The economic implications of generative AI
Tune in to hear Dr. David Kelly and Stephanie Aliaga, Research Analyst and Michael Albrecht, Global Market Strategist on the Multi-Asset Solutions Team, discuss the implications AI may have on economic growth and the future investment environment.
David Kelly (00:00):
Welcome to Insights Now, a series of conversations designed to shine a light of clarity on the complex world of investing. This season is a special summer series entitled Market Movers as we'll focus on key themes that are the potential to move markets today and over the long term. If you're listening to the audio version of this podcast, you can watch along on YouTube by tuning into our JP Morgan Asset Management Channel.
(00:23):
Hello, I'm David Kelly, chief global strategist here at JP Morgan Asset Management, and welcome to Insights Now.
(00:30):
So let's get started. There are a lot of questions surrounding AI and its potential to change both the economy and society. So I'm here with Mike Albrecht and Stephanie Aliaga, who have both been doing a lot of research into what AI might mean for the workforce, for productivity and for economic growth. And you're actually putting this out in a forthcoming research publication.
(00:52):
So Mike is a strategist with our multi-asset solutions team here at JP Morgan Asset Management, and Stephanie's a research analyst in my team who spends a lot of time focusing on the US economy. She's also helped a ton with these podcasts. So thank you for that, Stephanie.
(01:05):
So it's great to have you both at this table here. So let me start with you, Mike. Why AI now? I mean, AI's been around for a while, so why is there so much excitement about AI right now?
Mike Albrecht (01:14):
Yeah, that's really the question of the day that there's a lot of excitement in the market right now, and really what sets us apart from the prior waves of AI excitement that we've had in the past is that the generative nature of AI. So it's really generative AI as it's known that people are excited about that. The best example of this, the most obvious one is ChatGPT, the product from OpenAI, that chatbot that you can ask any kind of question and it applies to you in a very natural language way. Now that's tech space, but there's also other technologies similar that work with images like DALL·E and can really generate images based on a text prompt.
(01:54):
Now, if you compare that to where we were several years ago, thinking back to 1997, I grew up playing chess, and I remember when Gary Kasparov got defeated by the IBM computer. That was a very primitive example of AI.
David Kelly (02:09):
Quaint old days. Yes.
Mike Albrecht (02:11):
But AI nonetheless, and we've come a long way since then there there's been progress on machine learning. So when I open up my phone, I can search bumblebee within Google Photos and I'll come up with a result of the pictures that I took a few weeks ago, and that's because Google has gone and tagged my picture with all the contents of what they understand is in that picture.
(02:37):
And so that's really cool, and a lot of resources goes into this. Every time you go on the internet and you have to verify you're human and very annoyingly, you have to pick the images with the fire hydrants in them. You're training a model, and that's very useful. The difference now with generative AI is that I can actually ask it to show me a photorealistic picture of a bumblebee wearing a hat. I can ask DALL·E to do that, or I can ask ChatGPT to write me a story about a bumblebee learning to fly.
(03:13):
So there's a lot of differences here, and maybe silly as it sounds, it's very important because you're not just writing a story. You could potentially write a movie script. And that's why the screen writers in America are on strike right now. To be sure there's only about 10,000 screenwriters in the US, but if you can write a movie script or at least help to do that, then you can probably write computer code and help software engineers debug their code. You can probably help lawyers draft contracts and write legal opinions.
(03:45):
So there's a lot that you can potentially do just in that text space. And then if you think about the visual components, you can imagine that the graphic designers, video editors, a lot of their work can be automated as well, ultimately. And so ultimately, when we imagine what generative AI can do across the whole economy, the impact can really be massive.
(04:08):
Sometimes it's actually easier to imagine what generative AI can't do, and the biggest category there is that it can't do physical things. It can't pick up a glass of water and give it to someone. But in our information economy, in our desk jobs, that there's a lot that we do that could potentially be automated. So the bottom line to your question, I think, is that what really sets generative AI apart is that there, there's so much that it can do that we do today, and ultimately, being able to automate that much can have a really big impact on our daily lives and the economy.
David Kelly (04:48):
So Stephanie, do you agree with that?
