AI’s Impact on Financial Services

Itay Goldstein: Hello. Welcome to the “Future of Finance” series. This is our third season in which we are talking about AI in finance, the various ways in which AI is introduced into finance, and how this is promising to reshape the financial industry, financial regulation, and the economy more broadly.

Today, we have two perfect guests to talk about how AI is changing financial services, and how AI is affecting the workflow in financial services. First, I have Jeremy Siegel, who is my colleague here at Wharton. He is now an emeritus professor at [Wharton’s Finance Department], and over the years has been one of the most trusted voices, familiar voices from [Wharton’s Finance Department], still appearing in financial media very often. Jeremy Siegel has written the book Stocks for the Long Run, which is one of the most essential reads for asset managers, and is something that has shaped the way people are thinking about investing in financial markets to this day. Hello, Jeremy. It is great to have you.

Jeremy Siegel: I’m happy to be here, Itay.

Goldstein: And then with him, we have Jeremy Schwartz. Jeremy Schwartz met Jeremy Siegel as a student here at Wharton and joined him very shortly after that in WisdomTree. Jeremy Schwartz is now the Global Chief Investment Officer at WisdomTree. Together, Jeremy Siegel and Jeremy Schwartz host the podcast Behind the Markets, where they talk about issues at the frontier of finance and asset management. Jeremy Schwartz is also on the advisory board of the new master’s in quantitative finance at Wharton that is being launched these days, and that we are all very hopeful about. Hello, Jeremy. It’s great to have you here as well.

Jeremy Schwartz: It is great. I’m excited to be on that board, Itay. I wish they had it when I was a student— I graduated in 2003. I wish they had that program. I’m jealous of the young kids going through it. It’s great to be working with everybody there.

Goldstein: That’s great. This is a great place to start. And certainly AI in finance is going to be something very essential for the new master’s in quantitative finance. So with this introduction, let’s dive right in.

To start at a broad level, let me ask you the following. We know that AI in finance has developed a lot in the last few years. As people who are working in the industry and familiar with the various ways in which AI can be used in finance, maybe you can give us a broad overview of why AI is so consequential for the financial services industry and what it might do to the financial services industry.

Siegel: Jeremy, do you want to start there?

Schwartz: Sure. I think there’s no question AI is the top market consideration. The professor can talk about all the macro level impacts and what it means for the markets. Back from when I was studying under him at Wharton in 2000, you had the big tech stocks. Now, where you are today, clearly AI stocks are driving the markets. [They are] driving earnings. We see, actually, the markets continue to be supported by very strong earnings, driven by a lot of the AI technology names.

For companies like ours at WisdomTree who are creating investment strategies for it, it makes you think: how do you position within the markets? What are the different products you’re offering? How do you do them efficiently? We’re certainly seeing things like our AI-thematic-oriented products being the top performers and leaders.

And then how do you run those strategies? You’re using AI tools to do that increasingly as a way to cover more companies, to expand your capabilities and really transform all of your workflows throughout the organization. So it’s both very macro level for the markets, and then specific towards, how do you run all of your investment strategies?

Goldstein: Great. Jeremy?

Siegel: Well, first of all, let me say the broad. We are very positive on AI. I’ve written an article [titled] “There is No AI Apocalypse.” We believe that the increased productivity is going to be very positive for the economy. I actually believe that this is really, perhaps, the new industrial revolution. And we survived that. Certainly certain professions go away, but we actually have survived that.

There are so many aspects on AI. Jeremy mentioned a few about the investments. We all know the Mag 7, and most of those are AI. We can now maybe add SpaceX and call it “Mag 8.” The AI firms are very, very top heavy in the market. But in contrast to the dot-com firms that we saw 26 years ago in the bubble of 2000, [when] many were not earning anything and it was all just a promise, these firms are. Now, there are a few that are not. SpaceX is showing a loss.

But on the Mag 7, they are earning. In fact, their price-earnings ratios are not unreasonable given their growth. We could talk about the impact on finance itself, but I do want to talk about the market. The growth is spectacular. If it would continue, they would actually justify much higher prices.

