Breakfast Briefing-AI's Evolving Role in Public Finance

Briefings on practical topics giving precise instructions and essential information.

Transcription:

Gregory Dawley (00:06):

Good morning and thank you for coming out so early to this breakfast after the late night many of you had at the BAM and Chapman events. My name is Greg Doawley and I'm RBCs, West Coast municipal manager based in Los Angeles. We want to thank the Bond Buyer for the opportunity to again sponsor this policy briefing. Now in our 10th year in this role, we're at the early stages of a period of vast technological changes in both government and business. Last year, our breakfast topic was on blockchain, something that seems still several years away from making a significant impact on government processes.

(00:42)

This year we discussed AI and its role in society and specifically within public finance. This development is moving out of speed, never experienced before with ramifications throughout the world in many facets of life. Earlier this month, Governor Gavin Newsome signed executive order N 1223 on AI that aims to study its risks and potential benefits with the eye to avoiding AI's, potential amplified bias. State Senator Scott Weiner from San Francisco introduced Senate Bill 294 at the end of the last legislative session that envisions creating a state agency to regulate the responsible development of technology while also putting the onus on developers to ensure their technology isn't used for malicious purposes. Wiener's view is that society made a mistake by allowing social media to become widely adopted without first evaluating the risks and putting guardrails in place. And that repeating the same mistake with AI would be far more costly.

(01:42)

The amount of medium public discourse on AI is seemingly nonstop, and there are entire conferences on just an narrowly defined sector and policy impacts. Today we will break the panel up into two sections. First, Rishi Jaluria from RBCs Research Group who specializes in software and is based in San Francisco will be our keynote speaker, focusing on the growth and potential impacts of AI and society, government and industry at large. The second half of this event will include a panel discussion of AI's evolving role within public finance. This discussion will be led by Brian Olson, a Director on RBCs municipal Underwriting desk, and the former municipal credit strategist at our firm. He will be joined by Alex Domenick at RBC, Janice Hofferber at Moody's, and Matthew Smith at Spline Data. I'll now turn the presentation over to Rishi. Thanks again to everyone for showing up for the breakfast this morning.

Rishi Jaluria (02:45):

Alright, wonderful. Thanks everyone for being with us today and giving me the opportunity to speak. My name is Rishi Jaluria. I co-head the enterprise research software practice here at RBC. I'm based out of San Francisco. I follow and cover a number of enterprise software names like Microsoft, Salesforce, zoom, household names that you've heard of. Really, I just want to talk a lot about the work we've done around generative AI. This is something that ever since Chachi PT burst onto the scene about a year ago, we've been writing extensively on looking at things like monetization, how we pick winners and losers, societal implications, who's at risk from here, even getting into the actual technology and differentiating between a number of the different large language models that exist out there. So I really want to talk through all that, but really to level set, I think it's important to emphasize how important and how seismic a change generative AI is.

(03:44)

Some of the skeptics out there say, oh, is this the next metaverse or Web3 or NFT, very, very high technologies that haven't had real impacts? And the answer to that in my mind is no. This is instead really the fourth big technological wave in my lifetime. Going back to the internet, going back to mobile cloud and now generative AI. And each of those technologies had a seminal moment where it became mainstream where everyone started talking about it and you could see a path to disruption with the internet. It was a release of Netscape Navigator that anyone could browse the internet on their own home computers with cloud. It was both AWS and Salesforce that really democratized access to the cloud and had a huge disruption to the enterprise with mobility. It was the iPhone that was really that big moment. And for generative AI, even though we've all been talking about AI for a decade, machine learning probably for three decades, generative AI is now in the mainstream because of the release of chat GPT a year ago that is not only incredibly powerful, but has actually democratized access to generative AI that anyone can use.

(04:52)

And we'll talk a lot about use cases there and things that you can do on your own just to experiment with the technology and better understand it. So let's maybe talk a little bit about how do we pick winners and losers over time? We've come up with framework, but it is really important to think about that because every technological shift does create winners and losers, right? Think about what Amazon and the rise of e-commerce did to traditional retail. Think about what the release of the iPhone did to the Blackberry. Think about what the cloud did to your traditional data center and hardware vendors. And I think similarly we will see generative AI do that. And to that end, we've come up with kind of this framework we use for software, but I'm going to try and make it a lot more broadly applicable to really any company or really any entity and talk about the four categories of winners and four categories of losers on the winner side.

