YT YouTube2Text

YouTube2Text → Transcripts → How AI Is Changing The Career Ladder | Stanford Economist

How AI Is Changing The Career Ladder | Stanford Economist — Transcript

by McKinsey & Company · 5,977 words · 929 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Ultimately, no company, no country, no
0:03person has ever succeeded by just
0:05focusing on the same thing over time.
0:07You need to expand your horizons. And
0:09that's what's helped make America
0:11successful, and I think the successful
0:13companies are the ones that do that as
0:14well.
0:15>> Hey everyone. That was Erik [music]
0:17Brynjolfsson talking about the risk that
0:19comes from resisting change. He is a
0:22professor of economics at Stanford
0:24University and director of the Institute
0:27for Human-Centered Artificial
0:28Intelligence. Brook and Brian and I met
0:31up with him on the Stanford campus. We
0:33talked about AI and GDP, what
0:36organizations can do differently to
0:38realize new productivity gains, and
0:41[music] how we should rethink what the
0:43initials AI stand for.
0:47>> [music]
0:48>> This is McKinsey Talks Talent. I'm your
0:50host Lucia Rahilly with talent experts
0:52Brook Wedel and Brian Hancock.
0:55>> [music]
0:58>> Professor Erik Brynjolfsson, [music]
1:00welcome to the podcast.
1:02>> Thank you. Good to be here.
1:03>> It's great to have you.
1:04And Brook and Brian, great to see you as
1:06always.
1:06>> Great to be with you.
1:07>> Great to be here.
1:09>> So, let's start with a little context.
1:10At McKinsey, we talk a lot about the AI
1:13paradox or what we call the AI paradox,
1:16the gap between this massive investment
1:18that organizations are making into AI
1:21and the um appearance of meaningful
1:25at-scale enterprise value and returns.
1:28Your research shows that we may be
1:31turning a corner on that. Talk to us
1:34about the J-curve and and where you
1:35think we're going.
1:36>> What we're seeing is what we call the
1:38productivity J-curve, that
1:40these powerful technologies, to get the
1:42full benefit of them, you also need to
1:44make process changes, you need to
1:46reskill the workforce, sometimes you
1:48need to invent new products and
1:50services, and that all takes time.
1:52That's actually the hard work of it,
1:53not, you know, buying the LLM. While
1:55you're making all those intangible
1:57investments,
1:58um there's a lot of work and it doesn't
2:00instantly translate into new output. But
2:02then once the intangible investments are
2:04in place, then you can start harvesting
2:06them, and things start taking off.
2:09I'd say we're seeing some evidence that
2:11it's just beginning to show up in the
2:12national productivity statistics. I
2:14think it's actually too early to say for
2:16sure, but we're definitely seeing it at
2:18the micro level for specific companies.
2:20When it's rolled out, you know,
2:21effectively, you can get really big
2:23gains. The issue is that that's not
2:25something that's happening pervasively
2:26through the economy yet. I'm sure it
2:28will, though.
2:29>> What we're finding is that you know,
2:31when we're working with organizations, I
2:33would say 70 to 80% of the work isn't on
2:37what should be automated or what is the
2:39tech enablement part of the workflow.
2:42It's what are the new roles? What are
2:44the behaviors of those people in those
2:46new roles? What are the skills required?
2:49Even what are some of the underlying
2:50mindsets that are going to enable
2:52success? Is that what you mean by doing
2:54the workflow?
2:55>> Yeah, you have to sort of get in there
2:57and get into the nuts and bolts. But one
2:59way that you can get more visibility
3:00into it is break the occupations down
3:02into tasks. Every occupation is a bundle
3:05of tasks. And typically, AI can help
3:07with some of those tasks, you know,
3:08helping you write a memo or analyze some
3:11documents. It can't help with other
3:13parts of the job, you know, if you have
3:14to lift a box or something, the LLM's
3:16not going to help you. And so every
3:18occupation we looked at has some parts
3:20of it that are going to be more
3:21effective than others. And that means
3:23managers need to sort of reconfigure and
3:24reorganize and rebundle those tasks.
3:27That's actually takes a lot of
3:28creativity.
