0:14 

Azeem Azhar: Founder Exponential View

What percentage of people will be in jobs that currently don't exist?

0:19 

Joe: Global Chief Economist

There's three areas of the economy that we think would need to happen for AI to be truly transformative to this personal computer.

0:26 

And those areas would include areas that are of high cost, that have been tough to automate in the past and are fundamental or unmet human need.

0:35 

And three areas that jump in our scorecard would be clean and inexpensive energy.

0:41 

Second would be more preventative medicine in healthcare.

0:44 

And the third one would be along the dimension of education.

0:47 

If AI is going to be transformed, starting to think about those new effects, it's not just about saving time.

0:52 

What is the unlock?

0:54 

Well, we're just identifying statistically and we do it out a sample with electricity and personal computer.

1:00 

We're projecting that roughly five years out given the back tests we've done against every other technology since the Industrial Revolution.

1:08 

Arvind Krishna: Chairman and CEO IBM

Yeah, I think can I build on, Joe, the point, let's take an example that is real but is historical.

1:13 

In 19147% of the US population work directly in a farm.

1:19 

By 1950, that was 3%.

1:22 

What enabled it?

1:23 

It was automation because of machinery, which industries came about because now people at time, if you're working on a farm, it's seven days a week, It's not five days or three days or two days.

1:35 

It's not hybrid.

1:37 

So so that enabled malls, it enabled fast food, it enabled restaurants.

1:44 

It gave rise to what people used to make fun of in the 19th century as the leisure age.

1:50 

It enabled all that.

1:51 

Now, of course, you also needed the automobile because that's how you got to all these other places.

1:56 

But literally 20% of the economy around the whole food, entertainment and shopping industry came about because you freed people up from having to be at the farm.

2:06 

That is what Joe's talking about is really hard to predict.

2:09 

But every one of those 44% was not unemployed.

2:13 

Other occupations came up because of what was possible.

0:12 

What we are finding is despite some fears, AI will unlikely be dystopian, meaning massive job losses because the number of tasks that are needed to fully automate a job, roughly one in one in five job occupations in the United States, however will see significant displacement.

0:30 

What's more important is over 60% of of occupations will be augmented more by some automation and then some power tools.

0:38 

Think of the personal computer and how you're able to get better insights.

0:41 

And it's everything in medicine, education, finance, IT and there's roughly 20% of occupations that are not have meaningful impact.

0:50 

So when you add that all up, what that is telling us respectfully, is that there's there's greater than a 70% probability that every economic forecast out of central banks is incorrect because productivity is going to rise to a level.

1:04 

And I'm not just saying this is a personal view, this is a quantitative estimate of what it will do.

1:08 

And The funny thing is, is that actually United States, we need automation.

1:13 

The biggest reason why growth has on average been low since the global financial crisis is not debt and not demographics.

1:19 

It is the biggest drag is the lack of automation in a service based economy, the biggest drag in over 100 years.

1:26 

And so if AI is transformational, which means has to continue to advance roughly threefold.

1:32 

So if that happens, the dividend from automation will be as if the baby boom generation never retired from the workforce.

Will we Innovate Faster?   

JOY ROBINS: Welcome to DealBook. My name is Joy Robins. I'm the Global Chief Business Officer here at The New York Times. One technology we can all agree that is pervasive and that we're all really contending with right now is the power of AI and the potential of AI. And today, we'll hear from Vanguard, one of the world's leading investment firms and most respected investment companies with a legacy of thought leadership on how AI will shape the global economy and the financial markets.  

AZEEM AZHAR:   Good morning, everyone.  I’m Azeem Azhar.  I’m an early-stage investor in AI companies. And I'm really delighted to have on this panel Joe Davis, who's the Chief Economist and Head of the Investment Strategy Group at Vanguard, and Arvind Krishna, the Chairman and Chief Executive Officer of IBM.  

We are two years on from ChatGPT, and ChatGPT has gifted trillions of dollars in market cap to the big tech companies in the past couple of years. It's created this seemingly unending demand for computer chips and now electricity. I read that enterprise spending on the technologies that underpin ChatGPT, foundation models, should reach $13 billion this year, a market that was almost certainly in the low single millions just two or three years ago.  

