Carmen Recio and Sergio Gago, from Moody’s Analytics

Carmen Recio and Sergio Gago, from Moody’s Analytics, are interviewed by Yuval Boger. They discuss the intersection of quantum and AI, the importance of quantum computing in financial risk modeling, machine learning, and optimization, the need for businesses to stay ahead of the curve by understanding how quantum computing integrates into existing processes, Moody’s role in bridging the gap between quantum companies and end-users, and much more.

Full Transcript

Yuval Boger: Hello Carmen, hello Sergio, thank you so much for joining me today.

Carmen Recio: Hi Yuval, thank you so much for having us

Yuval: It’s my pleasure. So, so Carmen, who are you and what do you do?

Carmen: Well, I like to define myself as an applied researcher in quantum computing. And right now the field that I’m applying it to is the financial services industry. I work in the Moody’s Analytics team together with Sergio and recently I started also as the team lead. So that’s basically my role.

Yuval: And Sergio, who are you and what do you do?

Sergio Gago: So I’m really happy because this is my second time in this podcast, right? So it’s great to be here again and hopefully have not just the same audience but more, I know this has been growing nicely. My name is Sergio. I’m Spanish. I am a CTO by trade, turned quantum computing specialist, turned AI specialist, turned everything that’s on hype, if you will. About a year and a half ago, I kickstarted the quantum computing team at Moody’s and Carmen started working with us right after. And today I lead, I am the managing director of both AI and quantum computing at Moody’s as well. Yeah, trying to bring the benefits and the adventure of quantum into the finance industry.

Yuval: That’s the first time I hear this title, the director of Hype. So congratulations. Let’s look at it. And if I can start by talking about hype, there’s obviously a lot of hype about ChatGPT. And you’ve been at Quantum for a while. Do you feel that ChatGPT or AI in general is stealing the spotlight or more importantly the budget from Quantum?

Sergio: I think that is the case. We talked with many financial institutions, many banks, insurance companies, asset managers, and the likes. And in general, with all these technologies, there’s a lot of misunderstanding on how they pair up together. One question that I get a lot is, “So quantum computers are going to power up chat GPT?” And of course, the answer is absolutely not, and they live in two different horizons. Who knows what the future can bring with quantum machine learning and quantum AI and some from the lab, but for now that’s only something very, very far away.

Now what happens is that in many cases, anything that’s horizon two or three tends to fall in the same innovation teams who are typically limited in resources and the number of hands that they can put on deck. And ChatGPT has taken the world by surprise. So, transformers have been around for almost four years now, but it was only since the last September or so when the world started seeing that. The biggest difference that ChatGPT brought into the world is that businesses could understand the use cases clearly. They could just start using it and immediately connect the dots and start figuring out how we could improve their businesses, their processes, internal or external. Whether it was difficult or not, because on generative AI, it takes 1% of the time to build a super nice proof of concept that blows everyone’s minds, but then the other 99% of the time is taking that into production. So we are seeing that some of the teams that were traditionally working on AI have switched part or a big part of their dedication over to generate from quantum to generative AI.

In some cases, the argument was, “Well, you guys know about algebra, right?” So this is pretty much the same thing. With one exception, those teams that had already prior AI centers of excellence and projects going on, these places kept their separate teams. We’d be trying to do some cross-pollination, but smaller banks, for example, T2 banks and asset managers, for them has been a massive switch that of course has made an impact on budgets and availability.

Yuval: What does Moody’s do, and specifically why would Moody’s need quantum computing? What is the product that you’re trying to generate with quantum computing?

Sergio: So I’m going to try to take a first stab at that and Carmen can get into the more real answer of that. The business of Moody’s is data and generate insights and signals on top of that data, of Moody’s analytics specifically. Moody’s is divided into components, Moody’s investor services, that’s the ratings agency that most people now are not aware of, but then there is Moody’s Analytics where we basically play with an enormous amount of data that’s available to ask them to enter our clients. And we do a lot of computation with that data.