The reason why they’re not, and still at 20-25 times earnings, is the fact that the breakthroughs could be so large that the moats that have given these stocks such high profit margins, record profit margins, could be threatened by breakthroughs. That’s what happened in the dot-com revolution with multi-flexing through the fiber optic cable. It might be very well possible that we’re going to see some breakthroughs, and all of a sudden the Blackwell chip that Nvidia has can be superseded at much lower cost. And that would cause a disruption. Now, all that is good for the user, but certainly would threaten the valuations that we now see in the marketplace. So you have to talk about what is the macroeconomic effect of it. Then, of course, what is going to be the stock market effect? There certainly could be differences.

Goldstein: In the article that you wrote, you touched upon the fact that you don’t think there is an AI apocalypse. Let’s dive in a little more into that. Because I think this is certainly essential and also very fundamental to how we view AI in finance.

I think the core of the argument of people who think that there will be an apocalypse is that AI is gradually replacing humans in all tasks. Unlike other technological revolutions that we have, where the technology can replace humans in some tasks and then humans are finding other tasks, what is unique now is that AI can then step up and replace humans in the new task and so on. So there is really not much hope for the labor force. I understand that you don’t buy this argument. Maybe you can elaborate on this a little more.

Siegel: There’s a couple of things. I know that Anthropic did a study on how much AI would actually impact them (economic tasks). And some were very great. Others were like ground maintenance. I guess maybe if you think of AI as robots that can do gardening and ground maintenance, that is possible in the far future. You still need a lot of people in some of the professions. So it doesn’t replace everyone.

But I think the important thing that I brought out is that if productivity shoots upward, that means that we could produce the same amount of GDP today in less time than we could without the AI principles. For example, if we double productivity, we can produce in two-and-a-half days — on a five-day week that we have today — what we used to produce in five days. If we can double the output per person hour, we will double the wage. So people will be able to work half the time, get the same wage, and produce the same GDP output.

But what a lot of people are failing to realize is that the tremendous increase in the real wage will incentivize people to say, “Well, if I work more — for instance, three-and-a-half days rather than two-and-a-half days — I will increase my income by 40%,” which increases GDP in our $32 trillion economy by something like $13 trillion.

Most of it will be in higher-end goods — things people will be able to do to afford types of consumption that are beyond them today. Cruises, trips, second homes, second and third cars, fancy restaurants — we can go on and on and on. Beyond the reach of most Americans, [but] not all. They will be in the reach of many more Americans. I’m saying the word American, but it is true of everyone in the world that will incorporate it.

What you’re going to see is a tremendous increase in demand. Now, combined with that, we have to realize that 90% of the white-collar labor force is a fungible skill. What I mean by that is that they’re trained in certain professions, they’re trained into certain skill sets into certain companies. But their skills, which is discipline, being able to follow orders and do those tasks, could be spread to many, many different types of industries. And they don’t have to stay in [their industry]. If their industry gets replaced, there’s going to be a tourism industry that is going to be exploding that will need their services.

One of the examples I like to point out is that travel agents at the high end have seen a tremendous increase in demand over the last 10 or 20 years. LinkedIn says it’s one of the top-growing of all professions. Many people thought that travel agents would disappear in the world of AI, in the world of everyone making their own reservations online. The thing is that that high-level, personalized type of services are still in demand. Radiologists, which some labor economists predicted 10 years ago would disappear because AI would read the scans, are now in demand more than ever, because the number of scans has gone up by 10 to 20 times. AI is going to reduce the cost so much that there’s going to be so much more output. A person could do it that much faster, but the increase in that output will still cause a demand for his or her services.

All that is really going to be multiplied. Some people call it the Jevons effect, that [when] you lower the price, you use it so much more. It’s a little bit different than that in economics, but it is, in effect, that the growth of AI will enable people to increase their levels of consumption, GDP output, and expand industries greatly that will absorb the labor that is lost because of the efficiencies in some of the industries through AI.

Goldstein: Yes, these are very important points about the fungibility of human labor and how new jobs can be created as a result of that. I want to turn to you, Jeremy Schwartz, and think about how this is reflected in financial services. What do you see? How is human labor interacting with AI? Do we see AI agents replacing humans or do we see humans finding new ways to do things that they have done before, but now do it better?

Schwartz: I can tell some anecdotes from how WisdomTree itself is being impacted by this and some other things I see in the industry. WisdomTree today has about 400 people globally, [and assets under management of] about $170 billion across the world. There’s this narrative out there that you’re going to want [fewer] software engineers, because now the first task AI is good at is it can replace all the software people. I’ve always felt constrained on tech resources. Do we have enough tech people developing all the support?