(05:46)

The first category is really the large incumbents who have a big advantage around data and distribution that they can leverage the reach they have, the existing customer base they have, as well as all the data they've gathered. Obviously data is a really tricky topic. Who owns that data? Who has the right to use that data to train models to build generative AI systems? But in general, we do think these are major competitive advantages. Look at Microsoft as an example of this. At the same time though, we think large companies have to move with a sense of urgency that they're not used to. Microsoft again, I think is a great example of one that has been able to, but large companies, large entities are not known for being nimble. They're known for bureaucracy. They're known for many, many layers of red tape. Just the fact that 70% of enterprises have actually banned the use of Chat GPT on corporate devices tells you how much bureaucracy there is and how much of a hold there is against what large companies can do.

(06:43)

But again, the data and distribution can be a major advantage. The second category of companies that we think have a real advantage when it comes to generative AI is those who have vertical or who have domain expertise. One of the great things about generative AI is you can do anything with it, but that's also one of the pitfalls of it. It is a blank canvas. It doesn't speak the language of specific industries, and we really think that entities that have, or companies that have that domain expertise that can teach these large language models to speak the language of a specific industry and get the model 70% of the way there can be at a huge advantage. Think about a finance specific LLM that understands DCF stands for data class, discounted cash flow and not data classification framework, which would be more appropriate for security or think about how valuable a government specific LLM that understood all the nuances and all the lingo of the industry could be.

(07:40)

The third category is actually the mid-market challengers. These are companies that we think can actually act more nimbly and go against the kind of fat, lazy incumbents that are in a market leadership position, but are slow to adopt generative AI and think about how much they can disrupt them. I would, again, go back to the Amazon analogy, right? Amazon was an online bookstore and now it's the everything store and also one of the biggest software companies in the world simultaneously. And they were able to use their nimbleness and their innovation and take down these giants in retail commerce and even software and internet and take on a pole position. And we think the same thing can happen with a lot of these mid-market challengers. And then the fourth category is actually any company that enables others to have a generative AI strategy right now, this is a narrow list.

(08:29)

It does include some of the obvious candidates like Microsoft, like Nvidia, but there's probably even consultancies out there that are helping companies come up with a real generative AI strategy. How do you embrace that technology while still being safe, while still having protection around your data? Now, let's get to the four categories of losers. And I was told to be controversial, so I'm not going to hold any punches here, but I do think there are a lot. Number one is exactly like I talked about, the fat, lazy incumbents that are either slow to move or it's companies that just have architectural issues because they're old. They have all their stuff sitting on premise. They haven't embraced the cloud, let alone generative ai. And we think that's category number one and two is lazy as well as legacy. But this could be companies like Cisco, this could be companies like Oracle, SAP.

(09:21)

These are all companies that are just very slow to move and that they are stuck with a lot of technical debt that all their stuff is sitting in data centers and it becomes really hard to leverage what they have with generative ai. They've also bought 50 different companies over the past decade, none of which are integrated. Again, that becomes a huge competitive disadvantage for them within the industry. You can think about this is something that we debate every day. There are banks that are like ours, thankfully, that are embracing generative AI, but there are a lot of banks out there that are I think going to be very slow and maybe won't even think about it for two, three years, and they will be in our minds at a competitive disadvantage. The third category is companies that have maybe benefited from being associated with AI.

(10:04)

That's a charitable way of calling them. What I actually refer to them is fake AI companies, companies that have convinced CIOs that they're doing artificial intelligence when they're not. Think about companies like Palantir, for example. They're not doing ai. They claim they are, they are not. All the industry checks you do out there will tell you that. And I think not only because generative AI is so much more powerful, but also because it's democratized. I think it exposes the weakness of these fake AI companies. And then the fourth category of losers is what I actually call AI roadkill. These are companies where basic functionality gets completely replaced by generative AI. Think about companies like Asana that do basic task management, online travel agencies like Expedia, Priceline, do you really need all that stuff when generative AI can serve as your travel agent for you, there's software companies out there that create legal contracts for you.

(10:59)

Again, generative AI can do that, so maybe that entire company isn't even necessary. And again, same thing happened with the internet. Same thing happened with cloud. Same thing happened with mobile, and we think the same thing is going to happen here with generative AI. I also want to talk a little bit about generative AI monetization. And again, I think this is very broadly applicable in terms of how we think about the framework and really there's direct and there's indirect monetization. But make no mistake, every single company, every single entity needs a generative AI strategy. Just like every company in the nineties and two thousands had to have a website. If you didn't have a website, you eventually kind of became roadkill. Of course, having a website did not make you a tech company, but that's a topic for a separate day. You can think about ways to monetize generative AI directly.

(11:48)

There is, Hey, we have all these fantastic generative AI features. We're going to discreetly charge for them like Microsoft is doing with a lot of their co-pilots that they've announced. Or maybe let's bundle all these advanced generative AI capabilities and put it behind a paywall, start to charge people for it. But there's also, and maybe even on a consumption model where you pay for the features you use, but what I think gets even more interesting is the indirect monetization. How do I use generative AI and my early use of it, my early adoption of it to be a competitive advantage versus others? How do I use this to improve my win rates? How do I use this to improve customer satisfaction, customer stickiness overall, how do I improve my lifetime value of my customers and maybe even lower my cost to acquire the customer? These are all things that I think companies should be thinking about.

