Bill Wisotsky, Principal Quantum Systems Architect at SAS, joins Yuval to discuss SAS’s vendor-agnostic, hybrid approach to quantum and analytics. Highlights include D-Wave warm-starts that let SAS prove optimal kidney-exchange solutions in seconds, as well as QML pilots for fraud and bankruptcy modeling and disaster response. He shares a pragmatic definition of “quantum advantage”, treating QPUs as just another PU alongside CPUs/GPUs, and why rapid progress toward error correction—and ideas like Penrose’s—keep him learning.
Transcript
Yuval: Hello Bill, and thank you for joining me today.
Bill: Hey, Yuval, thanks for having me. This is a great experience.
Yuval: Well, it’s a great experience for me as well. So who are you and what do you do?
Bill: So my name is Bill Wisotsky. I work for SAS. I’ve been there for about 24 years. Most of that time has been in professional services. I started quantum there about four years ago, and it’s progressed to where it is today, where we actually have a dedicated team and I’ve been moved over to R&D as a principal quantum systems architect.
Yuval: And for those that don’t know, SAS, that’s not the airline, right? That’s not the British Special Forces. Tell me just a little bit about SAS.
Bill: So SAS has been around since the early 1970s. We’re the leader in the, anything analytical, right? So machine learning, AI, we do business intelligence, data management. Our solutions are used by virtually almost every company that is doing some sort of analytics. And you missed the shoe company. We’re not the shoe either.
Yuval: And you’re not software as a service either, right? So it’s not S-A-A-S and you’re not S-A-S-S, right? So it’s really just S-A-S.
Bill: S-A-S. Either all-capital or all lowercase.
Yuval: Very good. What was the insight that led SAS to go and start exploring quantum, I think you said, four years ago?
Bill: Well, I mean, from my perspective, I’ve always been interested in quantum. When I was in graduate school for behavioral neuroscience, my research was in biophysics doing EEG recordings, specifically something called visually evoked potential on the human cortex. And I had to learn about the way photons were being absorbed by the retina. So I had to learn a little bit about quantum physics, and I was introduced to the double-slit experiment. And it turned into almost like a rabbit hole where I was so interested in what I was learning about quantum physics and I was spending more time doing that. Fast forward, around COVID time, obviously everything shut down. My kid’s sports were all canceled. I had a lot of unaccountable time. And I said, “Let me explore this concept of quantum computing.” And I realized it was more advanced than it was years ago when I first researched it. So I started playing around with it. I started teaching myself some of the software development kits and figuring out how I could integrate it with SAS. And, um, I saw that, you know, some of the real promising areas of quantum computing around areas such as optimization and quantum machine learning, you know, that’s in our wheelhouse, that’s our bread and butter. So it was a natural sort of marriage between the two.
Yuval: So in your case, in your company’s case, it was sort of bottom up, right? You knew about the potential or was it some directive from the CEO, “Hey, we should really look at this quantum thing.”
Bill: Yeah. I kind of went off on my own when I started doing this and I kind of navigated through SAS to try to figure out who to talk to, to sell this idea to. And it was presentation after presentation and proof of concept after proof of concept. I wound up in a department called the Solutions Factory that bought into this. And they were the executive sponsors, and we did a whole bunch of more organized proof of concepts, I should say. And eventually, the chief technology officer saw it, and he came on board, and here we are.
Yuval: So I very much want to hear about how it’s going, but let’s actually talk a little bit more about how it started. So you said you studied biophysics, but I don’t think you have a PhD in physics, right? And so you didn’t have a PhD in physics and you had to convince the company that this is worth something, this is worth looking at. What lessons have you learned or what advice can you give to a friend who is in a similar situation, sort of similar educationally and similar to the company, how to get going and how to get started in a business?
