Fireside Chat with Mary C. Daly at the 2024 Fintech Conference

Kevin Ortiz:

Here at the San Francisco Fed, we believe that being early students of emerging economic trends leads to better understanding of where the economy is headed. This knowledge is key to fulfilling the responsibilities that Congress entrusted in the Federal Reserve system, that’s a safe and sound payments and financial system and a sustainable economy with full employment and price stability.

Now our commitment to looking ahead is why we joined forces with Sunayna and the Federal Reserve System’s Innovation Office and launched the EERN earlier this year. At its core, EERN brings together economists, technologists, academics, business leaders, workers policymakers to share information on new and emerging technologies, exchange ideas and learn about research and insights from different sectors, all in service to understanding the economy of tomorrow. Now, as part of this work, we’re so happy to bring to you the conversation today on innovation and the economy with President and CEO of the San Francisco Fed and my boss, Mary C. Daly.

President Daly embodies the early student mentality. Her career as a researcher, as a policy advisor, and now as a policymaker proves the point that we do our best work when we have a lifelong learning mindset. Throughout her time at the bank, she has pushed us all to think beyond what’s just in front of us today, and more importantly, how economic developments affect individuals, families, and communities across the country and the world. President Daly is joined today by Paresh Dave, a senior writer at Wired Magazine, who covers big tech companies, emerging trends in tech, and impacts of these trends on business and communities. So please once again join me in welcoming President Daly and Paresh. Over to you.

Paresh Dave:

Thanks so much, Kevin. So I want to start with sort of the economic news of the week. Obviously, inflation is still staying high again. Are we still on track for three rate cuts this year or has your view changed?

Mary C. Daly:

You didn’t expect that question, right? So I think the news of the week has been the CPI report, but it’s a really good time for me to remind everyone here that the Federal Reserve and policymakers are not data point dependent. We’re data dependent. And so we’re looking at the whole dashboard of indicators that come in, not just the most recent reading on a particular series.

When I look back and think back at the year so far, what I see is a strong labor market, a robust consumer spending pattern, good growth and inflation that’s not falling at a rapid pace like it was the second half of last year. And what that does is it says there’s absolutely, in my mind, no urgency to adjust the policy rate. Policy’s in a good place right now, and I need to be fully confident that inflation is on track to come down to 2%, which is our definition of price stability before we would consider a rate cut. So what we will do by the end of the year, I think really at the moment is less relevant than what are we going to do in response to this incoming information. And I’m going to reiterate three things that are important. The economy’s in a good place, the policy rate’s in a good place, and there’s a lot of work to do before we can be confident that we have price stability. We’re committed to those three things to thinking about that and our commitment hasn’t yielded.

Paresh Dave:

So does that mean two cuts then?

Mary C. Daly:

Again, I know that’s what everybody wants to know, but I’m going to just follow up with this. I actually think there’s too much discussion about is it going to be two or three or four or one and not enough discussion on what are we trying to accomplish and are we still committed to accomplishing it. So what are we trying to accomplish? Bringing inflation back to 2%, our definition of price stability, as gently as we can. We are on track to do that. The movement of inflation down to 2% was always going to be a bumpy ride. That’s just historically what is true and there’s no reason to be surprised that it’s happening now. But the commitment we have remains the same, restore price stability as gently as we can and maintain our policy stance as long as it’s necessary to be fully confident that we are on that path. So that’s the thing about the world. And if you remember nothing else from that, it’s not data point, it’s data. Data’s a plural word and it means the dashboard of indicators.

Paresh Dave:

So as you look across that dashboard, is there any evidence… In inflation?

Mary C. Daly:

I think it’s really too early to find a place in the economy where you can say, “Oh, AI, that’s why we see that statistic.” What you see is firms across the entire span of all types of firms invested in technologies and workforce training that improve their productivity. You see that regularly among firms and you especially see that among firms historically when we have tight labor markets and rising costs. Then every firm trying to make a profit is thinking about, “How do we get ourselves to be more cost-effective, meet growing demand, ensure we can do both sustainably over time?” And they invest in technology.