(12:37)

The indirect piece, it's hard to quantify, it's hard to measure, but I think it will be really, really important. And again, I think generative AI can be a major competitive advantage. I'd be remiss if I didn't talk a little bit briefly about the impact of generative AI on profitability because that's what we all care about at the end of the day is how do we make more money, right? And generative AI is really expensive technology. My estimates, it costs five times as much as a traditional cloud workload, although that will come down over time. So there is an impact on your gross margins and expenses there. However, there are a lot of efficiencies that you can get out of generative AI. Your marketers can go send 500 emails instead of 50 in a day using something like chat GPT. Your software developers can now be two times, three times five times more productive and write so much more code.

(13:31)

In fact, 45% of new code today is already being written by generative ai. So your developers become more productive. You can come up with a lot more features and capabilities than you could in the past, and this extends to every industry. I was talking with a friend of mine who's a cardiologist, and he said, look, once generated AI gets to where I need it to be, I can read 20 EKGs, 200 EKGs in a day instead of 20, which means I can diagnose way more patients than I ever had the ability to do before. And so if we put all of this together, I actually think you're long-term profitability being enabled by generative AI can actually be way higher because there's so much more efficiencies that comes out of this. I also think this is really important when we think about productivity at our own jobs.

(14:15)

I mean, I gave some kind of examples. I'll give a few more out there, but what I think maybe even in my industry or my job, I can use it both on the professional and personal level. On the professional side, I already use generated AI today to help me rewrite things that I'm working on, notes that I'm working on. How do I make what I'm saying more clear to help draft emails to clients and to companies based on information I have, based on my writing style, based on publicly available information about that company. When I'm having conversations with a brand new company, that is my starting point for research now is what should I know about the company? Who are their biggest competitors? What are customers saying about them? Maybe even here, give me 10 suggested questions I should ask and build off that these are amazing time savings and I'm so much more productive and better at my job as a result.

(15:05)

But the great thing about this is all this technology can be used in your personal life. You can ask chat GPT to act as your travel agent. Say, Hey, look, I'm going to go to Paris for three days and I like history and find wine. Build me a custom itinerary, and it will do exactly that. That will save you 20 different Google searches that you would've had to do and scouring the websites and it's all custom tailored to you. You can have it be your sommelier. I'm a little into wine in case these examples didn't make that obvious, but hey, here's the a hundred wine bottles I have in my cellar. Now here's what I'm going to eat for dinner, or here's where I'm going out for dinner. Please give me a recommended bottle of wine that you think would pair with this. These are, I think, amazing even personal use cases that all of us can use with generative AI.

(15:49)

And if you have not had the chance to use chat, GPT, highly recommend just playing around with these sort of things, having a conversation like you would with any expert in the industry. And I guarantee you, you will be very surprised by how good and how accurate a lot of these results can be, and this is just scratching the surface of what we can do with generative AI. Lastly, I really just want to talk about the societal implications of this. As you can tell, I am incredibly bullish on generative AI's technology. This is game-changing technology, but it will have massive societal implications. Everyone's having the conversation about is generative AI going to take jobs? And it's a very, very complicated answer to that question. The way that I'm thinking about it right now is, yes, there will be some jobs that maybe become a little bit obsolete, and I'll get to that in a second.

(16:38)

But the bigger thing is within every job, within every industry, within every sector, it's going to widen the gap between the top achievers and the bottom level. If we think about software developers, for an example, if you are the rank and file software developer who has just listened to your pm who is saying, Hey, look, I need you to build this feature in this piece of software, your job probably isn't as important anymore because tools like GitHub copilot can write that code for you. On the flip side, if you're a very visionary developer and you're saying, here's ways to improve the product, how do we improve accessibility? How do we surface new features to users to actually drive more revenue over time, you become more valuable if you're a lawyer, but all you're doing is drawing contracts or you're that stereotypical paralegal who's spending 10 hours in the law library trying to find precedent, court cases generative, I will probably replace your role and make you less valuable if you're the courtroom lawyer, if you're the Harvey Specter from suits, you become more valuable.

(17:41)

The places where relationship building, where specific domain expertise, where people where manual interactions are important, I think become more important because a lot of the manual work can be automated and you get more time to spend on where you're differentiated. I think one thing though, if we do think about jobs that get replaced, the thing that makes generative AI so unique is, as you can tell, these are white collar jobs. We're talking about all the previous industrial revolutions. Were replacing blue collar jobs, and that's why this is so different from the invention of the printing press or automating factories or anything like that. And it's not even just knowledge workers, it's creative that can do this, right? You think about all these image generation systems out there that are actually starting to threaten artists. Think about Dali Mid journey, stable diffusion, Adobe Firefly. You may have heard some of these.

