Podcast with Steve Flinter, Vice President, AI & ML, Mastercard

My guest today is Steve Flinter, Vice President, Artificial Intelligence & Machine Learning, Mastercard Labs. Steve and I talk about specific quantum applications that Mastercard is exploring, and how they are different than the typical financial services firm, what’s holding them back from moving to production, and much more.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval: Hello, Steve. Thanks for joining me today.

Steve: It's my pleasure. Thanks for having me.

Yuval: Who are you and what do you do?

Steve: So my name is Steve Flinter. I work for Mastercard in our R&D division and I, among other things, lead up our research areas around emerging technologies. And that includes obviously things like quantum, but also areas that we focus on like 5G, looking at new forms of payments. So anything that we think is going to be relevant to Mastercard or our customers now or in the foreseeable future.

Yuval: Excellent. How did the quantum computing group start at Mastercard? Sometimes we see it starting from the top. Sometimes we see it starting almost like a skunkworks project that someone does in his spare time. What was the case in Mastercard?

Steve: I think in our case it was probably a little bit more bottoms up. It was started, I guess, based out of ideas that our executive VP has around technology. And he and I discussed where this was going, was it at the right time for Mastercard to get involved. And I guess our first formal foray into this space was a research project being organized by IBM locally, IBM Research here in Ireland, as well as some universities and startups. And that was co-funded by the Irish government. And that gave us the impetus to get involved. So we started with this multi-part collaboration with other leading players and we've built and grown the team from that.

Yuval: And how large is the team? And if I may ask, what's the composition? Is it mainly physicists? Is it mainly finance experts? Is it all of the above? How is it built?

Steve: We look at people with a data science and mathematics type background, less so with the physics side. I work with other colleagues in other parts of the organization who are looking at other applications of quantum technology, so quantum networking and so on, and they bring specific expertise around cryptography and networking type technology. But of the areas that I'm particularly interested in, it's mostly on the applications of quantum computing to solve business problems.

Yuval: We sometimes see that the quantum computing is relatively small because the computers are relatively small these days, but companies are sometimes working to build broader support, prepare for the future, educate other colleagues around quantum, maybe go around the organization and look for use cases where that could be applicable in the future. Is that something that you're doing as well? What effort are you engaged in broadening the knowledge and imparting it to others?

Steve: I think all of those things that you mentioned are definitely very relevant to what I do and what we in R&D are trying to do. You can think of it in a few ways. One is we're trying to do that awareness raising at an executive level, at a senior level, trying to help them to understand what are the technology trends at companies like yourselves and IBM and D-Wave and the hardware vendors, what technologies are coming out of the tech space that we think is going to be impactful for us and for our customers and how should we think about that and invest in that as an organization. We're also focused on getting into the weeds of what are specific use cases that we think are going to be relevant, how do we start to bring some of those to life, how do we help our product managers and product owners think about what are the relevant applications for quantum.

Because, as you well know, and as your listeners will know, you need to think about this in quite a different way. And the classical way of thinking about approaching problems from a software or engineering or computer science approach may not always be the right ones. And so there's definitely a degree of education and awareness we need to raise there. And then probably I'd say the other angle that we're looking at is about helping to educate our broad group of technologists and engineers right across the Mastercard organization, not just in R&D. And so we work closely with our learning and development colleagues to think about what internal training and education resources can we put in place, how do we help interest the developers in empowering themselves and go on their own learning journey, even if it's not directly connected with a product development effort or an R&D effort, but to build their own skills and awareness, particularly for those who have an appetite and an interest in this area. So there's an element of all of those things.

Yuval: Sometimes when people talk about quantum computing, they talk about hype. And I think certain level of hype is good because it builds excitement and funding and helps recruit people, but when you go around the organization, do you typically run into, oh, this is 20 years away? Or do you see more of a thirst to learn and understand how it could be applied to a business problem?

Steve: It is a challenge and it is definitely something to navigate. There are those out there who would hold an opinion that it is 10 to 20 years out and it's still largely in the realms of science fiction. There's others that want it tomorrow, and when can we have it? So there definitely is a balancing act between those two and trying to set realistic expectations around where the technology is, the pace at which it's developing and when we think it's realistic to expect quantum computers to start to deliver business value. And so I think for Mastercard and for some of our customers, that two to three to four year horizon is where we try and place the work that we're doing, what is happening within that time horizon that can potentially deliver value.