But I think AI gets a lot of play because it’s the newest tool, but they’re investing in all kinds of technologies. And when they’re investing in AI, they’re investing in AI across the whole AI continuum, right? From machine learning that they might not have invested into the RPL to some investing in generative AI of course. But there’s a continuum of that many companies had never even started to partake in, but now have started to work on. Too early to see it in my judgment exactly in the statistics. But as Robert Solo, a famous economist once said, you can often see productivity everywhere except in the statistics. So it’s not surprising. That’s something to write down by the way.

Paresh Dave:

I’m curious, is anything on that continuum helping you study the economy? Are you using AI to study what’s going on out there?

Mary C. Daly:

So I’m generally an early adopter personally because I have a PhD in economics. One of the things you do if you do that kind of work is you’re always trying to find a better way to solve the complicated problems that you’re interested in. And so thinking about machine learning, we were using machine learning tools long ago basically to do textual analysis, try to understand better what sentiment of the consumer is, what newspaper articles are telling us. There are many people in our shop who study what Fed speeches say, do textual analysis of the history of fed policymakers and what we’re leading and lagging indicators of what policy changes might be. So we absolutely use that. I use these types of tools with co-authors on different things.

And I think it’s very much like what Sunayna was talking about this morning. There are standard tools that have long been around like textual analysis and then the generative AI tools, which we only use at the Fed in that sandbox that she spoke about earlier today. And that’s because we really are in the experimental phase and we need to make sure that we’re doing in a non-production environment as Sunayna has taught me to say. She gives me all these technology things that make me sound smarter in technology communities, but no, it’s in a non-production environment. I think that’s really critical. That’s where a lot of businesses we talk to are actually practicing, doing experiments so that they have full guardrails before they see two things. Is it adding value? And two, is it safely scalable?

Paresh Dave:

To me though, one of the most fascinating parts of your job is the old-fashioned part of it, which is these round tables that you have with business leaders, community leaders. In an interview with my colleague for Wired this week, you talked about this example of a retailer that’s using generative AI to sort of augment the work of copywriters, which I think is a fascinating example. What else are you hearing recently from these round tables? I think you had one this week with manufacturers.

Mary C. Daly:

We did. We had one this week with manufacturing firms. I mean, it’s just fascinating how people are thinking about using this and how creative they are. So if you’re a manufacturing firm or you’re a design firm or even you can think of architects, I mean, I’m going to talk about a whole range of firms that could do this, you have a lot of what I think of as… Well, I think of it this way because that’s what it is. I think it’s what they told me, it’s a bunch of drawings. You have drawings of things that you’ve designed and created over time. And you might have a full library of these things spanning the length of the time your firm’s been in business. And now you need to do something new. And instead of going and saying, “Well, this is a brand new bespoke project,” you can use generative AI to scan all those documents and say, “Here’s a good starting point.”

It’s very similar to a coding system or a document you want to produce, like a legal document. You could go and look at all the previous legal documents, have the model trained on those, generate a starting point. And then the next point is to have your, if you’re in an architect’s firm, an architect look at it, machine, an engineer look at it, you’re augmenting what the starting point was to get the final product. But think of all that time you’ve saved. And what we’re hearing from firms, whether it’s the copywriters, so for those of you who haven’t read the article, in Wired, the copyright example was, if you have 100,000, 200,000, 300,000 SKUs, those little item descriptions you have to write for your products that you sell as a firm, some of them are extraordinarily small with a low profitability, low margin, and they’re also quite boring to write about.

So if you’re writing about a screw, that’s less interesting than writing about the highest value thing that the firm sells. So what they did is they used generative AI to write the descriptions for everything as a starting point, audit it of course to make sure the screw was the right thing, but then use their copywriters to write the cool stuff. And then it allowed their copywriters to spend more time with the marketing and sales teams to get feedback and even write better things, increase the demand for their products, generate more output growth. And ended up being a story as we heard from this person, and we heard this week as well, is that any technology can replace, augment and create jobs.

And what’s very interesting about AI, and generative AI in particular, is in firms, it’s doing all three simultaneously. It’s replacing skills or tasks, really tasks, replacing tasks, augmenting the talents of individuals, scaling them up, essentially giving them more time to do the cool work, productive work. And then it’s creating jobs because now these firms have to go out and hire prompt engineers and AI strategists, et cetera, to find other places where they can gain from that.