Think about most of the regulatory requirements that financial institutions have, which are from simulation problems, multi-cables, and so on and so forth. Of the use cases that quantum computing promises and finance affect directly some of the products that we have in the market, where today we excel in classical ways. So for us, it was a no-brainer to see a technology that could potentially disrupt, improve, or complement the classical solutions that we have today, regardless on the timeframe for that to happen.

So whether it was machine learning improvements, simulation improvements, optimization improvements, or full optimization, for example, we can get people into the use cases that we work on. For us, we turned every stone from an innovation perspective. We evaluate this thing in 10 years’ time, in five years, in three years, whether it is quantum, generative AI, or blockchain, the same way that it was cloud computing a long time ago. So, it was a natural fit for us to create a team that deeply understands the domain of how we’re doing those things.

This is a company full of quants. Not quantum, but quants. And many of them are physicists who in the ’80s or so went into finance because that was where the bucks were and they’ve been having fun ever since. So, we are paying out that very nicely. And for us, the question is, when is this truly going to bring a pan of thought? We just want to be right there at the forefront whenever it happens. I don’t know, Carmen, if you want to expand on some of the use cases that we work with.

Carmen: Yes, absolutely. So well, what Sergio has explained, I think it’s one of the four pillars that we’re working with. And basically, this is the one that looks at use cases and that looks at how to enhance the current products that we have. But we’re also focusing not only on the opportunities that quantum brings, but also on the risk that it brings with it. And this is why we are also making an effort to understand what’s the appropriate response to the thread that it represents to cryptography.

So we’re working on coming up with a post-quantum crypto strategy. And that’s something that we have been working on for many months now. And actually, we are on track working with our cybersecurity teams and understanding what new protocols will need to be implemented. What actions should we take for that? Should it be on us? Should it be on our vendors? These are the kinds of debates that are going on. And besides that, of course, an inventory on all the data that is important and the current way to secure it.

There are two other pillars, one of which is engaging with the community. Because nowadays, since this is such a new field, well, not new, but it’s a field where there is innovation happening very often. So we need to keep track of what’s happening and we need to be close to talent. There are many reports that mention the talent gap that exists in the industry. And this is why we work with universities that have master’s programs or research groups that are working on this topic. Also so that we can work with the students, that will be the experts of tomorrow, but also stay close to research so that we can influence a little bit, that is, filling in the gap between academic research and real-world applications.

So some examples of activities we’ve done here are two hackathons that we did last year. So for instance, we did a hackathon in ETH Zurich that was organized by the Quantum Engineering Commission. That is a great student association full of ideas and energy. And there we proposed a challenge on Quantum Monte Carlo. And the results were amazing. After that, we also did another hackathon in Barcelona with Barcelona Supercomputing Center, IBM. There was also another company that is called Qilimanjaro, which is a hardware company. And also, we established an alliance with Trinity College of Dublin which also has a program on quantum science, a master’s program. The members of the alliance are are IBM, Microsoft, Horizon Quantum Computing, Algorithmiq, and Trinity, of course. So these are some of the examples that we’re doing to stay close to both academia and the student community.

And the fourth pillar, besides use case discovery and product enhancement, would be thought leadership. So we want to be there — and in fact, we did some research together with Corinium. And one of the biggest gaps that we identified was training among the companies that are in the financial industry. So this is why we’re working on being there for our customers in their quantum journey. And we’re working on preparing training materials, but also publishing the research pieces that we do and pure outreach pieces.

Sergio: And I think if I can add, one of the main points that makes a difference between ours and your typical financial institution, we are not a bank, we’re not a known client. So we sit right between quantum companies and the end users. Whenever we talk with some of your typical partners, many of them have passed through this podcast, either hardware companies or algorithm companies and solutions, we sit right between them and the bank at the end of the process. And since these clients rely on movies, either for data or sub-leadership or advice and this, we cannot have this responsibility of understanding where these technologies are going. The post-quantum cryptography is one very good example of, all right, so what should we do? Should we wait until NIST provides the final encryption algorithms? Should we start playing and doing theosis? Should we participate in some sort of common group or something like that? So those things are very relevant in terms of creating a community with agnostic providers.