When you get a general-purpose technology, think of electricity, it enables lots of other industries: the modern automotive industry, which is dependent on assembly lines, of course, the entertainment industry, pretty much every part of our modern world. And they can also strand businesses, leaving successful businesses with business models that are not viable. So, if we have this mega-trend, Joe, I'd love to come to you first, how does your research illuminate whether it's real and how it might play out?  

JOE DAVIS: Well, I think the more important question is we're trying to quantify the evolution of technology. And as you mentioned, Azeem, we think of it, there's three forms of technology in terms of it impacts both our work and our daily life. So, there's innovation, and there's two components of that. Think of the personal computer and IBM. It augments the work that we do. It also can automate certain tasks.  

Then there's that additional component, which is what I would call transformation that changes how we live and there's knock-on effects. Some technologies do one, but they don't do significantly the other, and only technologies that do all three rise to the level of general purpose technology.  

What I'm very proud of is we spent over two years quantifying, going back over 150 years, capturing the evolution of technology. And we believe we can actually predict it out of sample, like, several years out. And so we see dynamics in terms of the workforce and job descriptions and labor, and we see it in terms of capital the investment companies make. And if you're in VC, you know the phrase J-curve, which is you're investing in labor and capital, yet the return on investment is low at the time. And I think the only reason why business leaders do that is because they see a future economic payoff.  

And so, our conclusion is fairly eye opening. And that is, the consensus view in the markets and from many central banks of economic growth is almost has a very low probability of happening.  

AZEEM AZHAR: Arvind, from the perspective of IBM, do you see this as a distinct megatrend? Is this different to the many, many technology innovations we've seen in digital over the past 20 years?   

ARVIND KRISHNA: Yeah. So I'll go back to the thought that both Joe and Azeem, you expressed. I think if you step back, artificial intelligence is going to be one of the, I'll say the eighth technology, and that is going back over 300 years, that has fundamentally transformed.  

What happens is, you first do get the productivity. And I'll quantify the productivity. We see that almost any enterprise can become 1 0% more productive without taking, I'll call it, substantial risk.  I'll take an example from ourselves. We spend about 15%, I would call it, on all these things operations, back office, those functions. I will easily see 30% of it come out completely over five years. Not in one year, but in five years. By the way, half of that, we already have out. Then you are going to get new industries coming up.  Because the cost of doing something will go back to being 1% of what it might be. So, it's always the same three: productivity, I think it's a given, you're going to get growth for those companies because productivity means you have more capital and more dollars to go reinvest in getting more market share, and then three, it will transform economies.   

A point that got hidden in all this, how much of the current US economy, the fact that it has the highest growth, the fact that it has more companies of high market cap is because of a fact that the US embraced the internet first.  Hence as a consequence almost all internet companies are based in the United States there is a correlation to getting a nexxus of that capital and that growth in this place.  And AI, I think, is going to be at that same scale.   

AZEEM AZHAR: What percentage of people will be in jobs that currently don't exist?  

JOE DAVIS: what we don't know in our system is what the new industries will be. But what it can tell us, is that there is a blip on the radar screen that it will happen with odds and probability. So, then it's going to be, what areas could it be to unlock what they call the knock-on effects, right?   

We don't have the entertainment industry without electricity, air conditioning, lights. Actually, think of this theater. We don't have the entertainment industry without electricity running through it. So, there's three areas of the economy that we think would need to happen for AI to be truly transformative since the personal computer.  

And those areas would include areas that are of high cost, that have been tough to automate in the past and are fundamental or unmet human need. And three areas that jump in our scorecard would be clean and inexpensive energy, second would be more preventative medicine in healthcare, and the third one would be along the dimension of education.  

I just don't know what that blip on the radar screen actually manifests itself in a new company, new business model. We're projecting that roughly five years out, given the back tests we've done against every other technology since the Industrial Revolution. 

ARVIND KRISHNA: Can I build on Joe's point?  

AZEEM AZHAR: Yeah.  

JOE DAVIS:   Yup, yup. 

ARVIND KRISHNA: So, let's take an example that is real, but it's historical. In 1900, 47% of the US population worked directly on a farm. By 1950, that was 3%. What enabled it? It was automation because of machinery. Which industries came about? Because now people had time. If you're working on a farm, it's seven days a week. It's not five days or three days or two days. It's not hybrid.  