So when you talk to the individuals using it, you can’t help but walk away with a sense of enthusiasm for how this can help. But what I also learned is they’re very mindful, that’s why they started in a non-production environment. They’re very mindful as we are here at the Fed, as Sunayna described, that you don’t practice in real life, you practice in a non-production environment. You make sure you feel it’s safe… In a machining area, the actual physical output it has, the low fail rates that you need to have.

Paresh Dave:

To the point of this conference though, have you had a recent discussion with sort of banks or financial firms? How are they introducing generative AI into their work?

Mary C. Daly:

That’s a great question. And what we’re hearing so far is that really to help with back office operations, things that are just… It’s a lot of processing. And in any financial institution, you have a compliance requirement rightly so. And the things that are challenging in a compliance environment is you want to reduce as many mistakes as possible while you continue to have the throughput you need to keep your clientele and customers happy and getting what they need. You don’t want to block the whole production chain of things that financial firms are doing, like financial intermediation because you are double and triple in the quadruple checking. So having generative AI produce first drafts, which then can be audited, actually it reduces errors and takes away the somewhat repetitive, many would call tedious work, of doing things, where humans, I mean, I’m going to just assume you all understand what I mean by this. If you have to do something repetitively, it often gets so boring, you’re not as attentive.

So if we can use machines and software and large language models and what they deliver to make sure… They don’t ever get bored. If they spit the stuff out and then we audit it as humans, well then we probably accomplish both an increased compliance and a cost saving and back office operations.

Paresh Dave:

But do you trust that that auditing is happening?

Mary C. Daly:

Yes.

Paresh Dave:

Are the results good from the machines? What are you hearing?

Mary C. Daly:

So what I’m hearing is that it’s early days, which is why most firms and certainly the higher the risk of a mistake, the more firms are keeping it in a non-production environment for longer. So I mean, that’s very similar to how we work at the Fed. And we heard this in manufacturing, we heard it in finance, we heard it in retail. You don’t start with things where there’s a lot of cost if you make a mistake either to individuals harming individuals, releasing information that shouldn’t be released. So I think we’re still in very early days.

So I think of it this way. Is it being used? Yes. Is it mostly for practice? Yes. Have we seen some early returns when they take it out of the practice environment like in copywriting and things? Yes. But even in finance and in manufacturing, both, those are things where a mistake is actually really costly. In finance, it can expose clients and customers to harm. In manufacturing, it can expose buyers and users of technology, of manufactured products to harm. So I think it’s early days on the finance side and it’s really would be in back office operations. Think about there’s multiple products. I’ll just say one that everybody knows, Copilot. Those are things that help you with calendaring, internal meetings, ensuring that you’re able to be more efficient. That’s where I see most of the work so far, in production anyway.

Paresh Dave:

So even beyond finance and in these sessions that you’ve had, what would you say is maybe the boldest or most debatable comment that you’ve heard about AI? And have you agreed with it or disagreed with it?

Mary C. Daly:

Well, AI discussions are filled with bold comments on either side. There’s many ways you can say it, but one way that I’ve learned that really kind of describes the range is we have the enthusiasts, who tell you we can do everything, people won’t need to do anything. Those are, I would call the enthusiasts. You can call them lots of it, but enthusiasts. And then on the other side there are the doomsayers. And so between the enthusiasts and the doomsayers, which I set the extreme, so the enthusiasts think that AI will solve all our problems and the doomsayers think we’re all doomed because of AI.

But what I don’t see, and this is why I will repeat what happens in general in technology, there’s always doomsayers and enthusiasts. If you go back in history, you can go all the way back to the Industrial Revolution and there were the Luddites and the people who thought the industrial Revolution would mean nobody would ever be hungry again. Neither turned out to be true, right? The Luddites weren’t right. And the people who thought nobody would ever go hungry again weren’t right. Because in the middle is where most of the technology takes place. And in the middle it’s completely up to us.