Yuval: I think when we spoke a few months ago, you mentioned that you did an internal use case discovery project and you came up with, I think, a couple of dozen use cases. Now obviously you’re not going to be working on all of that at once. Which use cases are at the top of your list right now?

Carmen: Yeah, actually part of the team joined in November last year and we took on the work that Moody’s had already been doing, working with the experts, with the quants that know the classical problems and know the struggles they’re facing today. But of course, as you say, we cannot tackle 24 use cases at once. And we also need to evaluate which ones were more realistic. For some, another thing that we’re doing, of course, since we are at a stage where quantum advantage is not here yet, it’s also important to understand what quantum resources you need for the advantage to happen. So even if you cannot predict when it’s going to happen, you can predict whether something can be done with quantum-inspired techniques that are run in classical supercomputing resources, or you can predict if something can be done in the near term with some error mitigation required, or looking at the longer term, what can be done when error correction comes, and we enter data of fault-tolerant, quantum computers.

So having all these into account and talking to our business units and the problems that they are facing today, we tried to match the literature surveys like the one that was published at the beginning of 2022 by Pistoia and others, where they did a very comprehensive survey on the status of quantum computing applications for finance. We are always following the literature, but we are trying to prioritize depending on the problems that are relevant to our business units.

So right now, for example, we are focusing on three big areas of use cases. The first one being numerical risk or stochastic modeling in general. The second one will be machine learning, how can it be enhanced with quantum? And the third one, is optimization. I can give you some concrete examples. For example, inside of optimization, one of the use cases where we are exploring the application of quantum-inspired techniques, like tensor networks, is the use case of optimization of portfolios. But also, well, we have also– as you know, we don’t just explore financial risk, but we try to provide our customers with an informed overview of all kinds of risks. So part of it is also climate risk. And this is another area where we are trying to solve dynamic risk models using quantum annealers. Another area that is big for us is simulation. So anything that has to do with Monte Carlo. And time series forecasting for different use cases.

Yuval: I’m sure it’s good for your customers to feel secure that Moody’s is thinking about that. And they can always turn to Moody’s for advice. But how do your customers think about quantum? Do they say, “Oh, tell us when it’s ready and until then we’re just comforted to know that you’re thinking about it?” Are they pulling to use it? And related to that, is there an explainability problem with quantum? I mean, if a quantum computer makes a decision, are there some applications where you say, “Well, I can’t really use it because I can’t explain what the black box is actually doing.”

Sergio: We see, on your first question, we see a lot of examples across the world. There are those who are heavily investing in quantum already. We know their names. They are very vocal about that and probably is really, really good research that helps the whole community move forward. But then the vast majority are actually oblivious to everything that is going on. That was a hypothesis that we had. We basically work with most of the financial institutions in the world in one way or another. But when you think about a bank, there are thousands of people, tens of thousands of people in that institution. To give you one example, sometimes we talk with the innovation team, we’re a quantum team of one specific institution, and then we send an innovation survey that we send to all our clients from all the different angles and perspectives. And that same institution tells us, “It would be great if our bank would be working in quantum computing today.”

So, there is a massive disconnect between innovation groups and the business side of the companies where there is no pollination between the domain, the use case, and the work that is being done. This is not a surprise and it happens also in other innovation or Horizon 3 technologies or even Horizon 1 like AI or Node 2. So yeah, we see a lot of divergence on what they believe about that. And that is why we wanted to create this survey that we did several months ago together with Corinium. And we published that report that shed some light on the realities of to what extent quantum is penetrating the technology industry.