3:29>> Yeah. I was I was actually intrigued by
3:32something I I heard you say in a
3:34different context, which was when you
3:37think about the agenda for AI and AI
3:39innovation, we should really be thinking
3:41about trying to hone AI as a capability
3:46and tasks that humans aren't good at
3:49doing, right? And so as you think about
3:52the work you've done on what is the
3:54automation potential, how do you factor
3:56that point in?
3:57>> So many technologists are obsessed with
4:00making AI that imitates human. Like Alan
4:03Turing said, the Turing test, the
4:04ultimate test of AI is if you can make
4:06an AI that you can't distinguish from a
4:08human, that perfectly imitates it. I
4:10think that's maybe kind of a cool
4:12concept, but it's a terrible business
4:14strategy. Um
4:15>> [clears throat]
4:15>> instead, what you should be doing is
4:17having the AI focus on what AI can do
4:19well and not what humans are already
4:22doing well. Um and that's something that
4:25um
4:25makes AI more of a complement rather
4:28than a substitute. And most of the
4:30managers I talk to, um their first
4:33instinct is to have AI sort of replace
4:35the workforce. I'll I'll tell you I I
4:37was talking to a CFO not that long ago,
4:39and she was saying, you know, we really
4:40need to measure the ROI of AI better.
4:42And I was like, absolutely, you got to
4:44measure and she said, "And therefore,
4:46we're going to go through each
4:47department and see how much head count
4:49reduction they're getting from AI." And
4:51I was like, well, okay, that that's one
4:52measure.
4:52>> [laughter]
4:53>> It's It's a very narrow measure, but
4:54it's so pervasive, this sort of cost
4:56mentality. Cuz let's face it, it's the
4:58easiest thing to measure and people
5:00think of AI replacing people. But the
5:02bigger upside is if you can get AI to
5:05allow those people to do new things that
5:06they've never done before.
5:08>> Yes.
5:09>> So So, what are some practical examples
5:11of where AI can extend the human
5:14>> I mean, one of the places where we're
5:16seeing it happen the fastest is in
5:18coding. And it's nice because you see
5:20there, you can see some of it
5:21substituting and some of it augmenting,
5:23often within the same occupation for
5:25different parts of it. And what we see
5:28is that some of the routine tasks, AI
5:30agents can do those. But what's
5:32happening is the senior coders now are
5:34finding themselves augmented. They're
5:37able to do so much more. They are
5:39working with fleets of agents, not just
5:41one or two. Some a lot of the routine
5:43execution, the machines can do that
5:44better. So, there's a reconfiguration, a
5:46rebalancing of what people do. And
5:48ultimately, it sometimes leads to, you
5:50know, creating more similar code, but
5:52the really exciting thing is when you're
5:54able to do new things that you never did
5:55before.
5:57>> What are the implications for wages for
5:59folks at that higher end who are using
6:02AI to augment what they do versus the
6:06replacement contact.
6:07>> So, we worked with our partner ADP, the
6:09world's largest payroll processor. Uh
6:11they gave us access to data so we could
6:13analyze it really carefully. And what we
6:15found was that there were some pretty
6:16noticeable effects on the employment
6:18side. Some occupations we had falling
6:20employment, some we had growing. That
6:22depended a lot based on the age and how
6:24exposed it was to AI. On the wage side,
6:27we did not see a big change in the
6:28wages.
6:29>> I wonder if one of the things causing
6:30the lagging effect that you you might be
6:33seeing as one of your hypotheses is many
6:35of the companies I'm working with today
6:36are just trying to figure out the new
6:38way they do performance management of
6:40humans first and foremost. So, what what
6:42do I really want to assess, right? Are
6:45there new behaviors? Are there new
6:46skills? And you know, some uh people are
6:49have been talking about for some time
6:51now, how do I do performance management
6:52of agents, too, to kind of go in a
6:54completely different direction. Yes.
6:56>> The companies that are most successful
6:57are the ones that are best at measuring
6:59things. This is really the frontier that
7:01that we're very much in the front of
7:03right now is getting better metrics of
7:06what's happening. You know, the old
7:07thing, you can't manage what you don't
7:09measure. And um we need a new set of
7:11metrics.
7:12>> So, I have to ask, what what are the
7:14metrics that [clears throat] you would
7:15prioritize?