Bill: So you’re right about my PhD. My graduate program was in behavioral neuroscience. And I originally went into that to do animal behavior and evolutionary biology. And after working in an animal lab, I decided that’s not what I wanted to do. So I switched over to humans, and biophysics was the lab that interested me. I actually never completed the PhD, because we got involved and we started a biotech company that ultimately kind of went south. And then I wound up working for SAS doing predictive analytics. I always was interested in quantum ever since my dissertation. I mean, I was all the way up to the point of just defending it for the most part. I just had to write it, do some edits. And it was a hobby of mine for years. I kept on reading about it. I would read textbooks. I’d read papers, all from, you know, I’d say early 2000s all the way up until, you know, when I started this. So, if I was to give advice, I would say to anybody, be very persistent, learn the topics, speak from some level of education. You don’t have to be a physicist to understand quantum computing, and keep pushing forward. It’s sometimes a longer road than others.
Yuval: You’ve been doing this for four years. And if I remember correctly, we met after I heard you speak at a D-Wave conference about a case study. Um, tell me about some of the exciting projects that you did in these four years, some of the successful, or maybe the not so successful ones.
Bill: Yeah. So some of them I could talk about in detail, others I can’t. At D-Wave, we did this sort of project called the Kidney Exchange Project, where we were looking at donor and recipient pairs and how one donor that wants to donate a kidney to a recipient might not be compatible. And it creates an — so the donor one could, you know, actually donate to somebody else that they are compatible with, and it’s called a swap in a way. And that network could be very, very large, and it has a lot of constraints, and it has some objectives. So it turns into a very large combinatorial optimization problem. And we quickly realized that when we ran this through quantum, we got a distribution of results. We got good results, but there was no, you know, guarantee that the result we were getting was the absolute best. We can only figure that out from a classical perspective. So we would take the results from the quantum solver, in this case it was D-Wave, and we would use them as warm starts into our classical solver. And then we were able to solve the problem in a fraction amount of time and prove optimality. From there, we went on to do the same technique on some other proof of concepts around optimization in consumer packaged goods. We did, we’re working on some POCs in the quantum machine learning arena around fraud detection and around bankruptcy modeling. And we’re also doing disaster response with you. We’re working on some projects along with yourself.
Yuval: Some of these POCs, I’m guessing, are non-internal POCs, but you work with customers, clients, Do you tell them this is quantum? Do they care or do they get excited or turned off when you say, “Hey, I’m actually going to use this quantum computer thing?”
Bill: Yeah. So that’s a good question. The only non-customer POC was the kidney exchange problem. And that’s because it’s a well-known problem. Every other one was some sort of customer-driven project. Now, when we tell customers, we tell them that it’s a quantum project. That’s how we wound up doing the POC, and they’re very interested, and they’re very excited. We go into this more in a sort of collaborative way where we don’t quite know what’s going to come out of the project. They don’t quite know what’s going to come out of the project. And we work with them from both the quantum perspective and the classical perspective to see what might come out of it. And it’s a learning experience for both sides. But the amount of understanding and appreciation from customers. They’re very excited about quantum and their expectation level is very, I wouldn’t say low, but it’s very negotiable, right? They’re not expecting to solve every sort of complex problem. They know that quantum is in the beginning stages and could only be used for very specific things. So in a way they’re fun.
Yuval: Have you seen quantum advantage? I mean, either a problem that you couldn’t solve classically or was too long to solve classically or an area where something like this warm start that you were describing actually delivered tangible benefits?
Bill: Yeah. So, you know, that kidney exchange problem, for example, to solve that to optimality — it took hours to solve. In the hybrid approach, we solved it in 30 seconds. So that is one area if you want to define quantum advantage of speed. The funny thing about working with customers at SAS is we get a little bit more freedom in how we define quantum advantage, right? So you hear these articles about some quantum company solving problems in two hours that would take 45 million years or whatever the number is. But it’s not only about speed in our definition. In our definition, quantum advantage is something that the customer will benefit from. It could be speed. It could be the expressivity of a quantum machine learning model by being able to project classical data into higher dimensional representations and then get a more predictive model, even though that model takes longer to run. That might be a level of quantum advantage. So with that being said, in some of our data sets that we’ve used around quantum reservoir computing and quantum support vector machines, we’ve seen some levels of quantum advantage, not necessarily about speed, but when you’re measuring things like area under the curve, we’ve seen some benefit. We’ve also seen benefit in being able to train these models with less data. We start off with small data sets, and then we move, you know, we get larger and, you know, quantum seems to stay very similar and classical varies with the size of the data sets.