(18:33)

You literally type in a very specific prompt and it will generate art based on exactly what you said, witness the writer's strike that's going on in Hollywood right now. I mean, you can go to chat GPT and have it write you a full fledged movie script with the right sets of prompts, and that's part of why they're striking is to stop AI related content and stop AI from taking their jobs. So this is everywhere. It's white collar jobs, it's creative, it's knowledge workers, so there's potential disruptions everywhere. And I will say alongside that, it is up to us in our jobs to figure out how can we be more differentiated? How can we lean more on the relationship side, but it's also up to society as a whole. How do we re-skill and up-skill everyone? How do we help people adapt to this new economy?

(19:19)

How do we get people to be AI enabled in their jobs, in their careers if they haven't been? Or how do we help them find new careers? We were at dinner yesterday and I was talking about the example of what some of these oil-based economies like Saudi Arabia, Qatar, UAE are doing. They're all investing heavily in technology to retrain and upskill their global citizens for what a post oil economy would look like. And believe it or not, one of the biggest innovations in AI today is actually coming out of a university in Dubai in the United Arab of Emirates. And I think that's a great example of where society getting involved in this upskilling and reskilling can be actually really beneficial to minimizing the negative impacts and maximizing the positive tailwinds that we're going to get from generative AI. So putting this all together, look, I think generative ai, it's super important.

(20:16)

It is really, really exciting. I don't think I've felt this excited about the impact of technology since the nineties and the advent of the internet, and I was a little kid back then and I was still really excited. I'm probably double or triple more excited now because we're in the middle of all this and we have the chance to actually shape how it's being used and what this looks like over the long term. So I think it is really exciting. I think there's going to be huge benefits as a whole, this society. I know there's always going to be concerns about things like job disruption around data privacy, around how do we avoid hallucinations? But again, I think when this technology is used responsibly, when it's used well, and when we're using it with the long-term in mind, it is going to be a huge benefit really to society as a large to all of us in our careers. People can spend more time pursuing what they're actually passionate about than doing road tasks. People can be more differentiated. We can be so much more productive and maybe we'll even have more free time. So that's where I get really excited about all of this. Thank you everyone for listening to me. Appreciate it.

Brian Olson (21:32):

Thanks, Rishi. Good morning everyone. I'm Brian Olson with RBC. Before I introduce the panel, I'm going to try to talk a little bit more about the people aspect. Rishi was focused a little bit more on the companies and then segued into the people a bit towards the end. So we're going to do a demonstration on chat GPT for those of you that have not tried it yet, but before we do, could you raise your hand if you think that a lot of white collar jobs will be displaced by artificial intelligence? Okay, so if this were the house, we wouldn't have to vote again, the house represents, so do we have any English majors or literature majors out there? Okay, give me a book that's really difficult to understand. Which one?

Audience Member 1 (22:20):

James Joyce.

Brian Olson (22:22):

James Joyce. A particular book of James. Which one?

Audience Member 1 (22:25):

Ulysses.

Brian Olson (22:26):

Ulysses. Julie, could you write a 500 word essay, literary essay on James Joyce's Ulysses? I don't don't know how to spell Ulysses either. This is an a plus by the way. No, it is. I've played around with this. It's amazing. And interacted with literary professors on stuff like this. So you can see it took about 10 seconds. Anyone want to change their vote that didn't have their hand up?

(23:10)

So, can we turn the next page? I think you can keep doing it. You can keep revising it, make it shorter, a hundred words, a thousand words zero in on certain aspects of the book. So it's a taste of what's coming. From a societal standpoint. I'm a big fan of Eric Hoffey. He wrote a book published in 1951 called The True Believer. It's a study of mass movements and revolutions, and he sort of uses those two words interchangeably. Revolution is a more extreme version of a mass movement, and he had two points in that book that I think are relevant for the discussion today. He said, the first thing is a lot of people think that revolutions or mass movements cause change in society, and he flipped that on its head and he said, actually what happens is there's change that happens first in society and that lays the seeds for a mass movement or a revolution that's captured in this quote here.

(24:04)

Next please. The second thing that he said is he did a good job of describing the agents of change, excuse me, agents of mash movements. The conventional wisdom, sort of the Hollywood stereotype is poor people have had enough. They grabbed their pitchforks, they charged the castle and cause the revolution. He said, that's not what happens. Poor people are too busy trying to survive. So he has this concept of the new poor, and those are people that used to have status, they feel disenfranchised, they're, and those are the ones that are more likely to mobilize and seek change. They've had something taken away throughout history. This has been the path that the humans have taken recently. This dynamic, I would argue, played out in the rise of populous movements like MAGA. MAGA was a populous movement. MAGA tapped into the resentments of blue collar workers that were displaced by globalization.

