Next in Tech | Ep. 264: Quantum CEO Series: Amir Naveh

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry Research at S&P Global. And today, we have the next segment of our CEO series on quantum computing, and we're going to dig into quantum software development stack. And with me to discuss it is Amir Naveh, Co-Founder of Classic, the developer of a quantum software development platform. Amir, welcome to the podcast. 
 

Amir Naveh: Thanks, Eric. It's great to be here. 
 

Eric Hanselman: It's great to have you here. We've been talking about the architectures of different types of quantum computers, the approaches from the quantum computing side. But I wanted to get into the software development end of quantum computing. This is -- I guess, we're now a few weeks away from NVIDIA's developer of the GTC Conference. Jensen Wang was there talking about the importance of the software development stack in AI. We've spent a lot of time building up quantum infrastructure, and yet we're only now in the relatively early days of really looking at what software development environments around quantum computing really require and the kinds of tooling we need to put in place to really make quantum computing more available to a broader set of users. Can you give me a little background in terms of what you see as what this progression is and I guess, where we are today in terms of the capabilities we've got? 
 

Amir Naveh: So very high level, quantum computers are using a whole fundamentally different logic. So I would argue building the software stack for quantum computers, it's even more complicated and even bigger task than building software platforms or tools or capabilities for AI. It's kind of, in a sense, we're creating what we've done for classical computing over 80 years, right? So we need to build operating systems. We need to build compilers. We need to build the whole levels that bring you from the ability to describe what you want all the way down to running on a quantum computer, which is a massive, massive task. So I mean, right now, at Classic, we're 150 people. We've been running for 6 years, and we're touching really the tiniest tip of the iceberg here of what needs to be done. So I think the one main thing I'm thinking when we're saying software stack for quantum, it's like this is a huge, massive challenge, and we're just getting started. 
 

Eric Hanselman: Well, I was noting that it's now -- it's actually 9 years ago this month that IBM released its first really organized software tooling, the Qiskit software package. If we think about how this has evolved, you were touching the fact that the basic process is something that we've done in classical computing. As the joke goes, it's exactly the same, only completely different, right? This is -- we do have to get to a point at which we can start to build abstractions for development. So you're not having to code directly to the actual quantum computer. As we've been talking in previous episodes, the architectures that we're really looking at can be very different in terms of not only the problems that they're trying to solve, but the architectures that they're actually bringing to bear to be able to attack some of these problems. Being able to have a development environment that can translate between these 2 is not exactly a simple process. How do you approach that? And what are the ways in which you can start to get to a flexible enough environment -- so -- because the idea is you want to be able to retarget software that you build today will be able to go target other architectures with other ways of actually solving problems in an environment that's still developing very rapidly. 
 

Amir Naveh: Yes, that's a really good question. I remember the kiss kit launch back in 2017, I remember like wiring those like a single and 2 qubit gates on like 3 qubits and seeing, okay, I got a bill state, I got a GHC state. That's -- I mean, it's really exciting. I think it was a huge milestone, and we've got so much since. To your question regarding the different architectures, I think it's an interesting time today because really, there are 3 or 4 leading hardware modalities that no one really knows who will be the winner or if there will be one winner or different applications will require different kinds of hardware, superconducting photonics, ion traps, neutral atoms all have like massive advantages and disadvantages. -- the happy part here is that Sure's algorithm was developed in 1995 before any of these modalities really got up to speed. And you can design these algorithms, and you can create working, at least in theory, working software on all of these hardware and it would be fundamentally the same because they all employ the same logic. So all of them work within Hatamarchy. All of them have 2 qubit gates, which up to some small rotations are all equivalent. So yes, the last mile of translation will always be very hardware dependent. And also, I mean, sometimes there's orders of magnitude difference in the run time, for example. But there are many concepts that are similar, right? So if I have -- let's take So's algorithm again, if I want to design Sure's algorithm in a way that actually runs, there is a lot of heavy lifting just to make it run into the logical level. And then you have to also make it run into the actual hardware. This is where we work very closely with the hardware providers to make sure it goes -- it's optimal all the way down. 
 

Eric Hanselman: And that's really one of these things when you were talking about the long history of building operating systems, building compilers. These are things that we've had to have translation layers between the abstraction at which the broader concepts get developed, things like high-level algorithmic design and the way in which we actually translate that into the environment in which the code is actually going to run. It seems though that when we start looking at some of the specifics around this, you were talking about some of the rotational capabilities. We're now -- I guess the thing I'm heading towards, is this something where these are relatively small differences between systems that optimizations can take advantage of? Because if you think today, we wind up in environments where we've got different processors that have got different sets of capabilities. Some have got a floating point, large floating-point capabilities. Others, you have to do the floating-point computations in software. It does make a significant difference in terms of how that actually runs. But we've gotten pretty good at translating between the various kinds of capabilities simply because the architectural components have been fairly standardized. With quantum computers, though, it seems like between some architectures, we're dealing with differences that are significantly greater in terms of the capabilities. Or are they close enough? I'm curious about what the extent of that translation really has to look at given the systems, the variety of systems we have today. 
 