There are of course several reports in the market today. Some of them come from quantum hardware companies, and some of them come from quantum software companies, but in general, a lot of those reports come from companies that have certain biases. We like it or not, right? At the end of the day, these companies rely on quantum happening to survive. And there are other types of companies like us, or like IBM, or Google, or Amazon, we don’t rely on quantum to survive. We have a business, and for us, quantum is a compliment that we believe will be beneficial in the future. So we really wanted to shed some light to this, as in, is it really true that, as some claims that we all have seen, most of the Fortune 500 companies are investing at least X amount of millions in the quantum? And we believe that that is not true, but the reality is a little bit gloomier in the sense that there is way less investment than we may think.

But it has a brighter side. That means that there’s a lot of work and a lot of things to be done in that sense. Most of the people that we discuss with feel that the industry is definitely not ready. So as you were saying, they’re waiting and seeing what is going on, which we believe is the wrong approach. It’s not enough to wait and see, and that is why we’re doing some of the things that Cameron mentioned before on AppSkate and your team, on getting ready, start learning the technology, and most importantly, making sure that you mix your domain, your products, your pipelines with what quantum can do.

So whenever we have a. We’re calling this today the chat GPT moment for quantum. Whenever that happens, you’re right at the forefront of that, right? And we can talk more about the results of this survey later. But then you have a really interesting point on transparency and interpretability of the models. It’s really funny because this goes hand in hand with all the conversations that we have with artificial intelligence and the regulations of AI, for example, in Europe as they come in very soon, and any type of model on whether you use generative AI or more classical machine learning models. That is fundamental, specifically speaking whenever you work with regulators. Your models need to be fully explainable, which means that a lot of the use cases that we have on quantum will have a big challenge with that.

Let me give you one example. A lot of the cases that you can work with when you evaluate a traded risk, relying on quantum Monte-Carlo and things like that, methods based on quantum actually to the estimation, They could potentially give you a quadratic speedup, as we probably all know, even though there’s a lot of challenges that we still need to figure out. But all it gives you is the final answer. You don’t have the granularity and the explainability of running your Monte Carlo, where you have all your fun scenarios ready for you to dig into that and zoom into those. So whenever we talk with our own people, they say, “Okay, fine, but how do I zoom into this part of the probability distribution or that part, or how do I change? Maybe the answer is not so much making the final calculation that gives you the last number of the reserves of your conditional value of risk, the expected shortfall for the regulator, but how do you get into those numbers? Maybe it’s a tool that helps the analyst read in the portfolio and make business decisions so then when you have to run that for regulatory purposes, you use your classical methods.

Incidentally, this is exactly the same thing that we do with artificial intelligence. We have a human in the loop that gets supported by AI, but then the final tweaks, the final decision, and the final validation relies on the human. And related to that, people are thinking about chat GPT and maybe soon if you write an article using chatGPT, it’s going to be watermarked in some fashion.

Yuval: Do you feel that that’s coming to quantum as well, that the API is going to be watermarked to say, “Hey, there’s a quantum computer in the loop?”

Sergio: That’s a really good question, but I don’t think so because, in my opinion, or at least for some of the use cases, maybe there are other use cases where this is not true, but most of the use cases, we’re not creating the problem. The problem is it’s a problem that takes ages to calculate or that we have to oversimplify. And the quantum computer is just the super duper abacus. Very super. But it’s a computation machine at the end of the day. So I’m not sure there would be some kind of signature except that you have been able to calculate it. That’s a signature good enough that you were able to compute that problem. But beyond that, it’s either a combination, a number, or something like that. There’s of course quantum machine learning and other techniques where maybe this makes more sense, but I don’t know, I would have to think about that.

Yuval: Now, Carmen: Sergio mentioned that quantum will be useful in the future. When is the future in your mind?

Carmen: Well, that’s a very tough question to answer. I have heard many lecturers getting this question in different events. No one wants to make a very tight prediction. But what we can say is, as Sergio mentioned earlier, that it’s not important when it comes. What is important, there’s a very steep learning curve for companies to adopt quantum computing and therefore you cannot wait until it’s ready in order to start understanding how it will impact your industry.