We can certainly take a perspective on what's further out and see how that is going to roll into longer term roadmaps, but we tend to be focused on that short to medium term time horizon and thinking about that in terms of what is our response to that area, what types of applications do we think are going to be practical or realistic in that timeframe. But I think coming back to your question, overall I think there is definitely an appetite to learn more. There's an appetite for people to inform themselves as to what's going on in the space, and partly it's my job and the job of our team to try and ground some of those expectations, make them realistic, get the excitement there, try and get a sense of urgency around it, but not oversell either.

Yuval: Listening to your answers sounds like you're very focused on practical applications on relatively near term. So let's talk a little bit about applications. I think I read on the Mastercard website that you're doing or trying to apply quantum to customer rewards, that sounds a little bit like quantum machine learning, and then on routing of contactless transactions, which might be an optimization problem. Could you shed a little bit more light or give some more detail on these applications or other things that you're willing to talk about?

Steve: First, if anyone Googles or looks up applications in financial services, very often you'll find things like derivatives pricing and credit risk scoring and these kinds of applications. And they certainly fit very well in an investment banking or a securities market view of the world, but probably less relevant for Mastercard and for our customers. And so part of what we've been doing over the last couple of years is to look at what applications are out there, how do we start to characterize them, and some of the ones you've pointed to are certainly in the areas that we see opportunity. Mastercard, for example, has a very large loyalty and rewards business. It may not be known to all of your listeners, but it's certainly a key part of our business and we're one of the biggest players globally in that space.

And there's a number of hard and interesting problems in that area around how do you find the most appropriate reward or offer or loyalty event to give to a given end user or end customer. There's lots of different optimization problems within that, subject to different constraints. So that's definitely one area we're interested in and exploring. Mastercard, at its heart, is a network business. We move information from merchants, from retailers, through an acquiring system to our issuers, the banks who actually issue your payment card, and we also work and are developing many other payment networks. And so there's large routing problems that exist within all of that to move all of this information around.

So, again, we think that there are definitely opportunities in there to apply quantum to solving some of those problems. So there a couple of great examples. You mentioned quantum machine learning, that's definitely an interesting area, and Mastercard is using machine learning in lots and lots of different avenues, not least which would be fraud detection, fraud reduction, so there are applications for where there can be specific quantum approaches to some of those areas. I would say it's an evolving landscape. We're continuing to develop and try and seek out some of those use cases, and particularly then, back to my earlier point, try and align them with where the technology is, where it's going to be in the next three, four years and ensuring that we're not attacking problems that actually are going to need devices that might be five to 10 years out. So just trying to find the right problems for the technology that's available now or will be in the next couple of generations.

Yuval: Just out of curiosity on the routing problem of transactions. Because when I think of transaction routing, it sounds, oh, we're just moving a couple or many bits around. Is this more of an arbitrage thing that you have to convert from one currency to another? Is that about transfer fees? How much does it cost to move this money around? What is the cost function or the cost drivers of optimal routing?

Steve: Cost may be one thing, but another longer term strategy of Mastercard that we've been building out over the last number of years is what we call multi-rail. And this is basically the concept of having different forms of payments. The consumer to merchant payment is the one that everybody would recognize for Mastercard, but we have evolved into things like account to account payments in certain markets, into business to business payments in other markets. As you grow these different forms of payments that you can support, the routing options grow exponentially or combinatorically. And so there could be different ways of managing the payments flowing through the network or collections of networks that we run and manage. So these problems grow larger and larger as we try and serve different parts of the payment ecosystem.

Yuval: If you had to guess, how soon before one of these applications is in production or maybe one already is in production?

Steve: Not in production at the moment, we are definitely working towards being in a position to bring quantum applications into production in the next number of years. It's hard to put an exact timeframe on it and obviously sooner is better than later, but I think we're definitely motivated to find those relatively near term, near being two, three years, near term use cases that can deliver actual business benefit. We're not interested in putting quantum into production just for its own sake, just to show that we can do it. It's much more around can we demonstrate that we can solve a problem through a quantum driven process that is meaningfully and notably better than the alternative that we could do through traditional CPU, GPU type computing. And better can mean different things.

