Podcast with Marcin Detyniecki and George Woodman of AXA

My guests today are Marcin Detyniecki, head of research and development and chief data scientist at AXA as well as George Woodman, quantum computing lead at AXA. Marcin, George and I discuss what enterprises are looking for when they get into quantum computing, how Marcin and George built support within AXA and much more.

Listen to additional podcasts here

THE FULL TRANSCRIPT IS BELOW

Yuval Boger (Classiq): Hello Marcin. Hello George. How are you doing today?

Marcin Detyniecki (Axa): Oh, thank you very much for having us.

George Woodman (Axa): Yes. Thank you. Thank you very much. Feeling great.

Yuval: So, who are you and what do you do?

Marcin: So George, maybe you start?

George: Yes. Okay. So, I am the Quantum Computing Lead at AXA. AXA is, if you don't know, is one of the leading insurers in France and the world. And I run their quantum computing project to try and see where we can use quantum computing within the insurance business and how it's going to impact us. Trying against the timelines and make people aware within the business that this technology could be on its way.

Marcin: Yeah. And I'm Marcin Detyniecki, and I'm the Chief Data Scientist of the group. And, among my jobs I'm responsible for the R&D part. And this is how I'm involved in quantum.

Yuval: Excellent. And I saw a presentation, that the two of you gave in 2019. So I suspect you started quantum before that. When did you start? When did access start with quantum? And, What got you interested in the first place?

Marcin: Yeah, I think we started a little bit earlier, I would say maybe a one year earlier than that. And why did we start? I think, if you think about quantum computing, we know it's kind of mathematically proven that this is going to disrupt our world, our classical world. If you think, let's say, something like Shor’s algorithm and prime factorization, right. And this is what we're able to do something we were not able to do before the prime factorization and this is disruption, right? Now we are AXA. We are a great, very large insurance company and our work is to think about the future and prepare for that future. So, in particular, if there is some sort of disruption, we need to kind of get prepared. So quite naturally we decided to prepare for it.

What I think makes our preparation unique, is that we invested time and effort to really understand the technology. It was not just about kind of trying to figure out, this is nice use case, and so, but really going deep into understanding. How does it work? How the mechanisms, how can we can program the machines and so on, so this is a particular routine. And, if you think about it, it's very fundamental because quantum computing for somebody who is far away, may think that this is just the next personal computer, the next generation, but it's not, right? It's not, because … it's nature, which is associated, it's very strongly associated, for example, with the way you program it. If you design circuits, it's very different of, object programming languages. So really we invest a lot of time into trying to understand the fundamentals. And, also because, I think, that this technology is very good for some very specific problems and use cases and not for others. I think, this is what it makes our project a little bit different.

Yuval: Now, quantum computing in the Financial Services Industry has a lot of different use cases, right? From risk analysis to portfolio optimization and credit risk and what have you. Is there a particular use case that, or use cases that you found are interesting to AXA?

George: I think, I'll take this one. So, I think, a majority of our work, over the last two or three years now has been to find this use case, because, finance of course has lots of interesting use cases, as you can see from all of the publications that are coming, out of all the big investment banks, but there's been less activity from the insurance side. And, what we are trying to find is, going through the different types of technology. So that's including, annealing and universal gate-based quantum computing and trying to figure out what are the small parts, because it's not going to solve the entire problem. It's fine trying to find the small parts of a big problem that we believe quantum can be solving in the next seven, 10, 15 years. Part of that, is of course, risk analysis and the Monte Carlo Speedups that are hopeful potential, but then you get more optimization based problems that could have a shorter timeline, based on how the technology grows and which technology we are going to use.

And then, we have to look into how the improvement of the technology affects just our customers. So, we don't do pharmaceuticals, but our customers do, and that could affect them. And we need to keep them informed about that, because that's the technology that's probably going to progress the most, the fastest, and they need to be aware of that. And there are certain other areas like that, that we think that quantum could really be used for.