(25:02)

And I would argue that if AI does the same thing in displacing a lot of white collar workers, white collar workers will have a mass movement. I don't know what that will look like, but it will most certainly happen if there's a lot of displacement. If AI does not displace workers and makes everyone much more productive, creates a lot of wealth, then we're all going to work three and a half days a week as I think Jamie Diamond suggested that. So that's a utopia scenario, which I think would be great. Next page. So the question is, AI will never replace human judgment. You may think your job is safe or our industry is safe because we use our brain power. Rishi alluded to this. Machines replace muscles. This is replacing brains, and we're at the very beginning stages of it. Next page. So this is a pace of change in technology that we're used to.

(26:01)

This is basically a measure of the speed of apple chips goes up about 20%. Every generation sort of moderate increase. The pace in AI is exponential. So the graph, it's so fast that there's nothing there really. It's 117 million parameters in the first version of chat, GPT, the most recent one's, 110 trillion trillions, billions gazillions. Hard to understand. If you go back in time 117 seconds, that's 2019 pre pandemic. If you go back in time, a hundred trillion seconds, that's 3.2 million years ago. It's when Lucy walked the earth. It's the chain. And this is the model T. This is just the beginning of it. So replacing human judgment, I would argue is probably going to be on the agenda. So now drilling down into public finance, we're in California, so we go with the Hollywood theme, the good, the bad, and the ugly, the things that I can see on the good.

(27:07)

An example, cyber attack prevention. New York City is using AI to greatly enhance its cyber attack defenses. Municipal governments, as we all know, typically don't have best in class technology and AI could help them close that gap in technology that they have with the hackers. The ugly. We sort of talked about that job displacement, that's going to require a lot of retraining of older workers, a lot of social services for those that have been disrupted the ugly. This was a thing that I read recently I thought was a very interesting take, systemic bias. So imagine a highway patrolman who maybe has a systemic bias on who they pull over. They can impact hundreds of people, and the bias may be difficult to detect. The officer may not even know that they have the bias. Well, if we switched over to AI deciding which cars get pulled over first, if the algorithm has a systemic bias in it, it could impact potentially millions of people.

(28:09)

Everyone in a state, for example, the good news is the algorithm is visible to policymakers so they could see if there is a bias there and correct it. I would argue that that process would be pretty messy. So public finance is definitely going to be touched by AI. And now we'll go into our panel. We've got three panelists today, Alex Domenick is a Vice President on the municipal, electronic and algorithmic trading and underwriting team at RBC capital markets. We have Janice Hofferber is head of US Public Finance at Moody's, investors overseeing ratings and research for thousands of state and local governments, hospitals, universities, charter schools, and other tax exempt obligors. And Matthew Smith is the founder and CEO of Spline Data providing real-time, quantitative pricing and yield curves for the municipal bond market. By the way, full bios are online chat, GPT condensed those bios down into one sentence, by the way. Took me four seconds. So we will start with Alex. Alex, can you talk a little bit about what RBC is doing with ai?

Alex Domenick (29:19):

Yeah, sure. And thank you, Brian. So we talked a lot about AI's potential in displacing jobs. This is an area where our market, the municipal market, is already using AI to solve a problem that started arising maybe a few years ago when with the rise of separately managed account institutional investors. We currently, with our AI pricing, we can bid on thousands of line items at one time. And this is an area where a human trader could never bid on the same number of line items. And this part of the market really increases the efficiency in secondary trading of municipal bonds. This will help broadly with bond pricing liquidity and increase the efficiency in our market. I heard Brian talk about something. Machinery is a muscles, AI is the mind. I think this technology is not replacing secondary trading in the municipal market on the quality side, it's more on the quantity side.

(30:48)

So the quality of our bid numbers and pricing and municipal bonds, secondary trading, it's not that the AI can price them better than a human, it's just that we can repeat that pricing thousands of times where a human can only submit it one time. So it canvases the market in a much broader sense. So if you look back before this became kind of a prominent presence in our market, if I wanted to sell a municipal bond in the secondary, you might get one or two bids, now you get six or seven. And these are institutional investors just providing liquidity. And I think this is a dynamic that we'll definitely continue. If you go on your brokerage account right now and you try to buy a small odd lot piece of municipal bond, it's probably purchased by someone like our algo, hopefully ours. So I'll turn it over now to Matthew Smith to talk about a little more in depth in how that pricing process worked. He is also in this space.