So it’s not important when, but it’s important that we believe it will happen. So we need to be ready by then. Whenever quantum advantage comes, we need to be ready to understand how to integrate it into our products, how to put these new solvers into production, and what benefit they can bring versus our current state-of-the-art classical techniques. We need to understand how to map our problems to something that can be an input for a quantum algorithm. These are the steps that we need, that it’s now the moment to work on, and then whenever quantum advantage is a reality, we will be the first there to make use of it, and we will not lose the competitive advantage.

Yuval: Let’s assume that I’m the CIO of a mid-sized regional bank, maybe Midwest serving a few states in the Midwest. How should I go about it? What would you advise, Sergio, that I do as the path to understanding quantum and getting ready to put it into use?

Sergio: I am so happy that you made me this question. So, a mid-sized bank is very likely, very likely won’t have a huge amount of resources to do the research by themselves, but maybe they have a few data scientists on payroll that have been helping with some specific machine learning cases and whatnot. Now these very companies would be most likely leveraging generative AI today to use AI almost if it was a low-code or no-code replacement of traditional machine learning approaches. In that same way, how can they leverage quantum whenever it happens, but in a way that they can get a competitive advantage by not being left behind?

That is precisely the solution that we want to put in the market, in the sense that some other people are going to do the research and build intellectual property and patents and all that nice commercialization that we are living today, but then it’s going to be a matter of licensing the algorithms themselves. How can we make an algorithm-as-a-service solution that gives a company like DCIO the benefits of doing the same thing? What are they doing? Optimization, regulatory capital requirements, and so on.

So what we’re working on and we’re going to be putting in the market later in this year, this is a spoiler alert, is tailored precisely towards these types of companies, where you play with algorithms as a service, with some of the use cases that we are self-developed or that we pattern with other companies in the quantum industry that we will be able to comment on later in the year. So this company can say, “Hey, these are my assets,” or, “These are the loans I’m giving away. These are my credit card transactions, met in any use case like the ones that Carmen mentioned before.” And then with a simple API call, I can get the best-of-breed solution that exists, whether is classical, quantum-inspired, which of course is fashionable lately, pure quantum with certain types of hardware, with digital analog.

We don’t know what’s going to work in the future, right? And many, many companies are pushing forward, which I think is fantastic and helps us. But that should not be the problem of this tier two bank. Should use whatever is the best solution out there.

Yuval: Carmen, when you were describing use cases earlier, you mentioned broad categories, machine learning and optimization, and maybe a couple of others. But could you be a little bit more specific? So how can economists start using quantum today? Would it be around machine learning? Would it be around optimization? If you had to pick just one application that economists could use today, what would that be?

Carmen: Well, I think this is also linked to what Sergio just explained. In the end, an economist, is a user that should stay at a high level when using a certain kind of technology, whether it’s GPU computation, whether it’s AI, whether it’s supercomputing or quantum. So I think that quantum should be abstracted from them. And this is why companies like us are working on the middle ground between the expertise in the domain that is going to just consume whatever output they need for solving a problem or for doing some research or for trying to predict what’s going to happen in the future. And we need to give them the most accurate and the best solution that we can at the most optimal time, when time is important.

So for an economist, I would say that they should try to understand where their current solutions are struggling today. And whether, for example, they experience that there’s some calculation that they would need to get daily or that they will need to do even more than once per day, but right now there’s no provider that enables them to do such a computation so often. There they would have had identified a bottleneck that potentially has some advantage from being executed in quantum. So I think that for them, what we need is to keep staying at a high level but tell us what their business needs are. What computations do they need to do more accurately? What computations do they need to do faster or more frequently? And it will be on us to work on providing the technology that can get them there.