Now really everyone has the ability to develop their own software. We have access at WisdomTree to ChatGPT, to Claude, and so we’re using both of those. I think over time we’ll probably want to get an open source model to manage your token spend. But you’re using tokens in a way to develop a lot of new output.

There’s this whole narrative about “I want fewer young people,” because there’s this fear that the young analysts [could be replaced], because now anybody could put in something to ChatGPT, and we don’t need as many analysts because we could get the basic common information surfaced much quicker just by talking with one of these LLMs. In some ways I actually want more of the young people, because you think about having to get people to change their workflow, change their protocols. The young people who grow up using these tools and trained using these tools, they know no other way.

But in many of what you see, you often need to deploy what we call “AI ushers” or “AI enablers” to work with the businesspeople to develop the capabilities, b it’s hard to get people to learn new tricks. When they have a daily workflow, spending two hours a day trying to change their workflow is not always the easiest frame of mind. So I want more young people. I want more of these people doing it.

And then we’re building more apps. In just the last few months, our teams built probably 40 different apps without the software engineers. It’s like my own team developing our own system with more analytics and capabilities.

So there’s a flood of new information. I think one of the things you see is the cost of asking a question goes to zero. You’re going to ask a lot more questions and build a lot more things. But now you’ve got to have judgment on those questions. So you still have to have people who can parse through all the additional insights you’re getting. But I see wanting more young people trained with these skills, not less. I do think it is definitely changing the workflows that we have.

Siegel: Let me follow up on that. Think about— You want to research an industry. You want to know whether it’s good to invest in or not. You want to know what the major firms are. Now it used to be [that] you got your research analyst. You gave them a week, you assigned them to different aspects of this, and then you got together to talk about it. Think about what AI can do. You now put the question in AI. Instead of a week, it’s an hour. And in fact, you ask it to put out recommendations.

So what are you going to do? You’re going to sit around and talk. You should talk about, “Well, how good are its recommendations? What are its sources?” and all the rest. You have just sped up the process by multiple orders of magnitude, in terms of where you’re going to start the discussion about ROI. This is going to increase efficiency, I think throughout the whole economy, because investigations and information gathering are just going to be so cheap that you’re already there.

The question is, is it able to actually make those innovations? Are we there to the point where AI makes the innovations? When we know it’s solving mathematical theorems — some of which have been unsolved forever, [say] decades — can it be used in finance? There are so many aspects of finance. But in terms of economics, can it be actually used to develop innovation?

I think eventually it can, which speeds up productivity growth and can actually suggest ways to go forward. I’m extraordinarily excited about the possibilities. Just in terms of freeing up time, I’m going to give you a very mundane example. Our podcast that you mentioned, Jeremy used to write it up. It used to take him an hour-and-a-half, or two hours. I would then edit it, and then we put it out in print a couple of days later.

Well, now, of course, AI takes what I’ve done, also remembers everything I said before, can refer back to previous podcasts, fill in data where I may only know the approximate number. And Jeremy can get me that back in 10 minutes or 15 minutes. My editing process is much faster. That’s just one example of how much faster things can actually go.

Goldstein: You’re both very bullish about the innovation and the potential for the future. Maybe we can take a moment to also reflect on what are some of the potential negative aspects or what things we should be concerned about or just watch out for, if there are any risks to the financial industry or more broadly that you still think we should keep in mind?

Siegel: I’ve always said this, with or without AI. AI may actually, I think, magnify the cybersecurity [issue]. Are we going to invent AI so good that it can break through our own systems? I think that’s the biggest threat. I mean, certainly on the financial end. I do know there are worries about biological weapons that AI could, in fact, develop, and that’s certainly something that we have to look into. But for the financial firms, it is cybersecurity. I think they have just got to stay ahead of AI.

Goldstein: Do you think that we are doing enough for that? Do you think there is enough investment in cybersecurity?

Siegel: Obviously good guys staying ahead of the bad guys, one step ahead — we need to do that. That needed to be done even before ChatGPT was developed four years ago [in November 2022]. We knew that there were people that were trying to hack into our system. Now they can do it even faster. And so we’ve got to stay one step ahead.

I think it’s a threat like a military threat. You have to stay one step ahead, militarily, of your opponents, to make sure that you can deter any attack. And we have to step up on the side of cybersecurity, one step ahead of our opponents there. So resources definitely need to be put in. They are being put in. Private enterprises are putting them in. The government clearly has obviously a role there, because national security is involved.