Matthew Smith (32:09):

So yeah, like Alex said, our company focuses exclusively on pricing municipal bonds, and this is in the secondary market currently. We also produce yield curves, but we operate in the space of AI and that we use machine learning and absolutely everything that we do. A lot of people think of AI and they immediately go to chat GPT, but machine learning is like the building blocks of chat GPT wouldn't say that I'm the foremost expert on generative AI by any means, but it's kind of cool how it works. It is almost like they're Ping different models to try and predict the next word and the answer. So it's not like chat GPT understands English in a human sense or understands the intricacies of the Muni market. It's just saying with this data that I have in my training set and somebody asks about how much does California issue in terms of debt every year, it could go reference back to that and say, okay, this is the most likely answer to that question.

(33:29)

And that's kind of why it has. Like Rishi mentioned, hallucinations, it'll give you answers that are sometimes very confidently and correct based on some bad training data that it might have. But anyways, we focus on the machine learning part of that. So the same models that are predicting the next word in that response, and I think people also kind of hear machine learning or hear a lot of these terms and they kind of glaze over. But machine learning is linear regression is machine learning. As long as you teach a machine what linear regression is, you give it the formula and then you give it the ability to reinforce its own learning. So as it gets new data incorporates that into the model and improves the fit, improves the regression, improves the results with the more data that you have.

(34:24)

It used to be the new wave. Now I think it's kind of generative ai, but I think you'll see machine learning in general be kind of a bigger focus as the access to how to build machine learning tools becomes more democratized through chat GPT was shocked when it came out. I was an autodidact myself, so I learned how to code just through various online resources and it was a slog, it was a pain. But now if I have a question about writing code or how to implement a certain model chat GPT makes that very readily available. And as soon as two years prior that kind of information wasn't available, it was very tightly held by some of the PhDs and the master's degrees in the world of applied mathematics. And you'd mainly only see it at Citadel and everything. But now anybody that knows how to type can access that kind of information and integrate that into how they're operating in their day-to-day job.

(35:45)

Sorry, I'm kind of jumping all over the place, but I think the topics of, Brian and Rishi hit on these both of inequality, but also the democratization of information is really kind of the defining piece of generative ai. I think this differs from previous revolutions and that you don't know, you don't have to know how to code to access this kind of information, and that's what makes it so special. I think I'm kind of the opinion that we're not going to see massive job displacement all at once. I think that that's kind of going to evolve over time. I don't think many people are to be fair, but I do think that you're going to see everybody's job kind of change and people become a lot more effective.

(36:44)

Kind of harken back to my first job, I was working at a bank, I was doing back office work, but every single day I would get the same spreadsheet and do the same thing to that spreadsheet over and over again. And I was so tired of doing that and I just wanted to get it over with and done with. So I unplugged my mouse from my keyboard or from my computer and force myself to use my keyboard to get faster at Excel. That was the best decision I could have ever made for my whole career going forward. And it was because I wasn't just resigned to the fact that this is my job. I quickly finished all of my work and I was done within the first 30 minutes of the day and I had all of this time to learn Python and say, okay, now how do I quit doing Excel? Right? So I put it in Python and it snowballed from there. And now by God's grace as I'm a quant somehow I think that the key point, if you kind of want to survive this wave and you want to thrive in this wave is don't be a Luddite. And I think the bar for not being a Luddite, it's now so low because all you have to do is know how to type a question that it's so easy to avoid getting in that job displacement scenario.

(38:13)

Speaking to the trading side a little bit, and I don't know how I'm doing on time. We focus on secondary, but I think that there are big implications for quantitative pricing in the Muni market as it pertains to the primary market, particularly because secondary pricing and secondary trading I think drives a lot of activity in the primary. So let's say that spreads compress 20 Bips, everybody has an algo. That's definitely going to translate into how a new issue is priced. So primary or sorry, negotiated or competitive deals. And I think we're going to start to see that first in the competitive space. I think you're going to see a lot more non-traditional Muni participants enter the competitive underwriting space just because it's kind of the democratized access version of underwriting. And you'll see it in small deals because as people build these algorithms and build optimizers to price these deals, issuers are going to see a lot better execution. And at the same time, it's going to be a lot easier in terms of software available and tools available to raise small deals.

(39:32)

And yeah, I think you'll just see a bunch of algos kind of enter that space and it'll kind of bubble up into lowering the cost of debt for issuers. And I think that's where it all kind of boils down to it. That's kind of the point of coming up with all of this pricing and making the market more efficient. Our goal is to create a pricing product where people can almost plug and play and become that algo and then they can build on top of our pricing just like some of the existing evals in the space and outcompete their components with domain knowledge.

Brian Olson (40:16):

Thank you, Matt. So Janice, what we'll do is we'll have Janice talk a little, and then if you all can think of questions you have, we we'll skip my questions and go to yours. Janice, at Moody's, what are you guys doing in terms of resources and how is it affecting the workflow at Moody's in the rating agency space?