Stephanie Aliaga (04:50):
Yeah, and I mean, I think the other main reason for excitement is that this has been able to become suddenly so mainstream. And the reason that this technology has, the adoption speed has really been able to take off is because unlike the advent of the computer where first for everyone to use computers, you needed to build these expensive computer labs and teach everyone how to use a computer.
(05:12):
Now everyone already has computers, smartphones, they can just search up OpenAI on their mobile phone and begin using the technology right away. So that's how there's that statistic that ChatGPT broke records by reaching 1 million users into just five days. I mean, that's really the key here is that it's approachable to the average non-technical human. And then businesses are also catching on. There are a lot of exciting things that this can do for businesses and making their processes more efficient.
(05:42):
And so we've seen this wave of investment dollars flowing into AI in recent years. So just in the five years ending in 2021, you saw global business investment increase by six times on AI technologies. AI startups are also exploding, but it's not just startups, it's also the largest companies in the S&P 500. So we actually look through earnings transcripts of the first quarter earnings calls this year, and about 30% of S&P 500 management teams were talking about AI.
(06:14):
And that's up from 13% last year, compared to the last hype wave around crypto that really only ever got to 7% of S&P 500 companies. So the business focus on this has been really impressive too. And it shows how AI is really well positioned to grow and scale in the coming years.
David Kelly (06:33):
So it sounds to me it's not just... The generative part is really important because it can actually almost think, which is really what's bizarre about it. But also, Stephanie, to your point, the fact it's this accessibility of the masses. I remember when I started with computers, and now I'm dating myself, you literally had to write in D-O-S for disc operating system at the first prompt in order to get the thing going. But it was behind a very obscure wall in terms of using these. But now anybody can use them because AI is not just good at coming up with ideas, it's also good at understanding humans.
(07:08):
So it could have a lot of productivity implications, obviously, but what do you think? Can you quantify what those productivity implications might be?
Stephanie Aliaga (07:17):
Yeah, so there's a lot of estimates out there. I mean, just starting on a firm level there, there's a range of empirical studies that have looked at this and they've observed that companies adopting AI technologies can experience something like 10 to 20% in labor productivity gains just in their one company. And then there's estimates that kind of expand on an aggregate productivity level for the US economy. And Goldman Sachs thinks that that could potentially contribute to two to 3% in productivity growth every year for the next 10 years.
(07:50):
McKinsey also has done their own analysis, they found similar numbers. And to put that into context, productivity growth has been pretty lackluster in the past 10 years. Me and you have done a lot of work on this. One of the more frustrating things when forecasting economic growth, we're just not getting productivity.
(08:06):
And that's grown at 1.4% since 2005. So we're talking about an incremental gain of two to 3% just from generative AI and AI technologies. Said a different way, I mean, this could add trillions to the global economy in terms of total output. So McKinsey has added this up and they think this could add between two to $4 trillion to global GDP, globally. That's like adding an entire economy the size of the UK each year. And this is just from...
David Kelly (08:37):
I hope it operates better than the UK economy.
Stephanie Aliaga (08:39):
I hope so too. So that we're really talking about some really major impacts here from AI potentially.
David Kelly (08:48):
Yeah. Mike, do you have some thoughts on productivity?
Mike Albrecht (08:51):
Yeah. So a couple of things. First of all, I think there's a tremendous amount of uncertainty here. So we have to really be humble when we're talking about any of these numbers and really trying to quantify anything. It's hard to pin down exact numbers, and you'll hear a lot of very, very wide ranges. And we have those ourselves. The uncertainties come from the technology itself. We see what ChatGPT is able to do today, we don't know what it's going to be able to do in 10 years from now. It's going to be able to advance significantly from where it is now.
(09:18):
There is some uncertainty about the willingness and ability of businesses to really adopt that. And then if that is the case, how do societies respond ultimately? And governments respond? Do people go and that they find different kinds of work after they've maybe been displaced from their job because of automation.
(09:38):
That's something I'm sure we'll get into. But then on top of that, finally, there's all these complimentary innovations that we don't even know yet what they're going to be. So that would be kind of the first thing that I'd add. But we've done a lot of these calculations ourselves, and I think just one thing to set the stage here so we know what we're talking about. A lot of people tend to think about this job is going to be automated by AI and this job is not. But really what the work that's being done is looking at more at a task level.