Goldstein: Jeremy Schwartz, do you see any risks? What worries you?

Schwartz: Even just [in] the week that we’re recording this, you saw the government add some executive orders to do quantum computing support. Quantum is one of these next generation technologies that people worry can break things, like the encryption for all financial services, for Bitcoin, all these things out there. So it is good to see some strategic investment. If we think of a lot of the great innovations that happen across time, some of the government funding has helped. So it’s interesting to see some of that coming — [such as] public-private [partnerships], combining to join forces on that.

The new technologies can be used for good or for bad. So it is a race of, “What are those risks of the bad actors using the new technology?” [Jeremy Siegel’s book] Stocks for the Long Run calls this optimistic bias. We do believe in a very positive future, and we do believe in this enabling us in the world at large to make progress forward on all these things.

Siegel: We haven’t really talked about the ordinary investor. We talked a little bit about AI stocks. There is some incredible growth, and the risks of breakthrough that are always present. But it’s also the whole question of will we be a totally indexed market? I mean, what is the role of the analyst?

In that case, if everyone has all the relevant information — well, don’t forget, AI still learns from what has been written. Of course it could take what has been written and try to push through logic. Can it develop a brain that could do better than the best investor? We know there are a very few handful of people that have beat the market in the long run. And oftentimes even they have their periods where they don’t. Will AI develop it?

If it’s freely available or cheaply available, then everyone will go into that and we’ll get better pricing. We’ll get more efficient pricing on assets, which spurs productivity. If prices are closer to being right, we won’t spend as much capital on projects that AI tells us, “This is not going to be profitable.” While today, without AI, we might do that. There. are a lot of questions going on there. We’re not there yet, but it’s certainly fascinating to try to envision.

Goldstein: Yes. Of course, one of the fundamental paradoxes that we have in finance is that if everyone is just learning from the price system, no one will invest the resources. The question is whether with AI, this paradox is—

Siegel: Yes, the Grossman-Stiglitz [Paradox] — the impossibility of a totally efficient market.

If everyone’s an index, no one is determining relative prices, which means the first person that does that is going to make incredible profits, right? In a way. So, there’s that balance. I don’t know. I haven’t really been thinking to infinity of what AI could basically do. But, I don’t think there’s excessive indexing today. I’ll forget that question, because the evidence is still that active managed fees cannot beat the index. And so, we’re not there at an over-indexing point.

The question is, if AI starts picking [stocks], it would be very interesting. Would a Claude-AI-optimal portfolio be the same or different than a ChatGPT or a Grok or Perplexity or whatever else comes in? It would be really quite interesting. Would they converge, or are slight differences in learning processes going to give them bigger differences in portfolios? These are things that are, I think, fascinating to conjecture. But we’re not quite there yet.

Goldstein: Right, yes.

Schwartz: There are a few interesting experiments being run online, including by a former Wharton professor who’s working on a bunch of these. But you could see on social media them talking about Claude-driven portfolios, Grok-driven portfolios, ChatGPT-driven portfolios, DeepSeek. There’s a whole host of the AI competitions trying to pick stocks and seeing which ones are doing better.

And people have been using quantitative strategies. What AI is doing is running a bunch of models and putting it all together. AI is very good at generating the consensus idea out there. So, can it find the contrarian idea? It is one big question.

But, at WisdomTree, we are exploring with machine learning models. We have a long-short hedge-fund-like strategy that’s in the market. It’s only five-month history, but it’s with a firm in Israel called AlphaBeta. There are four different machine learning models that are trying to rotate between facts. Some of these use dynamic factor timing in some way. But those models have been adding a lot of value in the five months it’s been live. And so I think there’s going to be further things like that where you could have advanced modeling and then things like the LLMs trying to pick stocks as well.

Siegel: Yes, I guess the individuals whose jobs are threatened are those that are good stock pickers now. Because AI might be better and freely available to everyone else.

Goldstein: Yes. Talking about education, [especially] financial education, this is an important part of what we do here at Wharton. And Jeremy Siegel, you’ve been on the faculty for about 50 years, I think. Jeremy Schwartz, you were a student here and now you are on the advisory board of our new program. What would you like to see introduced into the curriculum in finance to prepare people for this new era of AI in finance?