7:16>> Well, so, you know, we want to boost
7:18productivity. And the way economists
7:19define productivity is, you know, it's
7:21output per input. And too many people,
7:23as we were just talking, focus only on
7:25the input part of it. So, you need to
7:26come be creative about coming up with
7:28better measures of the output side. And
7:30that includes uh better quality, uh
7:33better customer service, new products.
7:36Uh actually less employee turnover can
7:38can be part of it as well. I think most
7:40companies underestimate how much
7:42administrative data they already have.
7:44Ultimately, as you instrument the
7:46company more, whether it's a
7:47manufacturing company, sales, um you
7:50start being able to have all sorts of
7:51fine-grained metrics. And you can see
7:53how they move as a company rolls out the
7:55technology. A related part of it is is
7:58getting causal inference from that. Is
8:00you don't want to just have the
8:01correlations between these things. You
8:03have to really see whether they're
8:04causally related. And that means using
8:07different techniques like differences in
8:08differences and instrumental variables
8:11that economists have been using. There's
8:12been a credibility revolution and one of
8:15my missions is to bring that to business
8:17and have that same kind of credibility
8:19in the business data.
8:20>> How have you seen kind of the ask that
8:23you have of clients evolve over kind of
8:25the past three or four years?
8:27>> What we're seeing is we're using the
8:28client's own data and doing a little bit
8:29more of a bottom-up assessment. Every
8:32company I've worked with has the same
8:33pattern, which is this sort of this
8:35power law distribution where there's a
8:37few people that are performing at 10x,
8:40you know, 20x. If you can capture what
8:42those best people are doing in each of
8:44the different occupations and kind of
8:46codify it and replicate it and share it
8:49with the rest of the organization, it's
8:50a really easy win to level people up.
8:53>> The point about that, you know, the
8:55mechanism for learning and and how to do
8:58that at scale. My observation is that is
9:00a huge need right now because I think
9:02many organizations,
9:04you know, especially leading ones that
9:06said, "We're going to really unleash AI.
9:08We're going to let a thousand flowers
9:09bloom." And now they're saying, "Okay,
9:11that was great. There was also a
9:13cultural byproduct of people kind of did
9:16some upskilling around AI, but now how
9:17do we move towards more of a
9:18value-backed, maybe top-down agenda on
9:21it?" And so they don't want to throw
9:23away all of that, but there has to be a
9:25more continuous way to scale.
9:27>> That's such an important point is is it
9:29while it's great to capture this
9:30knowledge, you also want to direct it.
9:32You've got to have I think it's from
9:33both sides. You've got to have the
9:34bottom-up and the top-down direction. So
9:36you understand what's possible, but also
9:39make sure it it's directed at some
9:41useful outcome and not just playing
9:43around.
9:44>> So Eric, you're talking about
9:45accelerating um the J-curve trajectory
9:49progress on that J-curve. Identification
9:52obviously would be a huge accelerant in
9:54that context. And you did some
9:57interesting research on the costs of
10:00agents and the surprise that um
10:03humans are really bad at estimating how
10:07token-intensive and how costly certain
10:10tasks can be relative to others. Do you
10:12want to talk a little bit about that?
10:13>> there's a big disconnect there. You
10:14know, just really just a few months ago
10:17people were token maxing and
10:19they weren't worried about like the
10:21token cost and now it's getting to be
10:23you know, a first order thing. Companies
10:25are spending you know, tens of millions
10:26of dollars on tokens. The problem is
10:29that you don't know how much
10:31token cost is going to go into a task.
10:34And the agents themselves are very bad
10:36at estimating them. And so this is a a a
10:38big gap now because you know, how can
10:40you start a project and you don't know
10:41what it's going to cost or you get
10:43halfway into it and it doesn't finish.
10:46And so there's there's an agenda there
10:47for doing more work on understanding
10:50how much you're going to have to spend
10:52you know, on humans, on tokens, etc.
10:54going first.
10:55>> You and many others,
10:57McKinsey included, have called this the
10:59most consequential
11:03transformation since the Industrial
11:05Revolution. Do you think we should be
11:08doing more to prepare for this kind of
11:11consequential change?
11:14>> [laughter]
11:14>> We absolutely know.
11:15>> ARE WE READY TO take stock?
