the company, the
00:01:55
jobs themselves that that you see that
00:01:58
is still going to require that human
00:02:00
connection?
00:02:01
>> Well, I think we really don't we know
00:02:03
very little about how these tools can
00:02:07
actually be used. So, the story has been
00:02:10
driven largely by people who build them
00:02:13
talking about what could be done. So we
00:02:16
will all remember by 2019 driverless
00:02:20
trucks would have taken over and you
00:02:22
need to get rid of your truck drivers
00:02:23
because there was you know they were
00:02:25
going to be obsolete and there were
00:02:26
companies that actually did that and
00:02:28
then of course they got completely
00:02:30
surprised when it hadn't happened at
00:02:33
all. So, you know, the forecasts have
00:02:35
been wildly wrong. And if you are a a
00:02:39
person concerned about risk, I would not
00:02:42
pay much attention to the forecasts
00:02:44
because they have been wildly wrong.
00:02:46
They've been driven largely by people
00:02:48
who build the systems, thinking about
00:02:51
what they could do rather than what's
00:02:53
cost effective to do and what is sort of
00:02:56
reasonable to do. Right? So, we're only
00:02:57
getting a sense of that now. Is there
00:03:00
some of this that we may see companies
00:03:03
make some of these moves and then have
00:03:05
to do a 180, a U-turn to bring people
00:03:09
back in as they learn what they can and
00:03:12
can't use?
00:03:14
>> Yeah, I I understand there's already
00:03:15
been a little bit of that in various
00:03:17
places, but yes, I think there we
00:03:20
probably will see that. And I I think
00:03:22
there's a a reason for it that u we may
00:03:25
have talked about before and that is you
00:03:28
know in the finance world there is a
00:03:31
real preference for companies to squeeze
00:03:34
down headcount. Part of the reason for
00:03:37
that is companies are assessed on profit
00:03:39
per employee profit you know cost per
00:03:42
employee everything per employee. So if
00:03:44
you can get your headcount down that's
00:03:47
really a nice thing. And employers don't
00:03:50
think very much about the costs of
00:03:51
losing people. And some of that is
00:03:54
because human capital has no accounting
00:03:55
value. Uh and so you know if you lay
00:03:58
people off usually the investors if they
00:04:01
do anything they just cheer. Nobody's
00:04:03
thinking about what has to happen if in
00:04:06
fact you fall short of talent. What
00:04:08
happens then? You know we know what the
00:04:10
answers are and they're not pretty but
00:04:11
in the investment world we're not seeing
00:04:13
them. So I would imagine that you
00:04:15
probably could bet on some of that
00:04:17
happening. Yes.
00:04:18
>> And there are probably and it's and this
00:04:20
may be a question where you know it's
00:04:22
going to depend on the firm or the
00:04:25
sector but there probably are jobs that
00:04:28
as you go along and you look at these
00:04:30
different companies different sectors
00:04:32
that there will be jobs that would be
00:04:34
more susceptible than others within the
00:04:36
firm to potentially get cut.
00:04:40
>> Sure. I think that's right. Now, what
00:04:42
we're learning is those jobs appear to
00:04:45
be the exact opposite of what people
00:04:47
said they would be. So, you know, up
00:04:49
until just a little while ago, people
00:04:51
thought the simple jobs would be the
00:04:53
ones most easily replaced. And that is
00:04:57
not turning out to be the case. One of
00:04:59
the reasons why is in simple jobs like
00:05:02
sorting and coding, right? Or moving
00:05:05
data from one pile to another, one set
00:05:07
of documents to another. The problem
00:05:10
with that is it has to be absolutely
00:05:13
right. And to make sure it's absolutely
00:05:16
right, you have to build the algorithms,
00:05:20
train them on real data to make sure
00:05:23
that it can tell what is in pile A and
00:05:26
what should be in pile B. And it takes a
00:05:29
long time to get that right and it's
00:05:30
really expensive to do. On the other
00:05:33
hand, things which are you can do right
00:05:35
now with no training is ask the internet
00:05:38
to provide a summary of what we know
00:05:41
about the tire industry in China, right?