ARVIND KRISHNA: So that enabled malls. It enabled fast food. It enabled restaurants. It gave rise to what people used to make fun of in the 19th century as the leisure age. It enabled all that. Now, of course, you also needed the automobile, because that's how you got to all these other places. But literally 20% of the economy around the whole food, entertainment, and industry came about because you freed people up from having to be at the farm. That is what Joe is talking about is really hard to predict. But every one of those 44% was not unemployed. Other occupations came up because of what was possible.  

JOE DAVIS: What we are finding, however, is despite some fears, AI will unlikely be dystopian, meaning massive job losses because the number of tasks that are needed to fully automate a job. Roughly one in five job occupations in the United States, however, will see significant displacement. 

What's more important is over 60% of occupations will be augmented more by some automation and then some power tools. Think of the personal computer and how you're able to get better insights. And it’s everything in medicine, education, finance, IT. And there's roughly 20% of occupations that are not have meaningful impact. So, when you add that all up, what that is telling us respectfully is that there's greater than a 70% probability that every economic forecast out of central banks is incorrect. Because productivity is going to rise to a level, and I'm not just saying this as a personal view, this is a quantitative estimate of what it will do. And actually, the United States, we need automation. 

The biggest reason why growth has on average been low since the global financial crisis is not debt and not demographics. It is the biggest drag is the lack of automation in a service-based economy, the biggest drag in over 100 years. And so, if AI is transformational, the dividend from automation will be as if the baby boom generation never retired from the workforce.  

AZEEM AZHAR: So, it, it, seems like what's almost required at an executive level is a degree of imagination and a degree of experimentation, which perhaps exec’s have not been rewarded for over the past 20 or 30 years. So, please… 

ARVIND KRISHNA: Look, I think that is the biggest unlock for enterprises. We have almost tried to take risk out of how you manage a process and how you manage an enterprise. It is the surest way to go fail. Now, you don't want to take risk that destroys the enterprise. But how do you perhaps take 2% to 3% of your internal budgets and reward people? Go try it out. 

Fully expecting that over half will fail when they go try it, but the half will succeed. Then you double down and scale it completely. And let the half who fail kill it quickly but let them try something else. So how do you start rewarding people for experimentation is going to be a massive unlock.  

You mentioned HR. Our HR leader stepped up about three years ago and said, let me see if AI can help us do it. Half the functions, no chance. Advising people on how the team should be structured, advising people on what are people's gaps. That portion of HR, AI is not being able to help today at all, I would say.  

The other half of HR is, I call it, transactional. Promote somebody, change an address, provide a letter. All of that can be automated almost completely. So, then you reward people. And by the way, her team of doing this is 6 people. It's not like it's a massive expense. But they try it out. They see if there's traction with the employees. If there is, they double down and go get it done and then move on. I think that is the experimentation that we have to do, but it does mean taking some risk because some are going to fall flat on their face.  

AZEEM AZHAR: So, do you think that there's some way of shifting investor expectations of what good performance from a firm should look like in this environment, Joe?  

JOE DAVIS: In terms of thinking about the ROI, the one characteristic without exception since the Industrial Revolution, is that there is some cost and the payoffs are not immediate. I think that's why also there's the risk of underestimating the S curve, right? It's the wonderful thing of, you're disappointed in the short run but then actually surprised on the upside. It's a nonlinear process, and I think that's part of it. So, answering your question, I think it's just having the, and I think it takes courage. We just know that the great companies, one characteristic in them is having the fortitude to have a longer horizon, and then to deploy capital.  

ARVIND KRISHNA: I actually think investors are far more willing to let people take risks. So, if you say, I am going to spend this, you quantify it, to your point, was 1% to 2% for a year or two, and then you're going to see higher returns and cash flow and returns, but it's going to be two years out or three years out. They will absolutely have that patience.  

Now, if you say it's 10 years out, that's a little bit different. But I'm not going to go spend that amount of money unless I see a return. So, it's not as simple as you take half of 3% and it's wasted. But on the 1 and 1/2% that works, I'm getting 10 times return. So that's actually right there, and that's the 10% overall that you can get to from those, because those little pieces then accumulate up. And within two years, you're getting far more back than you're actually expending on the experimentation.  