But we are, I think, we're trying to keep it as, trying to be realistic with the technology, because I think there's a lot of over hype from maybe the technology providers. And then, I think I, well, we try to give them a good baseline, not to oversell the technology, but so they, so that our business lines understand when it's going to be available. And that, this process is going to take a long time. The use cases that we're going to build, are going to be very small at the start and they only, they of grow at a slow rate until you reach that quantum advantage stage in, however long that's going to be. And then hopefully by then, you can really take off and prove, how good quantum can be, but it's going to be a long way.

Yuval: One of the challenges that I think organizations have when new technology comes in is how to sell it to the business units, to get organizational buy-in. Otherwise it just two guys in a room doing something and no one will ever know. What can you share from your experience at AXA that might be relevant to other people who are going into quantum and saying “how do we get organizational buy-in? What should we do? What should we not do to get this going in our companies?”

Marcin: Yeah. Maybe George you can comment a little bit or what we're doing with our internal communities.

George: Yeah, sure. So, I think, it's really good for, I know quantum computing is quite confusing, it's got a steep learning curve at the start. But, I think, if you get over that curve with maybe a few technical people that do have access to all of these use cases and teach them what the potential of quantum can do, and also keep them aware of, how the technology is growing and be realistic in expectations, and I think, that would mean that the future for that use case can be very profitable.

But, it's creating this quantum computing community where people, lots of different people are interacting with each other with different quantum levels of quantum computing knowledge, so that, they can work together to prove these use cases. Because we found the hardest thing to do, is to find these specific use cases that quantum computing can be used for, because there's all of these algorithms that are coming out, all the technology advancements, but aligning them all together and finding the technology that's not only going to be useful in like 10 years. But also going to have a big impact and is scalable from now, is something that we've battled with a lot. And, I think, having an internal community of a hundred stakeholders with various backgrounds in the business can really help, grow that and help...

Yuval: How close are you to something in production? Where are you on the quantum journey from understanding the technology to building a couple of simple circuits to prototyping, to actually something that would be deployed in production?

George: So I think, the deployment in production is a hard thing to gauge because, you don't want to produce something in production if it's not better than its current classical counterpart. We have done a few say proof concepts of certain areas, of where we think a quantum computer could be useful, but they've always been quite small toy models. And then we're waiting for the technology and the hardware to improve, so that we can build that into a production environment that we can prove is faster than what we're currently doing. But a lot of the areas that we are looking at, the calculations that we need to be doing are just so massive. The technology that we are acquiring is in the... I don't like to gauge quantum size by Qubits because, there's so much more to assess a quantum computer than just by the number of Qubits, but it's the easiest thing to gauge.

And, the Qubit numbers just aren't there at the moment for the size of the problems that we want to be working out. We do have proof of concepts and ideas for and, add the algorithms change as well. We keep up with those and hopefully they can, bring forward that timeline and that point that we can actually start deploying them and not just using them as a research project. But I still think it's a number of years away until production based environments are going to be affected by quantum. And, when that does happen, there'll be lots of other. Because at the moment you're just working on the cloud on a pay as you go basis. But, at that point it might need a bit more of a serious conversation with how it works.

Yuval: I have a hypothesis on what companies want. And that hypothesis comes from both speaking to a lot of companies as part of my job at Classiq, and because I interview a lot of fascinating people on the podcast. So, I want to ask you a few bullet points, short answers, if you think it makes sense. I mean, maybe I've got it completely wrong. And so this is my opportunity to learn from the experts. So I think that when companies get into quantum the first thing or not the first thing, but an important thing for them is to build internal competency. So they don't want to say that they say quantum is just too important to outsource everything to a consulting group. Do you think that's a fair statement?

George: Yes. I think, not to jump in too much, but I think that, because it's such a complicated technology, you need to internalize some of that knowledge because all these consultants or technology companies might be coming with all of these hopes that technology, is going to break the world in three years and you need to put the ground to down and be no, this is actually what's going to happen for us and not just leave that up to an external provider.