So I disagree with the polar extremes, but where I find myself in vigilant agreement is with people who say how this ends up is completely up to us. It’s not up to the technology. The technology is not driving. We drive. The technology is a tool. So the choices we make and the parameters we set up about what we should do and what is safe to do and what is productive to do will be the importance of it. And so what I see from the firms is they take that attitude, they’re taking the attitude of, there’s too much reputation and business risk to just go blindly into this and there’s too much opportunity risk to not do it at all. And so they’re really doing those types of things.

So when I go on the round tables, I don’t find myself interacting with anybody who says things that I find too extreme. It’s when I go to conferences where you hear a lot of economists talk where, “We’ll have no jobs,” or “It’ll take all our jobs. It’ll take everything and we’ll be doomed.” But some are happy about it, some are not happy about it. So I don’t think that’s true.

Paresh Dave:

So one of the new hunger is that AI will eliminate financial exclusion. That seems to be one of the arguments. My understanding is you don’t buy that. So what, if it’s not AI, what is the solution then?

Mary C. Daly:

So I’ve been working at the Fed since 1996. And when I started, they said that community… It’s always something, like getting instant… Not instant. Payments on payments on your phone, that would be the end of financial exclusion. Having policy change be the end of it. Having now the newest thing, FinTechs would do it. I’m sorry, but they said FinTechs would do it. Then it was Facebook would do it. And it’s always something, right? And I don’t mean Facebook itself. I shouldn’t have even mentioned it, but it’s like a big company of people who are trying to do good in other places will fix it. So just going through the list. And what is always the same is those, again, our platforms, tools, their ideas or thoughts, their networks, but people make decisions. And when you think of financial inclusion, there might be barriers of technologically oriented, but there’s also barriers about how do we get people in. Tom was just doing the housekeeping and then talking about the rest of the afternoon. You can include everybody at 299% APR, right? But is that what we really want? And ultimately, those are people-based decisions.

So I think of the AI and other models of financial intermediation as being helpful tools that could aid us along the way, but if we don’t use them well and get organized about what we really want out of financial intermediation for everybody, then I don’t think we’ll get… We won’t solve the problem. It’s just like when they said in the Industrial Revolution, industrializing the economy would produce no hunger anywhere. That didn’t come to pass, not because we don’t have mechanized farming, but because we have decision-making that doesn’t adjust with those possibilities.

Paresh Dave:

Decision-making or profit-seeking?

Mary C. Daly:

Well, we live in an economy that is under the model that firms, we work in a competitive environment and other things, and firms are meant to seek a rate of return on their investments. And that’s what spurs a lot of innovation and a lot of experimentation. At the same time, we have, as you know, social… We don’t make these policies, by the way. Just so everybody’s clear, the Federal Reserve has a narrow set of mandates that Congress has given us for the payment system and the financial system and monetary policy for full employment and price stability. At the Federal Reserve banks, the 12 banks, we don’t make any regulatory policies. So this is all being set by other entities rightly so. So I’m not commenting on the specific policies, but what I will say to answer your question is I think it is decision-making. It’s not just that if people didn’t make profits, everything would be wonderful. I do think it’s about the decision-making of how we go about our business, what we think is important to a healthy and sustainable economy.

Those conversations have to be had in every part of our institutions here in these conferences, among firms when they go to their conferences, at universities, when you’re thinking about what we want. So I don’t think we can point to a single entity and say, “If you changed your behavior, this would work perfectly.” It’s just never been true. It takes all of us thinking collectively about how did we do our work better and more effectively? And there’s a lot of people who don’t have financial inclusion. It’s not just a few people. It’s a lot of people who don’t have the full slate of things that are available to everyone. And those things can, we absolutely can be improving on them, but we can’t rely on a technology to save us. We’re going to have to save ourselves.

Paresh Dave:

Well, I think about a Citi bank stat, I think from 2020, that there would be $5 trillion added to the economy in the US over the five years if we closed gaps in Black education and other areas. But I should have asked this first. What is your preferred way of actually measuring financial inclusion/exclusion? What is the metric by which you even study that?