AZEEM AZHAR: Two or three years ago, these large language models could hold about 100 words in their working memory. Today, that number is up to a million and a half or 10 million. The cost to access these models has dropped by a factor of 250 in about 18 months. That could lead you to say, well, I'll just wait a little bit longer till it's a little bit cheaper and somewhat better and less hard to use before I make my decision. And so there are a lot of confounding issues around something that is as dynamic as this. How do you think firms should approach this? Joe, we'll start with you and go to Arvind and then come to questions.  

JOE DAVIS: And I can provide perhaps the macro sort of backdrop. I think, there’s advantages of learning by doing every technology on this planet for the past 500 years. There's been a critical element of learning by doing, and it's that reinforcement loop that leads to unexpected developments and so forth.  

And so, in fact, we've actually looked at where great ideas come from. You can think of the iPhone. We looked at the Wright brother’s plane. We looked at all patent records, and we actually are able to trace every idea in its evolution, 6 billion ideas over the past 25 years. This is the amount of data we've brought to bear.  

And the one common thread through all those rates of innovation is the unexpected discoveries it's the learning by doing mode. And two smart people in two different fields say, I actually never thought of combining this with that, peanut butter and jelly. The risk I think is actually not doing the experimentation because of the learning that employees themselves, which can lead to an insight down the line.  

ARVIND KRISHNA: Look, it's that-- I'll pick up on the learning by doing. That two to three years.  So, suppose you decide to wait. You could wait two years. Now you've lost two years of your people getting to know the technology, people getting to know what are the risks and limitations, people figuring out, will this or will this not apply? That two years in modern business, can kill a company. Because if your competitors are that far ahead and they're waiting for that price point, it comes, and then they embrace it because they're completely ready, and they pick up that 10% productivity.  

You have two companies. Average bottom line in the S&P is about 10%. So somebody sitting at 20. That means they have twice the investment power of those who are sitting at 10, they're going to run away with the market at that point because they can do everything possible. And you can see this again and again and again. I almost invert it. Let's get ready so that we can take even more advantage of it two years out as opposed to wait for that price point to come down.  

JOE DAVIS: Even though we are projecting a 2 to 1 odds that AI is more transformational than the personal computer over the next 15 years, it's not foregone. 2 to 1, it's effectively 2/3 to 1/3 because AI still has to develop. Now, I bring that up because now if you look at the equity market today in the United States and its exceptional performance, and you calculate the same odds based upon the same algorithm, you get the US stock market assigning a 90% probability that that will happen.  

So even though Vanguard would be fairly constructive, Vanguard would be fairly constructive on AI's potential for economic growth, it is not as high as 90%. So, one of two things have to happen. Either our model will catch up to where the equity market is, or where the strong empirical evidence is that there's a disappointment at time, even though the technology ultimately transforms the economy. And that's the irony between investing in technology and what it does to the hopefully, that's helpful.  

AZEEM AZHAR: So how specifically should investors think about positioning their portfolios? 

JOE DAVIS: Well, what our research was, this was eye opening to me, because the intuition I had going into our whole research project was, if this technology is transformative, investors in the companies that do the transformation. And that is certainly true in the first phase of, say, a decade to two-decade long cycle.  

But what actually what history shows, and the empirical evidence supports is that when there's general purpose technology, again, it is broadening out. The very fact that the economy is being transformed means that hospitals are using it, and utility companies are using electricity, not just General Electric. What that means is the profitability of the companies outside the technology sphere are actually their profitability is going up as the economy is being transformed. So, the lawn play is actually overweighting value-based companies in a period of growth, whereas it's overweighting growth companies in a period where growth is at actually is at a premium because growth is low.  

So, here's the catch. If you can identify companies better than the average investor, I'm at best average. So, 4% of the companies explain half of the S&P 500 climb over the past 100 years, 4% of the companies. So, if you want to index it and you want to deviate from the market, you would be going outside of technology. But if you have the foresight and the skill, the ability to say, know what, I think I know how technology is going to evolve. It's the IBM's of the world or so forth. Then that's how you would want to deploy capital.  