11:17>> We are not ready. We you know,
11:19um
11:20it's I'm impressed by these improvements
11:22in capabilities and how rapidly things
11:24are getting better especially around
11:26here in Silicon Valley. People are
11:27spending literally hundreds of billions
11:29of dollars pushing those capabilities
11:31even faster.
11:32At the same time our understanding of
11:35what's going to happen to you know,
11:36employment, productivity, wealth,
11:39income, wages,
11:41you know,
11:42centralization of power or
11:44decentralization, even meaning, we don't
11:47have a good grip on that at all. There's
11:49very little being invested in it and
11:50that I think is the biggest challenge
11:52for the next 10 10 years or so is
11:55figuring out and how to close that gap
11:56between these capabilities and our our
11:59economic understanding. I worry a lot
12:01that we're going to a very disruptive
12:03period. I mentioned that I'm optimistic
12:05about productivity, but I'm worried
12:08about inequality and centralization of
12:10power and the disruption that's likely
12:13to happen even if we have a bigger pie,
12:15it could be very disruptive for I in
12:17fact it probably will be very disruptive
12:19for a lot of people. And uh and we're
12:22not ready for that.
12:23>> And it's also one of the things I'm
12:24worried about cuz when I kind of
12:26aggregate across the client
12:27conversations I'm having, when you hear
12:29people talk about augmenting, talking
12:32about the front of office tasks. These
12:34are people in sales or innovate and
12:36creating the products. And when we're
12:37thinking about automating, it's the back
12:40office. And then if you start to layer
12:42on, well who's in the front office,
12:43who's in the back office, it no longer
12:45looks like, hey now all of the sudden
12:48this wave of automation is coming for
12:51people with master's degrees or you know
12:54advanced capabilities, it looks like
12:55it's a different disruption pattern.
12:58What are you seeing in the research and
13:00>> It seems like it's cutting across lots
13:02of different groups now. I was you know
13:04stunned when a student came to my office
13:07a month ago.
13:08She was a graduating senior from
13:09Stanford and she said, you know she
13:11didn't have a job, her friends didn't
13:13have jobs. She said quote, is my
13:15generation doomed? And I said, oh my
13:17god, that's really hard. Um
13:19and and I see that kind of angst. People
13:23are really worried.
13:24What I tried to give her some hope
13:26because I think that yes, there's lots
13:28of jobs being destroyed. I don't want to
13:30sugarcoat that. There's also all this
13:32opportunity being created. You have to
13:34have much more agency and aggressiveness
13:37to lean into that. It's not that
13:39someone's going to tell you what your
13:40job is, it's that you have to be the
13:42creator. At the last my final class for
13:44my Stanford students, I told them that
13:46you know when you hear the words when
13:48you hear AI, you should not think
13:51artificial intelligence, you should
13:53think amplifying intention. Because
13:55that's what it does is if you have some
13:57intention, this is going to allow you to
13:59do a lot more than you ever could have
14:00before. I understand that's not what the
14:02conversation is about, but I think we
14:04need to change the conversation and
14:06remind people that this really could
14:08amplify whatever intention they have.
14:09>> Yeah. I actually want to loop back to,
14:12you know, understanding how to prepare
14:14better for an AI future. What are your
14:17hypotheses on what could be a potential
14:20solution or set of solutions here? Is it
14:22solving a collective action problem? Is
14:24it an incentive problem?
14:26>> Every company's going to have to
14:27reinvent invent itself. Um part of it is
14:29a collective action problem like you
14:31defined. Um what we're seeing is that,
14:34you know, the pyramid in lots of
14:36companies can McKinsey and and and
14:37universities elsewhere where people come
14:39in and they kind of work their way up is
14:41becoming more of a diamond. And that
14:44base isn't there anymore. And you know,
14:46that saves them cost in the short run,
14:48but that means where those middle
14:50managers, middle skill people going to
14:52come from and where the senior people
14:54going to come from if they're not people
14:55in the entry. And that's a real
14:57challenge. Part of it I think you have
14:59to have public investment in education
15:01and supporting training. Part of it I
15:03think enlightened companies are leaning
15:05into it. And AI itself can be a great
15:07teacher. Um you can get a lot of
15:09personalized education. You were talking
15:11about that earlier in terms of the mass
15:12customization. Um so there's an
15:14opportunity there.