Mary C. Daly:

Actually, that’s a very good question. What do we mean by financial inclusion? I think it’s hard to define, and that’s why people pick their favorite way to define it. And if you study financial inclusion, sometimes you count, are you banked or not banked, right? But not everybody wants to be part of a bank. And the reasons people aren’t part of a bank vary across the groups you’re talking to or even the individual. And so I’ve learned, and this is a very poor… This wouldn’t be something you would write down and say, “Okay, now we’ve got the master truth on what financial inclusion is.” But I think we have to accept that It’s a complicated concept.

And what we really, I think, mean is that our financial or access to financial tools like savings devices, devices to accumulate wealth, devices to get a loan so you can have partake in buying something that’s longer lived and being able to do what we think, are there barriers to those things that prevent people from doing the other things that are necessary in their lives or that most of us take as important to do, which would be, save for your future, invest in things that produce wealth, get a loan to get an asset like a car or a house. And those barriers, those things have to be removed.

And then the question is, how do you remove barriers? Well, sometimes it’s about education. Sometimes it’s about getting individuals ready for those types of things, knowing that savings is important, knowing that this is going to build their wealth, getting them ability to accumulate wealth. And sometimes it’s about the tools. Sometimes if you don’t have a bank and you don’t want a bank, you might have more access if you use a different entity. We have Lending Club and other things and microlending and consortiums that lend to each other. There’s a group of people and they develop a microlending community that we’ll share our resources and as a group will lend to each other. So there’s lots of modalities you can use. I think the first thing is identifying where the problems are and why the problems exist. And they’re going to be different depending on which community you’re trying to serve. And I think that’s what we have to accept. The answers are different because the problems are different.

Paresh Dave:

So you have to use 10 different metrics, 50 different metrics?

Mary C. Daly:

There’s nothing wrong with a dashboard. I started with a dashboard. I’m going to reassert that a dashboard. I mean, I’m a microeconomist by training. And the thing you learn in microeconomics is you never have a single piece of evidence that tells you exactly what you need to know. So you have to have what we call a preponderance of evidence or a preponderance of information. I learned that preponderance is hard to say repeatedly, so I went to dashboard.

But seriously, a dashboard of indicators tells you what is true because the economy is filled up with thousands of different issues that all look the same at the aggregate level, but when you try to make a single policy to solve them, you find they actually are really different. And so I think acknowledging that we need a dashboard of financial inclusion indicators and that some are going to be more relevant for one community than others is just good health and good practice and not overwhelming. It’s just a good way to collect all the information. Part of being a policymaker, one of the things you have to learn is you need the entire set of things to understand. And trying to get a single summary statistic that tells you everything is really not only hard, but it usually ends in poor decisions.

Paresh Dave:

So I just want to underscore the one thing. Is AI going to help us more fairly assess credit worthiness or no?

Mary C. Daly:

It depends on the models. So the models are only as good as the data that goes into them. I think the common way to say this is garbage in, garbage out. But seriously, as a researcher by training, you know that all the empirical work you do is only as good as the data you have. And one of the things I learned early on in my career, because I studied a lot of issues about disability and inequality and how we treat it and how we work on that in the United States, and what you realize early on when if you do that work, this could be broadened, is that if people don’t answer the surveys and you do an analysis of the survey, you have very incomplete information. And then you might come up with a policy formulation that completely doesn’t meet the demands of that community.

So the important thing, back to your question, is it can only help us if the models have complete information. And what we see again and again, and this is another place I’ll bring in the round table… So we had a great set of round tables where we learned about how incomplete the model coverage is. It’s not that a particular model is incomplete, it’s that when firms have to train their models, they bring them in and they have the data. If you’re a healthcare provider, a financial firm, et cetera, you’re missing components of the very populations you’re trying to reach.

So if you’re trying to score their credit, for instance, as a example, and you don’t have information on how their lives work in terms of you might be trying to score credit for a population that has what we call seasonal work, so they work a lot, say a fisherman, you work a lot, make a lot of money in one period, but you’re not working 12 months. But if your model is only on 12 month people and you look at their income, you think, “Wow, this is their monthly income” and you think this is not good if you happen to get a bad month, right?

So I just think we have to recognize that, again, it’s a tool and these models are only as good as the information that comes into them. And a big growth area that I’m seeing is companies working on data acquisition and cleaning so that the private information is not there, but the data and the attributes of the individuals that it’s been gained from are there so that the models can be trained better, but you have to do data quality, data capture, and then data provisioning. So I think that’s the big area where we still don’t have the full sophistication and expertise that really can take advantage of what these models might do. It’s early days.