It can be just a better result, a better optimization, but it equally could be we can get the same result in a shorter amount of time, or we can do it at a lower cost in terms of energy or compute power, whatever. So better can mean different things, but really it's that idea that can we deliver an application or set of applications to the organization that can, as I said, deliver that commercial benefit. And that's where we're focused. And to add maybe one other point to that, part of what that then drives into is also thinking through, well, what does it take to actually put something into production? And for us that's not just being able to run something by hand through a Jupyter Notebook, a Python interface. It's how do we plug it into a pipeline? How do we run it as part of a 24/7 operation? How does it get scheduled relative to other things? So there's that operational side of putting it into production, as well as the act solving the underlying problem piece of it.

Yuval: As far as moving into production, what are the key things that might help you move faster? Do you need better computers? Do you need better software development platforms? Do you need just more people? What is it that if you had your three wishes, what would you need to move faster into production?

Steve: I think one of the big challenges that we have as we approach the whole quantum computing space is a lot of the problems and a lot of the systems that Mastercard have are big data by nature. We deal with vast amounts of transactions and the data that those as transactions generate. And as you know and as your listeners will know, quantum computers struggle with that. They're not big data processing machines. So a lot of what we do is trying to figure out, well, how do we take a problem that might have originated in a big data environment and turn it into a compute problem that is practical for a quantum approach.

And that can be looking at areas like compression or clustering or other approach is data density reduction or data size reduction. So there's some of the things that we're trying to think through, how do we solve those problems and being able to solve problems that have that large data characteristic stick to them I think is one of the central parts of what we're trying to tackle.

Yuval: As we get close to the end of our conversation today, I wanted to ask you about two unrelated topics. The first one is global geopolitics, if I may. Mastercard is a global company and there seems like at any given time something is going on around the world. And especially these days, there's a lot of focus on somethings called a quantum's arms race. Who's going to have the bigger computer and the faster, and who's going to be able to crack who's encryption and so on and so on. Does that worry you? Does that impact you? Or do you leave that just to the governments to take care of?

Steve: There is certainly a big aspect of that that is at a government level. And for Mastercard, we are present in almost every country in the world and have relations with those countries. And so we, I guess, try and steer clear of some of those issues. I'm probably not the best person to comment on that at a corporate level, but I think the key things that, as I said, we're interested in is very much focused on working with the best technology and that's coming out of different regions, and then ensuring that we can use that technology to deliver benefit to our customers. Who, again, are regionally distributed.

And what we may well find is that each of those regions may have their own policies, their own governance around how they want quantum computing to be used in their areas. And I know for example Europe is very keen on ensuring that it has its own quantum industry and can compete competitively with the US and China. And we may end up having to allow for some of those developments in our own strategy. But we do approach things as a global company and trying to solve things for a global customer base.

Yuval: And I'm guessing you also are monitoring the impact that quantum computers might have on cybersecurity and in cryptography.

Steve: Yeah, absolutely. As, again, anyone who reads about this area in the press, it's often one of the first things that they'll find is a doom and gloom scenario that quantum computers are going to break all cryptography. And, again, back to our earlier conversation around the hype around all of this, that's definitely something that we need to manage and, I think, level set around how the potential that is there, but maybe 10 to 15 years out. So getting the right level of interest and action and reaction to the potential that's out there. But for sure I'm certainly tracking how people are thinking about it, how they're responding to it, bodies like NIST, for example, who are looking at cryptographic standards and how they need to evolve.

And that's definitely a big part of, and will continue to be a part of our strategy going forward to ensure that the cryptography that we're using now and into the future will be quantum resistant or quantum proof. Actually we published a standard last year, and I think you alluded to it earlier, around contactless payments, and that standard took a deliberate set of decisions around the cryptographic schemes that it was using so that it would be quantum resistant by the time that standard was fully implemented and rolled out. So I think we'll definitely see more of that where system architects and security architects are taking into account the potential future power and capabilities of quantum devices in the standards that they write now, because these standards take time to write. They take potentially a long time to roll out and so we're not just solving for issues that are here this year or next year, but potentially 10, 15, 20 years out.

Yuval: Steve, how can people get in touch with you to learn more about your work or to see if you've got openings on your team and so on?

Steve: LinkedIn is probably the best place to get me. Steve Flinter. There's not too many of us out there, so it should be easy enough to find in LinkedIn. Happy to engage with people who are interested in this area. Definitely, we're hoping to grow the team and continue to add expertise as we build up those use cases and as we get a close and closer to building applications. So for sure I think it's a very ripe area and one in which I think we will be growing our investment over time.

Yuval: Excellent. Thank you so much for sharing your insights with me today.

Steve: It's been a pleasure. Thanks for having me on, Yuval.