AZEEM AZHAR: Right. So, looking out over the next two years, do you think you are more likely to be surprised to the upside or the downside when you think about the impact of AI? We'll start with Arvind, come to Joe. Brief answers so I can sum up.  

ARVIND KRISHNA: In terms of deployment in the enterprise, because that's what we are. We're not a consumer company. We're an enterprise company. We're going to be surprised to the upside. I actually think 90%, I think I'll be more precise, 88% of what has been out there so far is in the B2C world. The enterprise world catches up. It's a bit slower. And it is going to surprise us massively to the upside.  

JOE DAVIS: I’d say on the automation side, we'll be surprised to the upside. The transformation, I think, will take more time, which is the knock-on effects.  

AZEEM AZHAR: Great. Thank you. Well, this has been a fantastic discussion. I'll try to summarize a couple of key points here that have emerged. Joe has made a strong case for AI as a general-purpose technology, as a technology that can be a significant boost to productivity, identifying the challenges to how it will reshape the labor market, but warning, cautioning us against any hugely dystopian futures.  

Arvind has described already IBM's incredible internal successes in improving its own operations through the use of one type of AI, but also the ongoing demand from enterprises and real businesses buying real technologies for real reasons. And I think that's a great summary of where we've got to and where we will come back to. And hopefully in a couple of years' time, we'll be able to sit down again and say just how surprised were we with how the next two years panned out. Thank you very much.  

Disclosure: 

All investing is subject to risk, including the possible loss of the money you invest. Past performance is not a guarantee of future returns. 

 © 2024 The Vanguard Group, Inc. All rights reserved. 

Joe Davis: Could AI be as transformative as electricity? This is a future that very few are actually talking about.

Could AI be as transformative as electricity? I mean, in one sense, AI uses electricity. But, no, could it rise to the same level? I have to be honest, I don't know the answer to that question. But, I can tell you this: if it is going to be, AI will have to do three things. Transformative technologies improve basic and unmet human needs—health, happiness, and progress.

Electricity was transformative because it helped us improve as a society along all three dimensions. And in addition, they created new industries and knock-on effects. They created new dimensions to the economy that saw improvements along health, happiness, and progress.

Let's take health. Electricity didn't increase health, per se, as we think of it today, but it actually evolved the health of society in some unappreciated ways. Street lighting reduced crime in cities. The electrification of hospitals improved the quality of care. And even the installation of fire alarms in cities, which were powered by electricity.

Just think of the leap forward in society over time by lengthening the time of education. Things as simple as being able to read at night under a light.

So, what will AI be able to do for health?

There's already glimpses of potentially AI's transformative ability. We're seeing scientists being able to detect new patterns by folding proteins in genetics, defining patterns in terms of drug discovery and other medical treatments.

What else could AI do for education? What else could it do for healthcare?

Happiness. Now electricity didn't improve—I don't want to say electricity improved human happiness, per se. In fact, I would argue human happiness is somewhat elusive. But, electricity did change how we spend our free time.

It helped power new forms of entertainment. It's called the movie. It helped give us the ability to talk to loved ones far, far away with the telephone, which was powered by electricity. Knock-on effects.

What could AI do for peace of mind in the years ahead?

And finally, progress. Progress. Transformative technologies create new industries and opportunities by creating knock-on effects.

By some estimation, more than half of the new job titles and occupations that evolve over time as society changes can actually trace their origins to transformative technologies.

How may AI transform, indeed, propel opportunities for progress in the years ahead?

© 2024 The Vanguard Group, Inc. All rights reserved.

Joe Davis: Could AI be as transformative as electricity? This is a future that very few are actually talking about.

This photograph was taken a little over 100 years ago this day. It's a photograph of a small mining town in Alaska. They had had the gold rush, and so there's about 300 people in the town. The town is Skagway, Alaska, about an hour's drive north of Anchorage.

Why, in many ways, this photograph is historical is because the men and women in that photograph, they're seeing an airplane with their own eyes for the first time. Yeah, they had read about a so-called "flying machine," perhaps in newspapers or they read about it in letters from loved ones. What an amazing experience that must have been to see an airplane for the first time.

It had not been that long ago that everyone on the planet believed human beings couldn't fly. Several years before this photograph was taken, factory workers in a relatively new factory in Buffalo, New York, assembled that airplane. But they didn't do it alone.