(10:11):
So there's these databases out there that look at every single task that you can possibly imagine for every single job. You can cross reference that with some other databases to look at the amount of those jobs in the economy and then come up with the task content, every single task in the economy. And you can go one by one and say, AI can do that. AI can probably do that, and so on.
(10:32):
And at the end of the day, you get a number that I think in our view, probably we think that some of these estimates out there are a little bit ambitious, but something like five to 20% of the economy could probably be automated at the end of the day over the next 10 years or so.
David Kelly (10:49):
Yeah, and I mean, you talk about automation, but productivity isn't just about automating existing tasks. I mean, there's more to it than that.
Stephanie Aliaga (10:56):
Yeah, exactly. And that's the other major point when it comes to AI is that when we think about the past transformative technologies that have really shaped the way we live today, so things like the steam engine, when actually the steam engine was first made to help pump water out of coal mines, and then people realized, "Oh wow, this technology's really neat. We can use this to build railroads and steam ships." Which then gave way to innovations and supply chains. It really created the industrial revolution more wholly.
(11:30):
Similar to the computer, it sparked a whole bunch of innovations digitally that populate our lives today. So when it comes to AI, I mean, I think what we're dealing with is there's a number of complimentary innovations that can come out of this that we can only begin to imagine. And the other thing that actually gets me really excited is the potential to accelerate innovation itself from AI.
(11:52):
So AI can help automate a lot of time-consuming tasks. Well, through the research and development process is quite long because of a lot of those time-consuming tasks. AI can also analyze vast amounts of data, and in doing so, it can uncover new insights, better ways of doing things, provide us with those insights, create a more seamless environment for human collaboration and creativity. We can get our own smart assistant in all of our team meetings to help take notes and suggest ideas. And when you hear from the leaders in AI, I mean this is what they're really excited about. It's less so much of the automation of tasks, it's the ability to really help solve a lot of important real world problems.
David Kelly (12:34):
It sort of reminds me of DNA, which not only is the product of evolution, but is an amazing evolution machine by the very frequency with which it makes mistakes and replicates. So a truly magnificent evolutionary process is not just one that evolves but actually speeds the pace of evolution.
(12:51):
This seems like very much sort of a parallel thing that AI can actually speed the rate of innovation of AI. Very interesting. So this is all very positive, but the downside, of course, is you're going to take our jobs. Are you worried about the employment impact of AI?
Mike Albrecht (13:10):
Yeah, I think there's definitely a tendency to think about that risk. And if there is a risk that that's definitely one of the bigger ones. I think there's a lot of work out there that involves repetition, very, very boring work that we don't like to do. And it's a fortunate thing, actually, that AI can maybe automate a lot of that stuff. But you can imagine for us, for instance, kind of reading through a lot of the research that tends to be saying the same thing is coming away with some summary points could help us do our jobs a lot better and save us a lot of time and open up opportunities to do more interesting things. I suppose that the downside to that is that if we've got a whole bunch of other time on our hands, we can spend that doing other things.
(14:01):
And ultimately, that can mean a few things. In a perfectly optimized world, perhaps we could all go out and have a lot of leisure time. The Keynesian 15 hour work week, if you will, really gets materialized. Whether that happens, I think is a big question. Probably the more likely kind of result is that some of us are needed to do jobs, but not quite as many people. And ultimately that means you don't need a team that's quite as big.
(14:32):
The other possibility is that clients wind up asking us to do a little bit more and we end up delivering much more interesting, wholesome work. And that's kind of the other possibility. So the reality is probably a mixture of both of these. The one thing that maybe a related point that I would add to this is that there, there's really a lot that the likes of ChatGPT can't do yet.
(15:01):
So you've got... The way I think about it, and I've used it a lot in my own writing and trying to help me, but sometimes it doesn't quite seem to understand what you're asking it to do it. In many cases, it kind of hallucinates some information that isn't quite true. And when it does do what you want it to do, it can be a little bit wordy. So it's not quite a refined product.
(15:30):
In some ways, it kind of reminds me of the new person on the team who is kind of coming up and figuring out how to do things. So they'll hand you a piece of work, which is very useful. It saves me a lot of time, but at the end of the day, it needs another layer of expertise on top of that. And I think that that's essentially what a lot of us can continue to do. The bottom line though is that you just don't need as many people to perform that, that really interesting work.