Janice Hofferber (40:34):

Well, thanks Brian. Thanks RBC for having me. I'm coming at the end of this presentation and as you can see, these are pretty tech savvy speakers. So I've been asked to come talk about what Moody's is doing in the space of AI, and I have to start by saying after 17 years of being at Moody's, I haven't seen our company move as quickly to adapt and learn as fast as we have in the last six months. It's been really a remarkable shift in the way we do our work, our mindset about our work, not just at the rating agency, but also in our sister company, Moody's Analytics that provides products and services to the market. Even the most risk averse of us are pretty excited. And so Rishi, I share your enthusiasm. We're excited to use AI and we think it really fits perfectly with our value proposition.

(41:24)

We provide analytics, we provide data, we provide software, and of course we provide ratings and opinions to the market. And so this collection of talents and capabilities really puts us at the heart and gives us the right to participate in gen AI, the reliability of our brand. We really think it stands for being a trusted thought leader. And then when you combine that with the richness and the depth of the data, it puts us in a pretty unique, or at least within top providers to take advantage of AI. So I'm going to talk a little bit about what we're doing externally as well as what we're doing internally in the rating agency. Let me start first by how we're addressing AI with our external customer proposition. We know, and I probably alluded to this, that our customers are really concerned about the reliability and the validity of the data that they use to make decisions.

(42:23)

So recently we developed something called research assistant, which I think actually has sort of ironic that we called it research assistant, but it just literally launched in the last few weeks. This is coming out of Moody's Analytics and this provides on-demand curation and access to a wealth of Moody's information and knowledge. And this includes over a hundred years of ratings and opinions, but it also provides customer access to all of our proprietary data sets. That's whether it's on climate or real estate or ESG. And using the research assistant AI function, investors and issuers can find the data they need, put it into a form that they can digest, whether it's pros or a chart or a table, and really facilitate their ability to do their own analysis very quickly. In fact, I was recently at a conference that we held for all of our employees and stakeholders and learned that Moody's has the deepest database on corporations and institutions in the world.

(43:27)

Remarkably, and I think it's remarkable because I'm a numbers person. We have data on 417 million companies. That just blows my mind because as an analyst and I covered 30 credits, I could barely keep up. So the idea of having information on that many institutions and combining it with the human interface that our data can leverage, you definitely can see how using AI can generate better insights and faster decisions. So another use case for Moody's Analytics is they're using AI for their insurance clients and using AI to evaluate climate risk and manage any kind of extreme event, whether it's weather related or wildfires. And you combine the data that the insurance company has with the data that we have at Moody's, even after an event, to be able to very quickly understand the exposure that they have in a particular market. The important thing here, and I'm not an expert, but we can overlay our data with the client's data and use through a Gen AI tool, allow them to really have a very secure and safe way of evaluating risk internally.

(44:47)

And that's where I spend all of my time. Obviously, we're really unlocking the opportunities of AI through our strategic relationship with Microsoft. So Rishi, I didn't realize that what we call copilot was actually the name that everyone uses. So you taught me something, I thought it was a Moody's branding, but what do I know? But through that relationship that we started earlier in the year, we are now providing generative AI access within a teams environment to over 14,000 employees. And that includes a powerful group of analysts of almost 2000 analysts around the world. The system co-pilot, which we all know and love now, is on everyone's desk in a very safe and secure environment. In fact, I think all of us are getting tired of hearing our CEO talk about how we are 14,000 innovators, or probably better said, a prompt engineers, which is a new term that I learned also pretty recently.

(45:44)

I know you're laughing at me, but Yes, you are, but we're really, yeah. Okay. There was one word you said I have to ask you about later on, but maybe others have the same feeling. But we're really challenging all of our employees to develop, to develop new use cases, and we're really championing them in the company across the range of analysts and assistants and managers really to come up with a new use case. And so what are we doing? Our primary objective is really to drive productivity, to eliminate time consuming tasks, not to eliminate people. So just the few areas that I'll mention on the ratings and research side, we're helping to write research outlines to edit our analytical writing. This helps to expedite our research. We're using it to produce charts and graphs and really be able to get our view out to the market as fast as we can with the analytic workflow.

(46:49)

We're using it to schedule meetings or just manage administrative tasks, and we're even using it for some knowledge acquisition to pull data together in a very quick, quick and comprehensive way. So there's obvious benefits, right? We've talked about the better insights and faster, but I'd be remiss in not pointing out that it also helps to reduce risk. We are very human centered business and we are regulated, and we have quite a few processes that need to be very well documented. And accuracy really matters, right? We don't have an opportunity to make a mistake. So gen AI can help us check and double check our work, and we're really training everyone in our business to really learn how to ask gen AI the right questions and use the technology as a tool, but we're importantly getting them to focus on the human elements of their lived experience, the intuition that they bring, the empathy they bring, right?