See, for years, they had been harnessing electricity as a new power source. They were trying to harness electricity to make that airplane at scale. It had been out for 20 years since the Wright brothers, but no one had really manufactured it at scale. Those technology workers had to change production methods. They had to change the "tech stack" to weld with power tools, which were now powered by electricity, to weld and stamp and hoist that airplane, which now weighed 3,000 pounds. No single human being can do that.

They had a job description change in the factory. If you actually look at the job descriptions, they actually change as tasks change at the factories. Power tools changed the nature of the workforce. And leaders and managers in that factory had to learn what we now call "organizational agility." How do you harness a new technology and integrate it to make things at scale? That's when it gets tough, right?

But innovation over the years at the edges. So much so, that months later, a new plane left the factory in Buffalo, flew to Chicago, it flew to Seattle, and finally over a small Alaskan mining town, changing the lives of those miners—of those men and women—forever.

The legacy of those technology workers back home in Buffalo—the legacy was innovation and transformation. They were the IT workers of the day, the most sophisticated building in new technology. And look at the legacy they had. With AI, that can be our legacy too.

Dystopian, no. Disruptive, yes, but not dystopian. That's the future I see. AI has the potential to be one of the most transformative technologies in over 100 years.

© 2024 The Vanguard Group, Inc. All rights reserved.

Joe Davis: Could AI be as transformative as electricity? This is a future that very few are actually talking about.

So, what effect may AI have on our work? My team and I developed a simple yet powerful framework years ago in [an] attempt to assess how the workforce, how our jobs may change as AI advances. Its ability to anticipate the future comes from two key insights, really just two.

The first one is that any job, each one of our own jobs, is the assembly and the compilation of hundreds of tasks and work activities. Just think of a job description. Is there just one line? No. There's usually multiple lines, multiple things we do in a job.

Now, jobs, per se, do not get automated away, but tasks do. And jobs can disappear if they're comprised of a small set of repetitive tasks.

And secondly, AI, like technologies of the past, will affect certain tasks differently than others.

I've categorized three types of how AI may affect our tasks and, hence, our work and our jobs in the years ahead. For some tasks, AI is believed—and I think we will show—can do a lot. It can do as effectively as a human being, almost equivalent. You can save [a] significant amount of time savings: 40% or more, let's say. That's a significant step in innovation.

I've labeled it dystopian not because that is dystopian, per se, but because there's a lot of smart people out there in the world, in tech, who fear that AI will be able to do most tasks of most jobs incredibly well; hence dystopian.

The second category is innovative. I think in many ways it's complementary. You can think of a power tool. It's the GitHub Copilot—like the humans in the loop, but it makes me more productive, faster. That's why I think of a power tool. If I'm in construction, think of that power gun.

And then finally, marginal. Really, AI, will it be marginal? Actually, that is the consensus view, outside of potentially the technology field. But if you ask most economists, by and large, their general assessment is that AI may be more marginal and, hence, we will not have the productivity lift that we have been hoping for, have been waiting for, for more than 15 years.

So, who's right around this debate? Well, it turns out we can harness AI itself and deep learning, ever the tools and techniques you know and work with set the pace for Vanguard for, it's really impressive. We can use those techniques to get a sense for every task in every occupation. There's over 50 million work activities occurring right now in the United States this very morning—50 million work activities and simulate them all using AI and deep learning to get a sense: How will AI shape our jobs for the entire workforce in the years ahead?

So, in many ways, we can use AI to forecast the future of AI. Isn't that cool? We can use AI to forecast the future of AI. Any Seinfeld fans here? It's like a coffee table book about coffee tables. It's pretty cool.

And when we apply our task framework across not just one job, but all 800 occupations, across the entire swath of the U.S. economy, this is the future of AI.

AI will be disruptive, but it won't be dystopian.

Yes, for some occupations, there will be significant automation. But across the majority of occupations, AI will be neither marginal nor dystopian. AI will be innovative.

If there is one number that you take away and think about the future, it's this: 4-to-1. Four out of five jobs will see significant time savings, changing their value add, which means earning more and propelling the economy. 4-to-1, that is the future that too few are talking about.