Stephanie Aliaga (15:58):
I think we're still ironing out some of the kinks with ChatGPT and generative AI, but the net result is that yes, a large swath of workers will be exposed to automation in their job. I mean, automation is not a new phenomenon. It's not necessarily something to be wholly scared of in its entirety, but I do think it's going to challenge workers to particularly hone in on those skills that humans are uniquely good at, complex problem solving, collaboration, creativity.
(16:26):
And then there's this swath of jobs too, that we don't really think we'll see much automation anytime soon. It looks like the use of robotics, at least on a broad scale, is still a long ways away. So I mean, a lot of the jobs that require manual labor, to my earlier point, will still be there. So you'll still need to mow your lawn or hire someone to do it.
(16:43):
And then there's a bunch of jobs where we just simply, the human component of that job is so valuable, it likely won't be automated. So when you think about daycare providers, celebrities, hopefully, market strategists.
David Kelly (16:59):
Well, yeah, and I suppose part of it is can the economy generate enough demand for new stuff to make up for some of the older stuff that's being done more efficiently? Which is really what the economy's really done. I mean, I've always believed, particularly in America, that people actually don't want to spend more time with their relatives, so they'll find something to do. Something to want and something to do.
Mike Albrecht (17:19):
Yes.
Stephanie Aliaga (17:19):
Yeah.
David Kelly (17:21):
So I think, as you've both alluded to, over the years, people have often talked about how some technology's going to kill jobs, and somehow the economy always managed to find the demand for new stuff, new services, new things, which keeps us employed.
Stephanie Aliaga (17:37):
Exactly. I mean, history shows this too. Over 80% of the employment growth that we've seen in the US economy since the 1980s is explained by this tech driven creation of new jobs. So yes, we've already displaced a lot of the farmers and textile workers that we used to have. And back in the 1800s, over 80% of the US workforce was working in the fields. But then the agricultural revolution, did it lead to mass unemployment? No, actually, well, first it made the cost of a lot of those agricultural goods cheaper. So people had more money to spend on other things and spend, they did.
(18:15):
So that created the birth of all of these new industries, which then needed to employ new people and so forth. So this long history of tech innovation that we've experienced in the US has been associated with the birth of new jobs that have offset a lot of that labor creation.
David Kelly (18:32):
But one of the downsides, even in prior technological revolutions, has been occasionally a widening of the income gap. I mean, certainly that was true of the industrial revolution where there were a lot of factory workers who were barely on starvation wages, even as some in society, really lived very well off the productivity gains of the industrial revolution. So I'm wondering, do you worry that the AI revolution could lead to even greater inequality?
Mike Albrecht (19:00):
Yeah, I think that that's a very fair concern, and it's probably the biggest thing in the next several years that we would be concerned about. And in terms of something that could actually threaten the strength of economies globally and societies and the individuals in them.
(19:17):
You mentioned the industrial evolution, that's certainly been true of that growing inequality over the last few decades, that you've had a lot of economic growth, but it really hasn't accrued to the masses. And ultimately the same could be true for artificial intelligence.
(19:33):
Just kind of thinking about this maybe a little bit conceptually here, if you have a robot taking a human job, that robot's going to earn those wages that the human was earning before, except those wages are accruing to capital, that someone owns the robot at the end of the day. So ultimately, we would think that probably AI and automation in general is probably more accretive to capital holders than it is to labor overall.
(20:03):
The other thing to consider maybe is that the jobs that are maybe the most exposed tend to be a little bit more middle class, kind of like paralegals, if you will. And so you might have more of a hollowing out and sorting people into more extremes on the low and the high side.
(20:22):
The last thing I'd add here, David, is just that as we transition to this new world and as there's potentially an excess supply of labor in certain sectors, that could put kind of downward pressure on the equilibrium wages in those sectors.
Stephanie Aliaga (20:40):
I will say that there might be some countervailing forces when it comes to AI in particular. I mean, I think the long era of computerization definitely led to this hollowing out of the middle. But what's unique about AI is that it could potentially lend expertise to those novice or lesser skilled workers that they haven't gone through the lived experience or gained the credentials to attain those.