Yuval: As we get close to the end of our conversation today, I want to go back to the study that you mentioned. You mentioned that you did the study and you mentioned that because you’re not a hardware or software provider in quantum, then maybe it’s not as biased as some of our companies who are just trying to generate more business for them. And I understand that it showed that companies are not entirely prepared for quantum or maybe to put it mildly, not entirely prepared, but what other highlights can you share from the survey?

Sergio: Sure, absolutely. And I would like to say that the report is available to download on our quantum computing website. So please feel free to go there and get it because it has a lot of really interesting insights. We tried to do both a qualitative and quantitative analysis and ask directly, specifically a lot of tier two financial institutions, but also tier one on those who were aware of quantum. We filtered specifically people who had a certain knowledge of the field. So we could have a sample that where people would say, “Yes, I would like to do this. Or not, I would not like to do this because X, Y, and Z.”

And the main message is, as you were saying, underinvestment. The most common case was that 87% of the respondents did not have any budget at all for quantum In some cases, it was one person who worked on everything, all the innovation cases at the same time, but had no budget to hire consultants, contractors, or companies, or even quantum hardware time or anything like that. So, it was more the biggest case as someone monitoring the technology to be able to make the case of investing in that when the time was right. And from all of those, 73% of the people couldn’t really define any commercial advantage or what could be those benefits.

Now, I think this is where we have the opportunity, all of us working in the quantum industry. We believe that there will be commercial advantage, otherwise, we would not be doing what we’re done, but there’s still a lot of communication work without the hype that can land this industry. Let me reinforce the without-hype concept.

One of the things that we’ve seen as well is people who went all the way around, who started playing with quantum, got fooled by the quantum hype, said that the results that could get today gave them maybe a piece of research, maybe some IP that arguably would have close to zero value in the future. So they don’t want to keep investing relevant amounts of money every year. So they decided to stop the investment or reduce it extensively until they could see some other value coming from other places.

And the other thing that I want to mention is some of the challenges that the respondents brought to the table. Of course, the applicability is one of the challenges. How to train people, how to get the skills from people. It’s not just about hiring quantum engineers. Now we are in the age of increasing the team size and increasing the scope. So how do you train your marketing people to put it there, your salespeople to sell those solutions, your data scientists to get more people into the fold, but also how do you integrate it? And this is one of the points that I want to try to land here. It’s not just about putting together a Jupyter notebook that you execute, works and it gets rotten in three months’ time. You have to start thinking about your pipelines, how it’s going to be embedded into your product, how it will grow together with your product.

Yuval: I want to end with a hypothetical question, and perhaps first Carmen and then Sergio. If you could have dinner with one of the quantum greats dead or alive, who would that person

Carmen: Well, in my case, I would love to have dinner with Feynman because I’ve actually read the book that is, I don’t know the title in English because I read it in Spanish, but it’s something like “Are you joking, Mr. Feynman?” And he seemed like a really interesting person besides being one of the parents of Quantum. So yeah, that would be my answer. Both because I think at the personal level he would be very interesting and to ask him how he envisioned the field could evolve.

Sergio: Well, maybe you can invite me to that and then I can still pick my choice. I don’t know if the last time I was here on this podcast, you asked me the same question. I think you did, but I don’t remember what I answered. Incidentally last week I watched Oppenheimer. I’m not sure if that changed my opinion a little bit, but I want to tell a little story about what happened to me a couple of weeks ago. I went to our New York office and now we do hotelling. So basically you book your desk or your room for today or for the week. You don’t know or no one should know where someone is going to be. I go relatively often to our New York office. And when I got to my desk early in the morning, I see that on the table, there were three books, three of the classics on quantum mechanics right there. And I said that we are a company full of quants and many of them are physicists, so it is natural. But the fact that we had those books in there was in deep ease. And one of those was one of the seminal books the “The React on Quantum Mechanics.” So I just took it and I felt something when I was reading this one. So now my answer would be having dinner with Dirac.

Yuval: Very good. So Carmen, Sergio, thank you so much for joining me today.

Sergio: Always a pleasure.

Carmen: Thank you very much, Yuval