15:15>> You did find in your research that
15:19early career professionals are being
15:22disproportionately affected. Do you want
15:23to say a little bit about that?
15:24>> did find that. The groups that were
15:25shrinking employment were the early
15:28career uh entry level workers age 22 to
15:3026 in the most exposed occupations. You
15:34[clears throat] could rank all of the
15:35occupations about 700 of them uh by how
15:39exposed they were. And if you take the
15:40top, you know, 150 most exposed
15:42occupations, the top quintile, um what
15:45you found is that the young workers in
15:47those categories had about a 13%
15:50fall in employment when we first did the
15:52study. Um we've been collecting every
15:55month since then. It's up to 16 or 17%
15:57now. The effect has just been kind of
15:59growing over time. And so that's why
16:01when that student told me about her
16:03troubles, you know, I I I I didn't, you
16:05know, dismiss it. I I saw that in the in
16:07the data.
16:08>> Do you think the trend will expand to
16:11more senior workers? In the sense that
16:15it might only be a matter of time that
16:19more senior
16:20employees who could also be susceptible
16:23to these kinds of
16:24>> Well, that's a big concern and that is a
16:26possible future. Um, but another
16:28possible future is that we lean in and
16:30and show them how to use these
16:32technologies to augment what they're
16:34doing, discover new products and
16:35services, expand the market. But one of
16:38the things I really stress in talking to
16:40the senior executives and for that
16:41matter the policy makers and economists
16:43is, you know, this is a design problem.
16:46Let's figure out the right incentives,
16:47the right structures, the right
16:48intangible investments, so we come out
16:51on the winning side of that. Um, we
16:53should not be passive.
16:54>> Mhm. Yeah. I think I think a lot of
16:57organizations though, I completely
16:58agree, are are a little stuck there and
17:01they want to know immediately how should
17:03the pyramid evolve? And the it's a it's
17:05a choice, just like any operating model
17:07or organizational design. There is no
17:09one right answer. So, thinking through
17:12how is value created in the organization
17:14and different parts of the organization,
17:16what would that mean or imply for what
17:18humans do versus what agents do? Either
17:21one can work and then only then once you
17:23have a view of that, can you say, "Okay,
17:25here's the talent system to support it."
17:27Right? But that's a process.
17:29>> Yeah.
17:29>> Um, and I I think many organizations are
17:32moving straight to the efficiency part
17:34of this versus
17:35>> I know I That's kind of like going
17:36straight to the efficiency part. I I
17:38honestly I think it's a little lazy, you
17:39know? It's it's the easy way to is to
17:41cut costs, it's not it's not hard to
17:43measure. But I don't think it's as
17:44sustainable as if you find new sources
17:46of value and you invest in your
17:48workforce and you're going to get more
17:49resistance from your workforce if, you
17:51know, you you tell them this is a tool
17:52for cutting heads. Um, and so I I I I
17:55really think there's a much bigger
17:57opportunity and I know McKinsey is
17:58leaning into this to try to find ways of
18:01creating new sources of value. It's what
18:04happened with the earlier technologies,
18:06you know, uh
18:07all through history, wages and
18:09employment grew even as new technologies
18:12came along and we want to keep that
18:14record going.
18:15>> But when we're we're talking about
18:16design, what I hear you say is we've got
18:17to be thoughtful in the design of
18:19multiple different levels. There's a
18:20societal level. How are we investing in
18:22the education and the training? How are
18:24we creating the right signaling
18:25mechanisms so that the skills you gained
18:27in one place are particularly relevant,
18:29another employer can pick them up. But
18:30there's also the
18:32company level design.
18:33>> Yes.
18:33>> When you're reinventing, how are you
18:36thinking about to use an analogy, you
18:38know, you you've talked about before,
18:40you know, electricity is another, you
18:41know, general purpose technology. It's
18:43not about just about putting electric
18:46lights in the factory. It's about
18:48actually electrifying the entire
18:50factory, which requires you to rethink
18:52the entire way the factory works powered
18:54by electricity. But that's a, you know,
18:56workflow level design. But it requires
18:58both the societal level and, you know,
19:00the company and workflow level design
19:02for it to work.
19:03>> Yeah, we've got work to do at all those
19:04different levels. At the
19:06level of specific companies, I think you
19:09need to understand the tasks so that you
19:11can then do the restructuring. Those two
19:13kind of go together.