Yuval: Perfect. And so, the other thing, I think people want is short prototyping cycle. So you've identified a number of use cases, maybe you want to do proof of concepts. It's probably better than a proof of concept takes two or three months and not two or three years. You think that's a fair statement?

George: Yeah. Everybody wants to have short. Everybody wants to have short...

Marcin: Short cycles. I don't know if it's always the feasible, but yes, that's what you want to do. You want to have, go through things and, try to get that done. I think the most important part of your hypothesis, is to get it done. It's not that much the shortness, it's... Of course, you try to make it short, but, it's more to make it, without, even releasing some difficulties. So for example, making the toy, instead of trying to wait for three years to make it the full size. So, yes, it's a cycle oriented, but, so you see the spirit. It's a little bit different just to think, I want, yes, of course, I want short cycles, which produce money at the end of the day. Right?

Yuval: Okay. And to George's point earlier, how important is it when you do the proof of concepts to choose algorithms that scale? So you can say, okay, today I have 20 or 30 Qubits, I can do this and that, but this algorithm, once computers are with 2000 Qubits, I can use something very similar on a bigger computer. How important is to choose something that scales in your opinion?

George: I think, It's two sides there, because obviously you want something that scales, but also a scalable algorithm might not have the same impact that a non-scalable algorithm that has a larger impact at a lower level. We are still a business and to sell the technology to the stakeholders, you need to show proof that it's working. So, there's a lot of balancing between proving something that could potentially work now, to something that could potentially work in 10 years, but won't work now. It's so, I think, a good analysis of that. Would be between, annealing and gate based. So, annealing is more of a realized technology, the actual hope as if it would be. Have quantum advantages still up for debate, isn't it. But that is something that you could potentially show now, to a director and be like, look at this amazing thing that you can do.

Whereas if you say, try and do a Monte Carlo simulation with an algorithm and try and show that to the directors that where you're only running it on three or four assets, because so much of the other is put into error correction, or just holding the qubits then, it does need a balance between the two, obviously. And also, because the research is being, it's perpetually, these algorithms are getting improved upon, by all of the software and the algorithm companies and the universities, and the technology companies. The algorithms of today are definitely not be the algorithms of seven years from now. Maybe Shor or there'll be the Grover and the Shor which are like the foundations, but the actual algorithms that we're using in the moment will be potentially different because of how the technology goes.

And also, I feel how you get superconducting IonQ trapped ion. So, it can spend all the different times of technology, which one of them is going to come out on top or are they all just going to branch off into their own. Because they all have pros and cons and which one's going to improve, and some do better with some algorithms on other, because of, their coherence time or like the fabrication, all the ability to scale them up. I think, there's so many different factors into choosing which algorithm is best for a certain problem. And then putting that to say 10 years in the future, is difficult.

Yuval: But, and that leads me perfectly to the next and next to last point in the hypothesis, which is hardware portability. So let's put aside annealing, and as you mentioned, there are many gate-based approaches and organizations that I speak with, are saying, "well, we're not sure if I am going to win or Honeywell or IBM or PsiQuantum or whoever”. I like everyone, but, how important is it for an organization, like AXA, to say, we don't want to write the code in such a way that it only runs on one gate-based machine?

Marcin: I think, this goes a little bit, it's very much in a relationship with their previous question, right? Because it's, your focusing very much on one dimension of the problem, which is the scalability. I think more important than scalability, and it was mentioned by George, it's I think the fact that this matches a real business problem. Okay. That's really a thing. Okay. Because, in some sense, quantum computing, it's extremely promising, but it's not delivering, let's say the full expected value yet, right. And, we're early stage, I understand, but you need to find this use case. And, I think the community in general is very much focused on the technology. Not enough, I think, looking at the business problem, the business problem is very, very tricky. And this will answer your question, right?

Because it's, it's very tricky because you need to find the exact place of your problem, which in which you can make it happen. And, since it's not full-fledged, let's say in terms of size or memory or whatever, they need to find a problem, which is kind of the right size, the right problem, and fits your process and has a big business problem. So if you are very close to that, okay, so let's say you do trading, okay, and you're very close to that, and you have an option called kind of problem, and you're really close. Then, you can be very specific to one technology, right?