Yuval: One of the things that struck me when we were working together, I mean, SAS was doing a project with QuEra, my company, is that you were asking about, or your team was thinking about optimality. Sort of quantum gives a solution, but it may not be the optimal solution. Um, do you find that customers are looking necessarily for the optimal solution or they just want something good enough within a reasonable amount of time?
Bill: It depends on the customer and the field that they’re in. In some fields, they need a 100% optimal solution and they don’t care how long it takes to get it. Um, usually there’s more at stake if the solution is not optimal. Um, in other cases, it’s about speed. They just want a really good solution, but they need it very fast.
Yuval: But with quantum, I don’t think you’re able to prove optimality.
Bill: No, and that’s where that hybrid process comes in, right? So when we take those solutions, so what we could do is if when we get our sort of distribution of solutions, and we have no idea how close to optimality they are, we could feed them into our Viya software in parallel and have Viya solve them, basically all at the same time with that warm start. So each one of those Viya processes will have a different warm start from each one of those solutions that we’re solving in quantum. And then we’re able to run that to optimality and prove optimality at that point.
Yuval: When you solve a problem using classical means, I’m sure by now you’ve developed, you and as far as the company, sort of a workflow. Okay, here’s how we capture the problem. Here’s how we think about expressing it. Is that any different in quantum? How does that work for, look for you when you try to solve a problem using quantum computer?
Bill: Yeah, you know, it’s interesting you say that because it’s a much harder thing to do that in quantum than in classical, right? In classical, it’s tried and true. We have all these different, you know, algorithms. We have X number of years of experience. We have all these specialists in the field. In quantum, it’s almost like sort of the event horizon of a black hole. You kind of not know what’s on the other side. And in some cases, we found that quantum works really well on certain problems, and then we take similar problems and quantum doesn’t work as well on those similar problems. What we’re finding is problems that have like a high combinatorial relationships amongst the variables. When that solution space is very large to search, we’re able to get a solution relatively quickly. And that’s from an optimization perspective. In a machine learning perspective, it’s a little bit more of an unknown and we have to kind of test different things. But it’s, and that’s why I said before, I’m gonna work on customer POCs. It’s very much a collaborative project where we have to understand the problem classically first. We need to understand how this works classically, where the areas in that classical process are very computationally expensive, take a long period of time, what the problem formulations look like, and then we have to kind of pull that back and say, could quantum help in these specific areas? And that’s how we kind of, you know, are looking at these things.
Yuval: Would it be correct to say that today quantum is just another arrow in your quiver, just another tool in the tool set, so it gives you another option on how to tackle a particular problem?
Bill: Yeah, I mean, we see it as another, you know, what we call a PU, right? We have CPUs, we have GPUs, and now we have QPUs, right? I mean, we don’t see it as really any different, and we want to take away some of that complexity from customers, right? A customer today, if they’re running a deep learning model in SAS, they’re not looking at the deep learning model and necessarily caring if it’s running on GPUs or CPUs. They just want the model to finish and as fast as it could with as much predictability that it can. We see quantum fitting into that paradigm as another sort of way of processing a problem.
Yuval: Do you have a favorite quantum computer? And the Lego quantum computer that you have is not a legitimate answer here. Um, and you don’t have to say QuEra, of course, a favorite. And why is that computer your favorite? What makes it, uh, sort of your first choice? If, if there is one.
Bill: I don’t, I don’t have a favorite. I’ll be honest. And I find that there are strengths and weaknesses across the board. What we’re trying to do is we’re trying to look at different modalities of quantum computing, right? We’re looking at neutral atoms. We’re looking at trapped ions. We’re looking at photonics and superconducting because they all have their sort of strengths or weaknesses. They all fit into the quantum puzzle in different ways. And then we look at vendors within these different modalities and then we choose which ones we want to partner with. Yeah. And, and yeah, there’s no favorite really.
Yuval: But what makes a vendor a good one for you? Is it the team that you work with? Is that the quality of the documentation, the availability of the machine? I mean, how can a vendor get on your good side?