19:15I think of the task as the atomic unit.
19:17Once you understand where AI can affect
19:19each of those, then you're in a better
19:21position to do the reinvention that
19:23needs to be done.
19:24>> Yeah.
19:24>> And I I think there's a skill gap there.
19:26I think there's I think companies are
19:27under pressure in some ways to use AI to
19:31drive efficiencies, but I also think
19:33reimagining a workflow or an enterprise
19:35is not something that most leaders know
19:38how to do.
19:39>> Right.
19:40>> And And so one of the things that I'm
19:42working with a lot of organizations on
19:43is creating a playbook for doing that
19:46and doing that consistently, right?
19:47What's that methodology? I don't I don't
19:49think we we actually really
19:51>> No, it's an it's an inherently harder
19:54problem. It just takes more creativity.
19:56You you can look at what already exists
19:57and say how we going to automate that.
20:00You know, that doesn't take a ton of
20:00creativity. But to imagine something
20:02new, that's harder. But my view is
20:04actually it's more risky to not do it.
20:08That if you just focus on what you're
20:10doing and try to, you know, hang on to
20:11things, that's actually the riskier
20:13thing. That ultimately no company, no
20:16country, no person has ever succeeded by
20:18just focusing on the same thing over
20:20time. You need to expand your horizons.
20:23and that's what helped make America
20:25successful and I think the successful
20:26companies are the ones that do that as
20:28well.
20:28>> I think one of the areas that's exciting
20:31for AI is
20:32small businesses.
20:33>> Mhm.
20:34>> Because now they're able to access, you
20:36know, talent and capabilities, you know,
20:39that they wouldn't have been able to
20:40access before or to access them may have
20:42had to, you know, sell their physician's
20:45office to a private equity practice in
20:48order to have the back office taken care
20:49of, but now hey, there's AI tools that
20:51can help them, you know, keep their
20:52business and grow their business while
20:54it's still their business.
20:55>> It's exactly what we talked about. I
20:57mean, my my masterclass, you know,
20:58you've all heard about the one-person
20:59unicorn and maybe that's kind of a
21:01little bit of an exaggeration, one
21:02person with a billion-dollar business,
21:04but it's not that much of an
21:05exaggeration to have a small group of
21:07people who now can use AI to leverage so
21:10much.
21:11>> On this topic of small businesses, I'm
21:12just thinking Brooke and Brian,
21:15do you see companies or Eric, do you see
21:18companies actually successfully building
21:21AI-native businesses within their larger
21:23organizations to innovate and test and
21:25then integrating them? Is that happening
21:27yet?
21:28>> One of my clients described as engine
21:29one, engine two. You know, engine one,
21:31the current business the way it's going,
21:33engine two, yeah, I'm going to invest in
21:35the disruption. So, this is going to be
21:37my new AI-native fill in the product or
21:40offering here and some others are
21:42saying, hey, actually I'm uncomfortable
21:44with that because what does it imply for
21:45people left in engine one as we're
21:47gearing up engine two? Does everybody
21:49know that that they are in engine one
21:51and engine one's on the wrong track? You
21:53know, so do we need to actually invest
21:55behind the lead innovators that we have
21:58today across the business and then scale
22:00that up. And I think the
22:03you know, the jury's a bit out among the
22:04clients of which ways
22:06are we able to do it because the engine
22:08two gets a lot
22:09more tangible output faster,
22:12but is that going to be the ultimate
22:14answer that, you know, gets you to
22:16scale? I think that's the question that
22:18that, you know, our clients are
22:19wrestling with.
22:20>> Yeah, I I do know one large um, global
22:24banking client actually in there they,
22:27you know, kind of took this experimental
22:29approach and some of those experiments
22:30became famous within the organization
22:33for the problems that they were solving.
22:35And so they actually have one that is
22:37kind of an internal go and see for how
22:39to do AI transformation well and how to
22:42use it as a force for augmentation. And
22:44they're of course using that as an input
22:47into this learning scaling model that
22:50we've been working with them on. So I
22:52think it's exciting. I think that's a
22:53kind of a more nascent approach or in
22:56this case it that model is is nascent.