Because when you're very close, then you want to optimize this very specific architecture, and then you can, you get this one-year advantage, let's say, okay, or, and... If your problem is a little bit more far away in time, in that case, what you want to do, is to be generic. So I would say us, we are closer to the more generic kind of, approach, because we have not identified use case where we are really kind of a route just about to put in production where we need to kind of squeeze the think and exploit the specificities of it technology. So, I think that we're underestimating, or let's say the community in general is underestimating the difficulty to nail the business problem.

Yuval: And in that sense, and that really brings me to the last part of my hypothesis is that it's really important to be able to integrate experts that are not quantum experts. So experts in trading or experts in life insurance or experts in chemistry, depending on the business problem. I suspect you agree with that?

Marcin: Yeah, absolutely. George can tell you, this is really, we spend a lot of time. It's the difficulty, is to kind of convince the people that you need to invest now for something that it's not proven completely yet, that this is out there. So people, all these people, all these expert or business experts, they have short term kind of, missions and results and objectives. And, the whole challenge is to bring them and tell them, what is the potential of this technology so that yes, we, they work together and it's really, as I said, it's not just one dimension, the technology, the scalability, the architecture. It's really everything to make it happen.

Yuval: I was really happy that you agreed to come on the podcast because I want to bring interesting content to my listeners. But, I'm curious, what did you, or what do you want to gain from being here? What would you like to gain from this podcast?

George: Well, of course, we are very honored to be here. I think from me, I think it's really important that, because the technology is such an early stage that there's not too much a competitive advantage for an external ecosystem. So, say other insurance providers, financial providers, or general companies interested in quantum to form a kind of ecosystem of, what, this is what we are doing, what are you doing. Pros and cons. Because there are not that many people in the quantum space at the moment and it's a space that's only going to grow.

So I think bringing knowledge together is only a good thing. And we've realized internally just how much interest there is in quantum because it's quite a fancy buzzword. People hear it, they get quite excited. So I think, maybe producing, I think getting the word out there, that there is an insurer interested in quantum computing and, if there are others, we can discuss what we're working on, and potentially work together in that aspect.

Yuval: Excellent. Now, you know what you're doing internally as AXA, but if you could influence what other vendors are doing, whether software or hardware or other enterprise companies or consulting companies, what would you like them to focus on in the next year or two?

Marcin: Yeah. So I mentioned a little bit this right. It's, I told you I think what's really challenging. I already said, and I will say something else also after that. But, the thing is, I really think that the business, the business part of the problem is very important. If people come and tell me, yes, I can accelerate Monte Carlo sampling from N to square root of N, oh, wow, that's cool. Right? Especially if N is big, but what does it mean in practice? Okay, for my business. And this is a little bit underestimated.

The other thing is, I think we're very at early-stage technology, I think it's very promising. I'm really enthusiastic, I wasn't initially to be honest with you, it's by getting deep into this, I really think I'm very enthusiastic about it. But I think since this is early stage.

Also the second thing, I'm... It's about trust, It's about sharing and working together. The people we work with, is usually the ones who are open to share and not try to kind of, oh, we have these secrets and we have these great things, but we cannot explain anything. They're kind of transparent because as you saw through the old podcast, really we're about understanding, you need to have the understanding inside the company to make it happen. So, if these companies don't transfer in some sense the understanding, it's going to be very difficult to this, to make it happen, because a long chain after. Usually the people you are entering is us, but afterward there's a long chain after us, still. So, yeah. So it's really about that.

Yuval: Excellent. How can people get in touch with you to learn more about the work that AXA and you are doing?

George: Contact me. I'm happy to talk to anyone about Quantum Computing. You can put my email in the bio or contact George Woodman on LinkedIn, and I'd be happy to have a chat about, anyone that's interested in quantum potentially collaborating.

Yuval: Perfect. Thank you so much for joining me today.

Marcin: Thank you to you and thank you for people listening.

George: Yes. Thank you very much.