(47:51)

And the aspects of human cognition that a system just can't replicate. I mean, maybe at some point down the line, probably I will be long and gone, but at this point, those are the key. That's the secret sauce to the work that we do. But we're not putting our heads in the sand. We know the technology is going to move faster than we can understand it, and we have to be very mindful of the risks that it can present. So we've put in place very robust governance and codes of conduct frameworks. We're mandating multiple hours of training on AI for all our employees. And so while we want people to experiment with it and come up with different use cases, we're also very focused on setting out ethical guidelines and accountability and compliance. So that's really what's consistent with the Moody's brand. So we're very focused on that.

Brian Olson (48:41):

Thank you, Janice. Alright, we have time for one or two questions.

Audience Member 2 (48:47):

I have a question maybe for Alex. I'm just wondering if you've talked about how AI could really increase the volume. I don't know if this is hot, increase the volume of trading. So if there's a, I think that Matt used the Brad training data. Does that take a pricing error and magnify it into a black hole?

Alex Domenick (49:19):

Not typically. I know, speaking from my experience with our algo, and this is true of others too, there are a number of thresholds that the model takes into account. So if a pricing prediction is way outside of a band that it also calculates on the side, then usually throws it out. So not to say that there aren't pricing errors because there absolutely are at times, but we try to avoid the tail risk situations and over time it's more efficient. But yes, that is something that we've definitely thought a lot about.

Brian Olson (49:56):

Anyone else? Go ahead.

Audience Member 3 (50:00):

I have a question for Rishi. So you obviously are excited about AI, know more than most of us in the room. So what is it about AI that scares you? What do you lose sleep over at night thinking about AI?

Rishi Jaluria (50:15):

Sure. Yeah. So beyond the whole pay, is my job just going to get replaced? I think the biggest worry I have with generative AI is it is an incredibly powerful tool. And if it's in the hands of the wrong people who have malicious intent, what does that do? Think about how dangerous stuff like phishing is, right? Where people pretend to be someone trusted and use that to extract information from you. You can use chat GPT to write a hundred of those emails a day instead of 10 and use publicly available information to make it even more targeted. I worry what happens when you get this technology in the hands of malicious actors and what can we see companies do to prevent that? I think a lot of these cybersecurity companies need to step up their game to help companies, to help people defend against how much more complex cybersecurity attacks can get as a result of generative action. So that's probably the biggest thing that worries me right now.

Brian Olson (51:14):

One more.

Audience Member 4 (51:20):

Thank you. And this is to the group. I'm just curious, you talk about job destruction with the technology coming out. Can you talk about the opposite? What about the job creation and also is there any studies or numbers? That's the first question. The secondly, is it in fact true that the Bay Area is really the leader in ai? Is that going to be the next frontier for Silicon Valley?

Brian Olson (51:41):

One quick thing. Last night at dinner, Rishi said that he thought that AI would separate the haves and the have nots. So the people that are very talented, creative, whatever, high end will do much, much better. The Luddites suffer. I heard that last night.

Rishi Jaluria (51:57):

Absolutely. And I mentioned that at the beginning. Maybe I can just kick off since started that. Look, I think there will be a whole new category of jobs that get created out of this. Even right now, there's a lot of generative AI specific jobs that are out there. The term prompt engineer is a new one, and it's probably a job that will last for two to three years before it becomes obsolete, right? It'll be like the next generation of Y2K specialists, but that is a new job title that didn't exist a year ago. How do I train to have people train the right prompt to ask the model to get the exact answers I want? But longer term, I think there will be a whole new class of AI enabled jobs that will get created and a lot that we just don't know about at all today.

(52:41)

So absolutely, I do think that there will be a ton of job creation. I'm glad you brought up the question around, is this the next frontier for Silicon Valley? Look, so I live here in San Francisco. I'm actually a 15 minute walk away from where we are right now. I walked here this morning, not only is generative AI part of my day-to-Day conversations in work day-to-Day conversations anywhere I go here in the front of it in San Francisco, you go out to a bar to walk to a football game. People are talking about generative ai. You go out to dinner people talking about generative ai. That's how it is. And for all the worries that the press has about San Francisco, I think AI and all the innovations that's happening around generative AI is going to bring it kind of back, right? San Francisco is a perpetual phoenix that maybe seems on decline and comes out of nowhere and comes back up. And I absolutely believe all the innovation that's happening in generative AI is going to maybe even save San Francisco.

Brian Olson (53:38):

Thank you everyone. On behalf of RBC, thank you. Enjoy the rest of your conference.