(21:04):
And it could actually empower that cohort of workers to provide better services, drive rate of value add. So for example, if you had a junior analyst that wanted to pitch a new digital app or interface for their company, you'd first need to hire really expensive computer programmer to write the code for you, and an expensive graphic designer. Now Google already has a service where you can keep all of the customization tools to build out an app yourself. And so now that analyst is actually empowered to do a lot more than it could have before. Maybe you couldn't afford those skills and such.
David Kelly (21:44):
It makes the power of tech accessible to those who really are pretty non-techy.
Stephanie Aliaga (21:49):
Yeah.
David Kelly (21:49):
Which actually feels really good to me.
Stephanie Aliaga (21:52):
When I applied for this job, there was a Python assessment.
David Kelly (21:54):
[inaudible 00:21:58].
Stephanie Aliaga (21:57):
Maybe we don't have that anymore.
David Kelly (21:59):
Maybe I'll skip Python altogether, but there are other downs, more nefarious outcomes from AI. Do you worry a little bit about some of the more doomsday outcomes from AI?
Mike Albrecht (22:10):
Yeah, I think that that's a long way off in terms of maybe AI being able to take all of our jobs or resulting in some kind of terminator situation where it gains free will and consciousness and all of that. I think in the near term, if we're thinking about the downsides here, going back to the point about inequality, that's probably the biggest one that we're really focused on.
(22:35):
We think that progressive automation is going to gradually erode the ability of people to find jobs that pay well. And for that, I think governments have a big role to play in terms of, first of all, very actively redistributing some of the income shares.
David Kelly (23:00):
But I mean, it sounds like people have to be pretty flexible themselves in just understanding how technology is changing the workplace landscape and adapting to it.
Stephanie Aliaga (23:11):
Yeah, I think that's the other main risk. I think I kind of talked about maybe a more sanguine outcome for labor, but the risk is that this is just happening so fast that it doesn't give labor time to adjust. So I think the potential for large frictional unemployment is very real.
(23:29):
And there's a lot of questions that we have to answer too as a society is now that we're kind of automate a lot of the routine tasks that a lot of workers, they've only been doing that for years. Can everyone be a great problem solver? How do we train people for that? Or are we teaching that in classrooms?
(23:49):
So these are all kind of big questions that I think the speed at which this technology is taking off really makes more urgent for business leaders and governments to address.
Mike Albrecht (24:00):
And maybe just picking up on that last point, I think this is the first time in at least a few decades where it's hard to imagine what the most popular jobs of the next 10, 15 years are going to be as it makes our long-term capital market forecasting a very, very difficult process. I mean, you mentioned coding and so forth, and it's hard to know, the world of tomorrow, is that going to be one where we need more coders to help with all this AI? Or is that something that AI can do on its own and you don't need those coders? Or is the job of coding just going to be something completely different?
(24:34):
So I think if there's maybe one takeaway from all this is the importance of flexibility and people's being adaptable to new environments and also being very active from a society perspective of retraining people.
David Kelly (24:49):
And policy makers also have to have a very clear eye and clear view of what AI is doing to the society and to the economy.
Mike Albrecht (24:56):
Yeah.
David Kelly (24:56):
Well, listen, thank you so much Mike and Stephanie. This has been fascinating and thank you all for tuning in.
(25:02):
Our team will be hosting webcasts and producing more AI content in the coming weeks. You can find links and additional information in the description box below. The next episode of the Market Movers Series, we'll discuss the demographic shifts of this decade that are shaping the investment environment with my colleague Jack Manley. Until next time, thank you all for tuning in.
Disclaimer:
This content is intended for information only based on assumptions and current market conditions and are subject to change. No warranty of accuracy is given. This content does not contain sufficient information to support investment decisions. It is not to be construed as research, legal, regulatory, tax, accounting, or investment advice. Investments involve risks. Investors should seek professional advice or make an independent evaluation before investing. The value of investments and the income from them may fluctuate, including loss of capital. Past performance and yield are not indicative of current or future results. Forecasts and estimates may or may not come to pass. JPMorgan Asset Management is the asset management business of JPMorgan Chase & Co. and its affiliates worldwide.