Bill: It’s a little bit of everything, right? It’s a little bit of, um, you know, we take a look at the quantum computer, the hardware itself, and we see which one is, you know, the most reliable or most advanced. We take a look at the software development kit that we have to integrate with and we have to use and the documentation. But really what’s very important to us is the team that we wind up working with from the quantum manufacturer. If the team is kind of not available or standoffish in some respects, it’s harder for us to do our development and work with them. If the partner is very engaging and easy to work with, that’s a big plus. It’s very important.
Yuval: You’ve been doing this for four years. What have you learned about quantum in the past year that you didn’t know before? What’s new in your mind?
Bill: I’m learning things every day. I’m learning new things about neutral atoms, things that I didn’t know before. I’m learning things about trapped ions that I didn’t know before. But the biggest thing that I’m learning is that you can’t stop learning. When I first started with this, they were talking about quantum computing being 20 or 30 years away, but that sort of horizon keeps on getting closer and closer and closer. What I’ve learned in the last year is that every quantum manufacturer is pushing that horizon of sort of error-corrected qubits and large-scale quantum computers ever closer to the present. So, you know, I think that is one of the biggest take-homes. Another thing that I’m learning is that there are ways of achieving some level of quantum advantage without having these large quantum computers with tremendous error correction, that there are ways of getting various levels of what we, I guess, would call quantum advantage. Just have to be a little imaginative, I guess.
Yuval: When you think of SAS competitors, do you also see them using quantum or do you find that SAS is ahead in that respect?
Bill: I mean, I don’t know if they’re using quantum or not. I mean, there are some vendors that are more niche vendors than us that don’t encompass such a large breadth of solutions. And they’re involved in quantum a little bit, but I don’t know if I would call them competitors per se.
Yuval: I think you knew this question was coming, but I do have to ask you a hypothetical as we get towards the end of our conversation today. So if you could have dinner with one of the quantum greats, dead or alive, who would that person be?
Bill: I think Roger Penrose. I’ve listened to him talk and I think that some of his ideas are pretty intriguing. One of his ideas, this Orchestrated Objective Reduction theory, it hits home to me because it’s all about consciousness and microtubules and neurons and quantum superpositions and entanglements that happen on the nervous system. And because my background was in behavioral neuroscience, I remember learning about action potentials and graded potentials and how On/Off receptors and parvo or magnocellular sort of arrays in your brain process vision. And it always got to a point where there is a hard cutoff between sensation and perception. And there was a lot of theories about how that kind of, you know, you could build up all these on-off receptors to make lines and shapes, but that sort of switch over to perception was always sort of theoretical. Well, it brings, you know, it connects various higher order cortical areas of the brain. And I never liked that. And then when I read about this Orchestrated Objective Reduction theory, it was really the first one that I ever heard about that talked about a quantum theory in neurons that was responsible for consciousness to some extent. So I thought that was very, very interesting. So I would love to be able to pick his brain about some of this stuff.
Yuval: You know, but as I was listening to you, I think there’s a question I forgot to ask. You mentioned that you moved from services to sort of an R&D position related to quantum. What’s the difference in, what’s the difference as it relates to SAS and what are you going to be able to do in R&D that you weren’t doing in services?
Bill: Yeah. So services, um, I wasn’t doing quantum in services. I was doing quantum on a part-time basis. I was doing it at nighttime after work or before work or at free time, but I had, I had customer engagements that I was working with that had to do with predictive modeling or data access or business intelligence or just performance issues that might arise when doing things on certain types of servers configured in certain ways. So this is allowing me to do quantum full-time. And my position now is it’s kind of varied in a way. I work with the developers on these quantum algorithms and what they should do and how they could get incorporated into SAS. I also work with product management as to what the… when we come up with our first stage of beta products, what they should look like. I work with partners like yourself. I work with academic institutions on joint research projects, on presentations, on guest lectures. It’s a fun role in that sense. I get to really get my hands really deep in the algorithm development, as well as, you know, high level stuff.
Yuval: That’s wonderful. Bill, thank you so much for joining me today.
Bill: Yuval, it’s a pleasure and this is a great experience. Thank you very much.