Amir Naveh: Right. I think in essence, you can translate in practice, it's much harder than translating between one chip architecture to another, right? It's not like, okay, do you have ARM or do you have, I don't know, something from Apple or something from Intel and then you just -- it's more fundamental than that. So if it ends up that the like most relevant processors 10 years from now would be photonic chips and also superconducting, for instance. So it's a massive translation between the 2. I think there is kind of a trade-off. You can wait and see which hardware advances and then start building the software, but it's the hardware will advance hand-to-hand with the software. And it's -- I mean, it's not -- probably not a good idea to wait until the hardware is ready. So yes, it's not -- it's not going to be an easy translation between all these modalities. You do need to have the optimization that is very aware of which technology is going to be used. What we do for IRAs, for example, and what we do for superconducting or for neutral atoms, it's quite different in several levels. 
 

Eric Hanselman: Well, it seems like something where if we think about the focus that many enterprises are putting on being able to actually wade into quantum capabilities today to start taking those first steps. We see a number of different paths in terms of those that are trying to tackle some of these problems. We've got optimizations that are taking place today in terms of where we're actually able to apply the quantum computing functionality that exists. I'm a chemist by schooling. I'm fascinated by a lot of the chemical reaction dynamics that are being built today. Those are things that can take place reasonably simple terms on a lot of the architectures that exist in computing today. It seems like maybe a lot of what enterprises really need to be thinking about is how they scope the kinds of applications they're bringing to the quantum question today. I'm curious about the kinds of journeys that you see some of your customers going through as they try to sort out the practical and some of that building experience and capability in terms of their functionality today. 
 

Amir Naveh: Right. Yes. So we started Classic in 2020. And back then, there were maybe a few dozen enterprises that were really looking at algorithms and applications. And even those teams were typically very small. It would be like maybe a handful of engineers that got some like part-time work exploring some algorithm or another. In the past few years, now really any large enterprise name you can think of probably has some quantum computing effort. It's in the thousands at least. We're working with many dozens of enterprise customers, and it's just the volume of the size of the teams and the number of relevant enterprises that are trying to understand how this influences them is really rising dramatically. I think #1 interest for everyone is to understand how quantum is going to impact their business. Before building any concrete algorithm or testing it or whatever, just understanding is this going to be the next AI for us. It's going to be bigger. What can we do with it? What should we be afraid of? What's competition doing? And this is really, in some terms, it's a must because the hardware timelines are getting closer and it's become -- you really do need to understand how this is going to influence your business. Then once you get this -- you're over this exploration phase, typically, what people are building is, one, the teams and the know-how and the independence. Two is the algorithms and the IP. So a lot of these teams are starting to build all kinds of use cases and applications, which they can either patent or keep as secrets and prepare themselves for when these computers are coming. That's the typical use cases we see. 
 

Eric Hanselman: That makes sense, especially now that there's much readier access to quantum capabilities. We've now got a number of different platforms on which you can consume quantum functionality today. These are areas where I know when we're talking to our clients, there's certainly a lot of interest. The question for most of them is how do they actually get started? And the practical end of this is they've got development teams that understand classical computing paradigms and getting knowledgeable about that next stage of real understanding how you begin to explore, yes, identifying use cases is critical. The next stage is really getting our teams to understand how you actually start to leverage quantum capabilities. How do you see some of those on-ramps? And what are the early-stage tools that you see your customers starting to use as they start to define some of these paths to quantum usability? 
 

Amir Naveh: Right. So I would claim not to be to centered on classic. But I would claim that's our main value proposition today for enterprises at least that are trying to enter this field. So this is a lot of what we do today is take teams that sometimes are built from AI engineers or classical algorithm experts or domain knowledge experts, and we teach them what they need to know in order to build the same things in quantum. So basically, I mean, it may sound a bit -- yes, it maybe sound a bit simplistic, but it's not that hard. For many of these teams, it's not that hard to -- so for example, you mentioned optimization problems or quantum chemistry problems. The problems are really complex. But if you know what is the molecule or trying to find the ground state of or if you have a clear model of what is the optimization problem, mapping it into quantum algorithms and seeing more or less what they can or cannot do, generally estimating the resources, seeing what algorithms come into play. This is something that a handful of smart engineers can be able to do within a few months. Platforms like Classics really help. And yes, it's not -- you don't have to have a PhD in physics in order to understand how these problems map into quantum algorithms. You need to be genius to invent a new algorithm, but to map it into the existing algorithms, that's something that a typical smart AI engineer can do. 
 