Siegel: A lot of people do ask, “How would you teach in a world where AI is giving such good answers instantly?” And often what I would do is the following. I would come in class and ask the GPTs, the various ones, a certain number of questions and see what they come out and then have the students discuss it. “Well, here’s a little difference between this one and that one. What do you think? Or do you think that it is picking something? Do you think it might be ignoring something that should be picked up? If it is, why is it?”

The teaching process is still there. In fact, instead of people cheating with AI, a lot of people said, “Oh, they’ll just get an AI answer.” I would give them the AI answer and say, “You write a critique on the AI answer.” That does need your own brain. Or say, “Why is AI ignoring this? We learned about this asset. Whatever it is, [there is] some correlation. It seems not to take that into account. Is it doing that right or wrong? Why do you think it ignored it? Maybe it sees something you don’t know, and it can be ignored.”

I think teaching is still as powerful and important. And that’s the way you should start. You should start with the AI answer and critique it from there. Of course, that’s after learning the principles that we teach in the 101 classes.

Goldstein: Right. These are great ideas. Jeremy Schwartz, what do you think?

Schwartz: If there’s a fear of using AI in the classrooms, I really want the students to come out of the program with specific training on how best to use the tools. The people who I really want to hire are natively using it to their best capabilities. We are hiring people because of their AI experience. I don’t want them to be afraid of using it in the classroom setting. So [it’s about] being fully up-to-date on all the developments, what is the best use case, how can you build your own tools with it? People who are experiencing building things on their own, using the tools, are very highly valuable.

I was here for 20 years. I am trying to set an example for my team of showing how to use it myself. But not everybody’s like that. People get stuck in their ways of doing things. You need to be able to use these new tools, because some of the young people can do 10 times as much as others because they’re using the tools.

So they’ve got to be trained in all those fundamentals and the principles of good data science work and quantitative finance and all those things. But they also need to know, inherently, how to get their stuff done using Codex and Claude Cowork and all these things to move things forward.

Goldstein: Okay, great. Our time is almost up here. Let me just ask you each one final question. If you could try to imagine how the financial industry, how financial services, are going to look like 10 years from now, what would you highlight?

Schwartz: Maybe I’ll start. In some ways, some things may not change. Like, if you’re a financial advisor working with clients, what might change is how we actually deliver the solutions. The back office, the middle office, the team supporting the funds. But finance is pretty highly-regulated. I think the end client experience may be similar.

As new generations inherit wealth young people might want to be served in a different way than the older generation. But [for] financial advisors, consulting clients, there might be more customization for people. More tools. More solutions that are more unique, because the cost of servicing a customized solution drops.

But I actually still think people will want advice. [If] they’re getting advice from themselves, it will lead to more do-it-yourself. It may. I’ve heard anecdotes from advisors saying their clients are saying, “Hey, Claude surfaced this ETF portfolio. Why do I need you?” So there’s a fear of that. But, a lot of people still want the trusted advisor. And so for the next 10 years, I don’t think it’s a complete shift to do-it-yourself. I think a lot of it won’t change, and it’ll just be how things get serviced — that is what’s really behind the scenes.

Siegel: Let me follow up. Even though I’m a big fan of indexing, I often said, “The biggest gain that financial advisors earn their worth and their fees is keeping people in markets when times get bad.” And that involves emotion. We all know about behavioral finance, which 30 years ago wasn’t a topic, but is now integrated very much with traditional finance. And human emotion, to keep that under control— I don’t know whether a trusted advisor that’s been with you for a long time will measure. And if Grok said something different, which one are you going to go with? I think that that becomes just still very, very, very important.

We talked about the pure indexing and picking individual stocks and the brainstorming that’s going to be around. I think it’s greater in efficiencies. We have not talked about transactions. We’ve not talked about our payment system. We haven’t talked about cryptocurrencies. We could spend an hour on that. And I think AI is related to that also.

But I still think the human fact of an advisor [is important]. When I wrote the book Stocks for the Long Run, I concluded, “Hey, stay in the market during all bad times if you’re in long-term, and that will be the best thing.” And that requires not the intellectual ability to do that, but an emotional convincing of that. That’s someone that’s trusted. That’s someone you just say, “I’m not selling right now. I know times look bad. But history has told me.” That is something that I don’t know whether AI will ever really [replace]. Will you ever trust a robot more than a human being that you have spent 20 years with trusting before?