22:59>> We've been talking a lot about
23:00augmentation about ways to push novel
23:03thinking etc. in order to continue to
23:05drive acceleration
23:07in uh, productivity and also in um,
23:11to keep humans engaged and employed and
23:15so forth.
23:16There's also a lot of discussion about
23:19um, information collapse and the
23:20potential for information collapse if AI
23:23becomes
23:25too much ballast for humans in the
23:29workplace. How do you assess that risk?
23:32>> Well, there is this concern that AI is
23:35generating more and more of the content
23:36on the web and more and more content in
23:38books and more and more content
23:40everywhere else. I mean, we're all
23:41humans. I can vouch for that. I'm pretty
23:43sure.
23:44>> Check.
23:45>> One of the things that we need to do is
23:46make sure we have incentives for content
23:48creators, human creators to continue to
23:50create the value so the LLMs can keep
23:53continue to become better and better.
23:55And right now existing copyright law
23:57doesn't really necessarily reward the
23:59content creators. Um,
24:01um, and you want to strike a balance. If
24:04you give the content creators too much
24:06incentives, actually then the downstream
24:08people don't um, have incentives to use
24:11it as effectively. On the other hand, if
24:13you don't give them enough, the content
24:14isn't created. And so I think going
24:16forward we need to sort of rebalance the
24:18way we reward creation versus use of
24:22different kinds of knowledge in order to
24:23have kind of a thriving ecosystem.
24:25>> We're we're seeing that very live
24:27happening um, at a firm level in
24:30professional services organizations. You
24:32know, okay, you're the expert, your
24:34expertise is what generates the client
24:37demand and the move.
24:39If you're not appropriately incentivized
24:42to do that, to put your information into
24:43the firm systems and do it, it kind of
24:45collapses. So, we're seeing a lot of
24:48client interest in, okay, how do I set
24:50up a knowledge management system to make
24:51sure that I'm doing
24:52>> clients because I've I've also seen the
24:54opposite problem where the opposite
24:56approach and it's it's really
24:58destructive. You know, in our call
24:59center paper, what we found was that the
25:02less skilled workers got the biggest
25:03boost from the LLMs and they were not
25:05performing almost as well as the most
25:07skilled workers. And some of the folks
25:09who talked to me
25:11using these call centers, oh, that's
25:12great. Now we don't have to hire as many
25:14of those
25:15most skilled workers anymore because the
25:16less skilled workers are doing almost as
25:18well.
25:19And that's a very short-sighted approach
25:20because exactly as you're saying, where
25:22does that knowledge come from in the
25:23first place? It came from the most
25:25skilled workers. In some ways, they're
25:26even more valuable now because they're
25:28not only answering the question for
25:30their own client, they're answering a
25:32question that then gets replicated
25:33throughout the organization. So, you
25:36want to make sure you invest in those
25:37highly skilled workers that are creating
25:39the really valuable content. That's how
25:41you upskill the rest of the workforce.
25:43But you need a company that sort of is
25:45sufficiently
25:46forward-looking that they understand
25:48that and don't just reward them on what
25:49they're doing today, but also how their
25:52knowledge creates value in the future.
25:54>> Earlier in the conversation,
25:56we talked about the CFO client of yours
25:59who reflexively turned to head count
26:01reduction cost cutting.
26:02>> Right.
26:03>> What kind of capabilities do you think
26:06leaders need to build? As Brooke said,
26:10it's a it's an entirely new set of
26:12capabilities that leaders need to build
26:14to re-envision processes end to end and
26:17reap the benefits of these intangibles.
26:21Are you seeing leaders do that
26:22successfully and what kind of
26:24capabilities do would you cite that they
26:26would need to?
26:27>> It's really hard. There aren't that many
26:29that are doing it successfully. I mean,
26:31we've been working with Nasdaq and
26:32that's one of the companies that's been
26:33doing a great job of this.
26:35They've got some great leadership there
26:38that is
26:39for a long time has been reinventing
26:40organizations even before LLM's came
26:42along. But now they're continuing that
26:44track record.
26:45But most companies are are struggling
26:47with it. It's understandable. I think
26:49people
26:51underestimate how difficult that is.