00:05:44
Uh those answers are probably not going
00:05:46
to be that great right now. They won't
00:05:49
be as good as some expert would give
00:05:50
you, but they're cheap and they're free
00:05:52
and they're quick, right? So, it turns
00:05:54
out that if you're doing some knowledge
00:05:56
work, um maybe those jobs are pretty
00:05:59
susceptible. But even there, the reality
00:06:02
is nowhere near the hype. Let me give
00:06:05
you an example. If you look at computer
00:06:07
programmers, which is one of those where
00:06:09
people are saying they're just going to
00:06:10
be obsolete, right? Well, there have
00:06:12
been for a long time tools that would
00:06:14
help you automate programming. They'll
00:06:16
suggest code that other people have
00:06:18
already written on this topic you're
00:06:20
trying to do with this question. So,
00:06:22
that stuff's been around for a while.
00:06:24
AI, generative AI, arguably better than
00:06:26
that. But when you look at what
00:06:28
programmers actually do on their job, it
00:06:32
appears they only spend about 30% of
00:06:34
their time actually coding,
00:06:35
>> right?
00:06:36
>> So what are they spending the rest of
00:06:37
the time on? Negotiating a budget,
00:06:39
talking to their clients and users to
00:06:41
see what is it you actually need.
00:06:43
Negotiating between the budget and the
00:06:45
clients. Here's what we think we can do.
00:06:46
That's where they're spending their
00:06:47
time. So if AI took over completely the
00:06:51
programming task, you still got 70% of
00:06:54
the work that so far it looks like
00:06:56
people do.
00:06:58
>> So what does this potentially mean then
00:07:02
as we look down the road for the next
00:07:04
generation? Because obviously we're
00:07:07
learning about a lot of this right now.
00:07:10
that next generation coming into the
00:07:11
workforce, they have to I guess if you
00:07:14
know if you're in high school now and
00:07:15
you're going to college now, you have to
00:07:17
have the expectation that AI is going to
00:07:19
be part of the mix. But you know, how
00:07:22
much could it potentially develop? And I
00:07:24
guess to a degree this there's a lot of
00:07:26
still unknowns that haven't been
00:07:28
answered yet.
00:07:30
>> Yeah. And I think that is the point. Uh
00:07:32
what is a mistake is making a guess and
00:07:35
investing a lot of money in that guess.
00:07:38
So saying for example, you know, they
00:07:40
can never automate art. Uh so I'm going
00:07:44
into art and then it turns out they can
00:07:46
automate art pretty well. You know, it's
00:07:48
just hard to know. So don't place a
00:07:52
single bet and say this is the field I'm
00:07:54
going into because, you know, we can't
00:07:56
automate it. And some of the ones that,
00:07:58
you know, it we thought they were going
00:08:00
to automate pretty easily, it turns out
00:08:02
it's hard to do. So I guess I would just
00:08:04
not worry about it all that much trying
00:08:07
to guess where the technology will go. I
00:08:10
think the problem companies have right
00:08:12
now and I think you hear it from the
00:08:14
CEOs is that they're under pressure to
00:08:18
respond to the hype and the hype is AI
00:08:22
is the way to cut headcount. So get busy
00:08:24
and do it. Right. Right. So there was an
00:08:27
interesting survey uh of top executives
00:08:30
saying 74% of them said that they felt
00:08:33
their job was on the line if they
00:08:35
couldn't deliver these cuts basically
00:08:37
and they can't. They also said about a
00:08:40
third of what they're doing with AI is
00:08:43
performative. It's not actually doing
00:08:45
what they say it's doing and they're
00:08:47
kind of pretending. That doesn't mean
00:08:48
they're necessarily lying but they're
00:08:50
counting a lot of stuff as AI which is
00:08:52
not even AI at all. Right? So the
00:08:55
problem is we don't know. The CEOs are
00:08:59
under pressure because the boards in
00:09:00
particular think we do know and so
00:09:03
they're responding in ways which are
00:09:04
probably going to be dysfunctional.
00:09:06
>> Peter, great to talk to you and get your
00:09:08
insight as always. Thank you, sir.
00:09:10
>> Thank you, Dan.
00:09:11
>> Peter Capelli, management professor here
00:09:13
at the Wharton School and director for
00:09:15
the Center on Human Resources.