Paresh Dave:

You mentioned the class of companies that are providing new ways to access capital or credit. Whether it’s neobanks or FinTechs, whatever you want to call them, do you think that there is stomach from the Fed, especially after what we saw last year with SVB to sort of protect those new entities? Are they providing value?

Mary C. Daly:

So this is one of the times when I have to say sorry, Paresh. So one of the things that’s really important in the Federal Reserve system is that we don’t speak about decisions we don’t have the pen on. And this is an area where I don’t have the pen. That’s a board of governors working with the FDIC and the OCC and Congress in some cases to think about these issues. So I really can’t comment on those things because they’re really important, but they’re not something I get to make a decision on. So if you want to go back to AI in the labor market or inequality or anything, I can totally comment, but not on that.

Paresh Dave:

Do you think the FinTechs are doing interesting things with AI? Are they providing value in sort of, you know?

Mary C. Daly:

I mean, I wouldn’t separate FinTechs as different from any other business work in the country or the world right now. Businesses across the full span of businesses, from FinTechs to financial sector businesses that are wealth management, banks, et cetera, to restaurants, to retailers, manufacturers, everyone… I think Sunayna talked this morning about the Eternal September and we had Eternal November with ChatGPT, the day that that was launched, and anyone who had a phone could get an app and could download the app and could interact with what generative AI can do, that started off a wave of interest, which I think has meant that firms of all types are working on it.

I don’t have any indication that one particular area is ahead of another one. I really don’t. I have an indication that all are in the same sort of point on the maturity curve, trying to figure out where they know what it is. And one of two things is happening. They’re either trying to quickly learn how they can use it to better their efficiency and productivity and growth, or their employees are saying, “I’m using it” and they’re trying to figure out how to harness that use and guardrail that use. And I think that’s true no matter where you are on this span.

And again, I think in terms of people actively putting it into a production environment, it probably is what we’re hearing, and I think this probably scales, but let me just say it’s what we’re hearing, is that the places where the risk is greatest for a mistake is the places where it’s not. It’s the least developed for production because those companies don’t want to go to production with something that can take their business down through one mistake. So I’d say we’re still in early, early days. You got to underscore nascent. The technology is much more mature than the uses.

Paresh Dave:

We’ve spoken a lot about businesses and what they’re telling you. I want to hear a little bit more about what advocacy groups for consumers, for workers, other just regular consumers.

Mary C. Daly:

Absolutely.

Paresh Dave:

What are you hearing from those advocacy groups?

Mary C Daly:

Sure. So consumers are just like the rest of us in this room. When you talk about with consumers, they’re excited, right? Because it’s cool and you can use it and they are using it themselves and they see uses of it.

I was just talking to people in the medical profession. Many medical hospitals, and well, maybe doctor’s offices, the big consortiums, healthcare groups are starting to use AI and generative AI to assist doctors. And so I was talking to them about you have to sign a waiver to accept that this is being used. And they have huge numbers of people signing the waiver. It’s really high percentages. And I said, “Oh, really? Would that surprise you?” And she said, “It did. But what it is we’re seeing, everybody’s excited. They want to see how it works. So now our doctors and nurses and other people have to show them how it works because they want to see how it’s generating the answer.” So I think consumers are fairly excited.

What I think is harder is if you’re a worker, because for workers, they’re wondering, especially ones who are highly leveled knowledge based workers, think of the actors and writer strikes, right? One of the principle concerns was AI and the ability to use generative AI to do the work that they were doing. But it’s not limited to those individuals. You hear it in engineers. We heard from our roundtable participants that engineers are pretty worried because you can do a lot of the complicated engineering. And so how do you help people see… And this is what workers are asking for and I think this is what firms are trying to deliver on. Workers are saying, “Help me see my future in a world where this is used.” And the obligation that I think firms had have is to show workers if they’re going to invest in these things, how it’s being used.