26:53>> So you mentioned agency,
26:56entrepreneurship. I go to you know,
26:58thinking boldly cuz many of these
27:00workflows are cross-disciplinary. And so
27:02in my experience getting a group of
27:04cross-disciplinary people in a room,
27:06first of all, everyone kind of is a
27:07little defensive. Well, my part of the
27:09workflow is working great, right? So we
27:10don't need to reinvent that and there's
27:12a there's a lot of you know, like
27:15learning in those spaces that require
27:17openness to actually get to the other to
27:20the truly reimagined end of that. And I
27:22think leaders that can set the
27:24conditions for that in a way that allows
27:27for that thinking in a bold, challenging
27:30way. Anyway, those are the qualities
27:32that stand out to me.
27:33>> It's really hard, you know, and and CEOs
27:35[clears throat] have to step up and play
27:36a role because
27:37a lot of the rest of management is just
27:40not
27:41selected for having that kind of big
27:44transition.
27:45They've got something that's working and
27:46they want to preserve that and so it
27:49really needs a jolt to the system. And
27:50you also need the openness that you
27:52talked about that hey, you can do this,
27:53it's going to be okay.
27:55We don't expect it to go totally
27:56smoothly and let's be bold.
27:58>> Yeah. Actually, I was with a group of it
28:00was
28:01you know, business executives including
28:03CEOs and then also some military leaders
28:05and we were discussing you know, some of
28:07these behaviors that are required for
28:08change. And the [clears throat] one that
28:10that or it's it's kind of a behavior but
28:12kind of a role, looking for the
28:14mavericks in your organization. And
28:16those can exist at multiple levels. But
28:18a maverick is someone who is going to
28:20challenge, who is going to be bold, who
28:21isn't going to be defensive. And you
28:23know, a question that might be worth
28:25asking is you know, to a CEO or or a
28:28CHRO is do you have a good sense of who
28:30your mavericks are? Right? And are you
28:32strategically allocating them to drive
28:35this transformation?
28:36>> I like that. This is a time for
28:37Mavericks.
28:38>> Yeah.
28:39Do you want to answer the $64,000
28:41question that is plaguing parents from
28:44shore to shore?
28:46What are the skills that kids will need
28:49in the future? Kids who are in college
28:51now.
28:51>> That's that's a really tough question.
28:52My framework is is less to go into
28:54specific skills and more into this
28:56broader framework. And it gets back to
28:58what we said earlier, having that
28:59intention, figuring out what is it that
29:01you really want to do. I think that's a
29:03skill that can be engendered, even, you
29:05know, taught. Um, and having that kind
29:08of openness to let's be the person who
29:11directs the project. Let's be the person
29:12who comes up with new ideas. If you can
29:15have kids who have that kind of an
29:16attitude, they're going to be much more
29:18successful in the future. Because I
29:20think almost everyone is going to be
29:22managing not just an agent, but a fleet
29:24of agents. They'll be like the CEO of
29:26their own little entity there. And
29:29they'll have to have those leadership
29:30skills, leading agents to to direct
29:33them. And the ones who are good at
29:35pointing them in the right direction and
29:36then evaluating them are going to really
29:38thrive.
29:39>> So, from a policy standpoint, should we
29:41have more investment in things that
29:44enable entrepreneurs?
29:46>> I think so. Yeah, one of the things
29:48that, um, you know, we look at over the
29:49past couple of decades is that, you
29:52know, technology's advanced, but
29:54productivity hasn't grown all that much.
29:56And part of the reason I think have to
29:58do with less dynamism in the economy.
30:02Um, there you know, even though we see a
30:04lot of it here in Silicon Valley,
30:05there's actually, uh, fewer startups
30:07overall in the United States. And if we
30:09can get more entrepreneurship and more
30:11dynamism, then I think the technologies
30:14are going to have a bigger, uh,
30:15beneficial effect.
30:16>> Erik Brynjolfsson, thank you so much for
30:18joining us today. This is great.
30:20>> My pleasure.
30:21>> And Brooke and Ryan, great to [music]
30:22see you as always.
30:23>> Yeah, it's great to be here. Thank you.
30:25>> Thank you.
30:29>> [music]

About this transcript

This page contains the full transcript of How AI Is Changing The Career Ladder | Stanford Economist by McKinsey & Company, generated from the public captions YouTube serves with the video. The transcript has 5,977 words across 929 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.