00:00:00
Well, with artificial intelligence being a growing part of our lives, both in personal and in business, many questions
00:00:06
are being asked about areas like productivity growth and just how much of an impact we will see from this area in
00:00:13
the future. A new report out by the Pen Wharton budget model takes a deeper dive
00:00:17
into these questions and pleasure to be joined once again by Kent SMEs who is faculty director of the Pen Wharton
00:00:23
budget model. He's also professor of business economics and public policy here at the Wharton School. Hi Kent, how
00:00:28
are you today? >> Good and great to be back. >> Thank you. All right, so let's start
00:00:33
with just how important you think it is to project this productivity path especially with how AI and technology
00:00:41
are having such a greater role. >> Yeah, absolutely. And you really see that with market concentration right
00:00:47
now. Um the bottom 493rd firms as uh the Apollo chief economist to Slack is documented that their
00:00:57
earnings projections really haven't changed but it's almost all the markets
00:01:01
driven by these seven firms that magnificent seven um and that's you know mostly an AI play and so AI is having a
00:01:09
very big um kind of uh salient you know uh impact the real question is how broad
00:01:17
is it going to across the economy and then um really can AI help deal with this major debt
00:01:27
problem that we have of of growing debt and that was the big motivation. And so what we basically did is we look at a
00:01:32
bunch of different occupation groups and really based on other experts and their
00:01:38
insights about uh the actual data and how to interpret things and everything from office administrative support
00:01:47
functions the business and financial operations, computer and math occupations. Those are the ones who are
00:01:53
uh most likely the greatest impact by AI. and then all the way down to like farming, fishing, construction and so
00:02:00
forth, buildings and groundskeeping. Uh those are the occupations are the least uh likely to be impacted. And so then
00:02:07
the big question is how much impact will this have in the macro economy and then
00:02:12
how much impact could this potentially have on government revenue um and then ultimately potentially even the
00:02:18
government debt over time. And there's been, you know, this belief by policy
00:02:22
makers that we're in this new era that AI, we don't have to be physically
00:02:27
responsible because AI is going to solve everything. And what we're showing is
00:02:32
that um that that's simply not true. We're not even nearly that's not even
00:02:38
close. Um and so they're making a lot of you know uh incorrect uh uh assumptions
00:02:46
about the impact of AI. Even with those incorrect assumptions, they're not then
00:02:50
mapping to the budget deficit problem correctly. >> Can you start to estimate then with so
00:02:56
many businesses bringing AI on board and I think the expectation that many others
00:03:00
will just how much of our GDP will be impacted by AI. >> Yeah, it's not a small uh uh amount. So
00:03:08
if you just look at the metric of uh of jobs that at least 50% of the task and so we have this taskbased model that is
00:03:19
uh very similar to Don Ace Mloo recent Nobel Prize winner in economics his taskbased model that he used for
00:03:25
modeling AI we get bigger results than he got uh for the impact of AI but nonetheless uh if you break it down
00:03:33
literally at the task level within each these occupational grips I basically about 40% of occupations or at least uh
00:03:44
based on the employment in those occupations are going to have at least 50% of their task that will be
00:03:51
essentially replaceable in the future uh by AI. So it's not a small impact by any
00:03:58
measure. um you know the uh primary author on this report out as the the supervisor of this report um Alex Arnon
00:04:07
he I think he put it really well uh to uh recently he says it's it's it's it's
00:04:12
roughly the impact of email um in particular email definitely had a big impact um yeah but it's not and if you
00:04:20
look across the occupations lots of occupations email email like technology um definitely had an impact grow We used
00:04:29
email this morning to communicate before this. And so, uh, but at the same time,
00:04:34
it's not one of those things that's like this magic bullet either. It's not
00:04:38
electricity. It's it's not refrigeration. It's not that transformative. And so because of that,
00:04:45
I is there an element of as AI comes in and the the level of productivity that we see grow in the years ahead that it
00:04:54
becomes to a degree kind of a known quantity just how much we're going to see the growth occur because companies
00:05:02
are going to add it and you know the technologies will change as we move forward. But once you install that
00:05:07
technology I think the expectation is you're going to do your job faster. you're going to have better
00:05:12
productivity, but it's not going to, you know, marketly change yeartoear. You're
00:05:17
going to kind of reach a level and continue to grow maybe incrementally. >> Yeah, that's right. And so almost like
00:05:24
so one of the things that uh uh we did in this study is we looked at adoption of past technologies those kind of
00:05:29
growth curves and rather fascinating um and we're talking about really the technologies as opposed to things that
00:05:37
like the invention of the transistor or you know refrigeration you know these general purpose big jumps that uh have
00:05:46
much bigger impacts. If you look at these adoption curves of technology and everything from the from the personal
00:05:52
computer to the smartphone to the internet to cloud computing rather shockingly their adoption curve are very
00:05:59
similar to each other where you get this what kind of call turnpike theorem like
00:06:04
stuff with pretty quick adoption but you can only go so far um you know 100% is is the upper bound and so many uh
00:06:13
occupations are going to be way uh less than that some occupations are going to quite a bit more than that than you know
00:06:19
50% you know we have office administrative support functions at 75% of uh their their task will eventually
00:06:27
be replaceable by AI whereas building and grounds clean and maintenance operation we said it's about 2.6% 6% and
00:06:34
we we literally if you're people are curious broken that down by over 700 different occupations um and then we
00:06:42
summarize it in big occupational groups but I think it is uh the case that you have quick adoption like you do with uh
00:06:50
internet and other technologies but then yes you get this leveling effect as a result of that what we find is that
00:06:56
during the first decade it will increase GDP by about 1 and a half% by you you know, within 30 years maybe 3% in the
00:07:05
long run by 2075 maybe 3.7%. So you're getting this leveling effect. Not everybody's been happy with this. It's
00:07:12
bigger than some you know results although we're kind of the first to kind of map it into this kind of this uh type
00:07:18
of change but nonetheless it is especially with the budget uh issues but nonetheless it is uh bigger than some
00:07:27
estimates but definitely smaller than the rhetoric that we are hearing about AI. And so what kind of impact might you
00:07:34
see play out with the federal budget as we move forward here in the next >> and this one is very much a work in
00:07:40
progress because the first step was on productivity more generally speaking and then GDP and
00:07:47
wages uh wage growth and so forth. The reason why it's harder to map that back
00:07:53
to the budget is that there we have to go category by category. So think about healthcare. Healthcare can actually be
00:08:00
improved by AI uh for certain things like scans and reading scans. I call it the TSA attack. When you go to the
00:08:06
airport, you don't have 50 people at the, you know, um the TSA uh line looking at your bags, but it's only
00:08:14
takes a couple people because we have very modern scanners uh right now. And so, uh you can imagine some innovations
00:08:22
uh especially for diagnostics, breeding, and so forth. But at the same time in healthcare
00:08:28
it's going to AI overall is going to increase cost simply because it's going
00:08:31
to allow us to cure more things and that's a good thing but there's costs
00:08:35
and so overall what we're finding for the first decade I would call this maybe
00:08:40
upper boundish maybe about $400 billion in uh uh deficit reduction uh through expended tax base but that is uh that's
00:08:51
a really rough first cut of all this and we're going to be looking at it going
00:08:55
forward. >> Kent, great to talk to you as always. Thanks, sir. >> Pleasure, sir.
00:09:00
>> Thank you. Kent SME, faculty director of the Pen Wharton budget model and also
00:09:04
professor of business economics and public policy here at the Wharton School.