And to maybe your surprise, certainly something that I was pleased to hear is the firms we’ve been talking to are doing just that. They’re engaging their workforce to say, “How do we do this in an effective way? What are the least desirable parts of your job? And what are the things where we can scale this?” I think all of us were here for Sunayna’s talk this morning. At the Fed, in the Federal Reserve system, that’s the same thing, helping individuals and business people in the Fed, our business lines, engage in this themselves. So it’s not the technology versus worker, it’s a technology with the worker, because ultimately, we’re all workers and we all want to be part of this. I think that’s the level of engagement that we’re going to need. So I definitely feel, just to recap, apprehension on the part of workers. I feel that normal apprehension that comes with any technology rollout that this could take our jobs. And then I hear wanting a bigger place at the table than they’ve traditionally felt they had.

Paresh Dave:

A desire for agency.

Mary C. Daly:

A desire for agency. I think that’s a great way to put it, desire for agency and determining the future. It helps frankly that we have these models. And there’s many examples so I’ll try not to keep naming the one that came out first. On your phone, there’s many ones you can use. And I think having people have access to that helps with that agency because then you know the power, but you can see how you might harness it to help as opposed to harness it to take.

Paresh Dave:

Speaking about the future but also the past, one of the things that I’ve written about AI is how, during the pandemic, there were a lot of models that just went haywire because it was such a shock. The old data just didn’t make sense anymore. As you think about the economy going forward, are you starting to look at how do you ensure that AI doesn’t cause some crazy instability the next time there’s some weird shock?

Mary C. Daly:

Sure. So I guess if you work in a Fed or any other, or a researcher of any type, we have research economists, models, et cetera, you deeply know that the model’s only as good as the history it’s trained on, right? And that’s true of any model. It doesn’t have to be a natural language model. It doesn’t have to be a generative AI. It can just be a statistical model, running a regression on data. It’s only as helpful as the history is to the current situation. And so in the pandemic, all models struggled, right? Every single model struggled. Most models struggled in the financial crisis because we hadn’t had a financial crisis for many years and the model data didn’t really help us. And so what you do in most models is you try to figure out how to maneuver around that, but I think there’s already that baseline awareness.

So where I would be focused, I would say focused more than worried here, is our focus has to be when we use these generative AI type models, that we recognize that when the world has a big shock, a pandemic, a financial crisis, a huge change in how things are done, well then the model won’t be as able to generate answers that are close to right and more auditing has to be done and sometimes you just have to say, “Okay, we need to retrain the model.” So these models that they’re working well over time will train themselves constantly. But when you have something like a pandemic, that’s a perfect example, the world closed down. We hadn’t seen a pandemic since 1918. So of course nothing since 1918 when we didn’t have natural language models and we didn’t have much data to feed into normal economics models, we just were flying without that kind of model.

At the FOMC, you might even notice if you look back at the minutes, that we admitted that our models weren’t going to be a good guide to this. And so we were going to take incoming information, search for new data sources, think about how to retrain our own models, not natural language models, but our own statistical models in terms of thinking through this. And if you’re trained well in this environment, you can rise to that occasion. The discipline has to be not reflexively relying on it when the world has changed. And I think that should be our focus and a caution. And that can be true not with big, big shocks that happen to everybody and with small shocks that happen to communities or to businesses, right? So that’s a discipline and a practice we’re going to have to get better and better at.

Paresh Dave:

I’m glad it’s on your radar.

Mary C. Daly:

It’s definitely on our radar. I mean, again, I live through this for many things. We have been through a financial crisis. I’ve been through a pandemic now as an economist and now as a policymaker. And both of those teach you what you inevitably know if you work in any kind of research is that the world changes. And if it changes abruptly, the past is only a partly helpful guide. And the worst thing you can do is just assume it’s a perfect guide.

Paresh Dave:

I want to end with a couple of questions more on financial inclusion, less on Ai. One of my pet peeves with sort of the financial ecosystem is that there’s a lot of artifacts of sort of a world where one person in a family or a household sort of made decisions and the sharing that we see through families and couples and friends and how they make decisions just isn’t contemplated in a lot of the financial ecosystem. How do you think about that and what’s your sort of pet peeve with how the bank’s financial ecosystem works today?