Eric Hanselman: It seems like there is a broad enough set of knowledge today in specific verticals that are starting to apply these capabilities that you've got quantum algorithms that are now relatively well proved out, and you've got applications and use cases that there's already been that groundwork done in order to actually start to lay some of these out. And I know that from a -- if you look at the early days of Qiskit, as you're pointing out, what you were able to actually do granted there was the promise that, oh, it's a Python library. All you have to do is write a little Python code and hey, you're on your way. Of course, you fundamentally have to understand what that translation is between a Python expression and the actual quantum logic you're trying to implement. So being able to get towards an environment that's got a little more abstraction, hopefully gets people a little further along. But you raised what I think is one of the points that we spend a lot of time talking with our clients about, which is when you look at and take your particular projection about where we get to some level of higher-end quantum, with my security hat on, we keep thinking about Q days, quantum capabilities to use So's algorithm to start potentially solving a lot of the hard encryption problems that are out there. We keep looking at 2030. So if we start thinking about where organizations ought to be, for most new capability and technology initiatives, 3, 4 years' worth of timeline to be able to master that is relatively short. So this is something where -- and again, we keep saying organizations need to get moving on this. We've gotten to a point at which their practical applications today -- now that next stage is really actually starting to get to that broader enterprise use case. There are a lot of capabilities out there that quantum can tackle and organization should be moving forward with today. 
 

Amir Naveh: Right. I agree 100%. I think 3, 4 years, that's the relevant timelines, and it's really -- I mean, someone who hasn't started today, that's a really short time to get started. So just building properly the team, understanding how it fits into what is happening today and then finding, I mean, not all use cases are the same. If something is using Grover's algorithm could take 20 years until a relevant quantum computer would be relevant. Maybe never. I don't know, Grover needs massive quantum computers to ever be useful. But then if you're looking at quantum chemistry problems or quantum or simulation problems, or even in some optimization problems, then the timeline is even shorter. And then part of the work you need to do is understand which of the applications are relevant in which timelines. And yes, that's -- be 100%, I think now is the time if you didn't start yet, now is the time that you really need to understand how this is going to impact. 
 

Eric Hanselman: So what are the sets of resources that you're typically pointing your customers to in terms of getting them started for those that are looking at relatively early-stage work, the kinds of things we've talked about, previous episodes, we're talking about a number of optimization problems, even things that seem to be relatively simple, but that from a classical computing perspective can be really hard to solve. A lot of the scheduling, the traveling salesperson kinds of things. But even things like production line optimization, one of the things we dug into in a bit -- those are areas where a quantum approach can give significantly faster results in places where those optimizations can actually have real dollar contributions in applications that exist today. What are the ways in which are you typically working with partners like global systems integrators that have specialization? Are there early-stage engagement paths? I'm curious about what those pathways are that you see your customers taking to get themselves up into greater capabilities. 
 

Amir Naveh: Right. First, I would really encourage all the listeners to open the classic library, just type in classic library in Google, and you'll find hundreds of different use cases and applications that we've built together with users or with customers. This is the largest kind of use case example application set available. And you'll find all of the optimization problems you mentioned, many of the quantum chemistry problems, many of the simulation problems, computational fluid dynamic problems, obviously, cybersecurity and cryptography. So really, I mean, all of the -- I'd say this is the set of applications that people are exploring today. Typically, in a call like when we do introduction calls with customers, in 1 hour, I would be able to listen to what they say are the main pain points and probably map 5 or 6 different problems that quantum computers are able to tackle just from hearing what they're saying are the massive computational challenges today. because, I mean, there are quite a lot of frameworks. And then once you know the algorithms and you've seen a bit about what can and cannot be done. So it's quite easy to map the optimization problems or the fluid dynamic problems or whatever partial differential equation problems or quantum chemistry problems or I mean -- so we were -- or machine learning problems. Yes. So that's typically how we work. We do work also with the system integrators, Deloitte, PwC, Infosys, Accenture, NTT in Japan. And they all have very large teams that are approaching enterprise users directly. So we're doing both. We're doing the direct approach, talking with the customers, feeding them what we've already discovered with many similar teams and then also with the service providers trying to do it maybe at a more horizontal level, teach them and have them work with as many as possible. 
 

Eric Hanselman: A lot of pathways to be able to achieve some of these benefits. But as we've been saying, I guess, the most important part of this is to actually get moving. So well, this has been great, Amir. I appreciate all the perspectives. Thank you so much for being on the podcast. 
 

Amir Naveh: Thanks so much for having me. And if anyone specifically is interested, how do we get into Quantum, then feel free to reach out to me or anyone from our team, and we'd be happy to take you on this journey. 
 

Eric Hanselman: We can point them in your direction. Well, so many different things we could be talking about, so many other avenues, but we are at time for this episode. That's it for this episode of Next in Tech. Thanks to our audience for staying with us, and thanks to our production team, including Sophie Carr, Feranmi Adeoshun, and Kyra Smith on the marketing and events team. If you like this episode, please subscribe or like us. Please keep in mind that statements made by S&P Global represent their own views and are not necessarily the views of S&P Global. I hope you'll join us for our next episode because there is always something, Next in Tech!