Mary C. Daly:

So I think it’s interesting. I try not to think of things as pet peeves, but as opportunities to make things better. So that’s just a difference in sort of approach, I think of opportunities to make things better.

But I think one of the things that can happen, it happens in the financial system and you see it in other systems so I’m not just going to say this is a financial system only, but it is something that you see easily in the example you just gave. The economy and society changes. And the infrastructure that we have to match, it lags. And when it lags, it doesn’t fit anymore. And so then what once felt like perfectly inclusive no longer feels perfectly inclusive. I think your idea of couples organize themselves differently, families organize themselves differently, that really matters.

I will say I had an early experience as a young person, I had to work when I was younger and I needed a bank account. And I went into the bank and they told me I couldn’t have a bank account. And I said, “But I have a paycheck and here’s my paycheck, here’s my stuff.”

“You can’t have a bank account unless your parents come to sign.” But my parents needed me to get a bank account because they weren’t really able to come and sign. And so the whole thing was this gigantic mess of how they couldn’t meet me where I was. So I actually went to a smaller place and I told the woman my problem, and she said, “Okay, I’m going to give you a bank account.” She opened my bank account, she showed me how to write the checks. She taught me how to do it. I managed to do everything I needed to do, but it was an interesting moment for me to recognize that she made that decision, not a machine, not a technology. She had to go outside of the rules a little bit that you would say… And so I think, how are we going to change these things? The technologies, they are to change it. We could change to be this. And so it’s practices.

And so my opportunity for improvement, or in your Farland’s pet peeve, is that I do think that all of us who are working in the financial sector have to step back and ask, “Are we building a system that we would build today if we started from scratch?” And I’m guessing we’re not. And so then if we’re not, how would we make modifications or change entire ways to look at it if it’s so much dissonance between what we would build today if we started now, and what we actually are carrying with us?

And that’s for me, always been a good device to know where to focus. Because if you look at these two things and they’re vastly far apart, you can say, “Oh, we should start there.” Because narrowing that gap… And what that woman did who helped me when I went to get my first checking account, what she did is say, “The lived experience of this young woman is different than the rules that the bank has for who can get a checking account, what you need. And I’m going to close that gap because I’m just going to take agency.” But if all of us started closing that gap by asking, “Where are we? And is this what we designed today?”, I bet we would find a way to be closer.

Paresh Dave:

But didn’t she risk being fired?

Mary C. Daly:

I don’t think so.

Paresh Dave:

Not following the rules.

Mary C. Daly:

I mean, so she didn’t break some major rule that was going to cause financial instability for organization. She just said, “I’m going to sign this poor kid’s application for a checking account.” I mean a checking account in the olden days, there was no overdraft. You ran out of money, your check bounced. Geez, it’s over. There was nothing that was going to make the bank at risk. She gave me a little paper, checkbook, and showed me how to fill it out, showed me how to take the register, took my paycheck. I went there and deposited every time. She took the paycheck for me, showed me how to do it, go to the teller, all those types of things. Zero risk to her. A simple act of kindness backed by a simple concept of what the bank was founded on. They had a big sign out there. “We are here to help.”

I gave a talk once over in Singapore on a bank culture conference talking about the culture of financial institutions and how we can make it better. The basic premise there was that the organizations are at their best when they’re trying to close this cognitive dissonance and remember why we’re here. Financial institutions are trying to help people, trying to help with financial intermediation. If you keep that in mind and you keep the regulatory risk-based guardrails in place because you have to internalize the risks that we’re taking for people, then I think you get to better outcomes. And so no, I don’t think she risk getting fired.

I think the main message of the story is she took agency within a sense, which was still risk-free for her institution. And if we did that, I mean, I’m not recommending if anyone’s working in a financial institution, you go start doing things randomly without talking, but you could absolutely see a series of things that you’re missing as an institution and move it up the hierarchy and see if that’s something you should fix. I think those are the opportunities that we all face that I think are worth pursuing so we can turn pet peeves into opportunities.

Paresh Dave:

Great. Well, thank you, Mary. I really appreciate this conversation. And thank you all for being part of it.

Mary C. Daly:

Thank you. We covered a lot of territory. We’ve been everywhere. Thank you.