Quantum computing has applications across communications, network building, financial modeling, cybersecurity, AI, and so much more. But getting involved with these projects has traditionally taken years of education at the most advanced academic levels. According to Helmut Katzgraber, Global Practice Lead at Amazon Quantum Solutions Lab, the scientific community could be doing a better job inviting more people to participate in discovering breakthroughs. In today’s podcast episode, Ben, Matt, and Ryan reflect on Katzgraber’s experience in the field, along with the steps he is taking to broaden access to education on quantum computing.
Katzgraber reflects on his time as a university professor up until 2020 and why he switched to working at Amazon.
He walks us through a quantum computing challenge that he hosted with BMW, through his role at Amazon (and what real world applications he sees emerging from these types of collaboration experiments).
We discuss what inspires him to stay curious — raising the bar for scientific research, crowdsourcing breakthroughs, and opening up the playing field for more people to jump in.
Follow Ben, Ryan, Matt, and Helmut.
‘Til next time, all.
[intro music plays]
Ben Popper Hello, everybody. Welcome back to the Stack Overflow Podcast, a place to talk all things software, technology, and extremely tiny particles. I am Ben Popper. I am the Director of Content here at Stack Overflow, joined as I often am by my colleagues, Matt and Ryan. How's it going, y'all?
Ryan Donovan Pretty good. How’re you doing?
BP All right. So my knowledge of quantum computing doesn't extend much beyond Antman parts one and two, but luckily we have an expert on today to discuss some of this stuff with us– both what it is, how it's being practiced now, and maybe some of the future potential applications. So yeah, it should be a really exciting episode. I want to welcome Helmut Katzgraber, who is the Senior Practice Manager at Amazon in the Quantum Solutions Lab. Helmut, welcome to the show.
Helmut Katzgraber Thank you. Pleasure to be here.
BP So I guess first, for folks who don't know, tell us a little bit about yourself. How did you get into this field?
HK That's a long story actually. Until December 2020, I was a university professor with a focus on computational physics and computational optimization for understanding physics materials in quantum computing. And I kept reading about quantum speedup claims and that tickled my interest so I decided to see how far I can push the bar on classical hardware to see what quantum computers can really do, and this is how I basically shifted my focus towards quantum applications. I then consulted for a couple of companies. And overall, I've spent now about five years trying to find valuable quantum computing applications in industry.
BP And I guess for listeners at home who are not familiar, we have had a few episodes with different folks on the program to discuss this, but explain it to me like I'm five. What is quantum computing? How is it different from classical computing or what people may be familiar with in their lives as users of computers or software developers? And what's special about that? Why are companies spending so much money and time and research pursuing this idea?
HK Well as a matter of fact, I'd like to just mention briefly that today's physics Nobel Prize was awarded actually for a quantum entanglement experiment, and that pretty much signifies how important quantum computing is these days. A quantum computer is a very different type of, let me call it a special purpose processor, so we're not going to have a quantum phone or a laptop anytime soon. But it's really a device that is intended to aid for very difficult computations that currently are very hard to solve on classical computers. The fundamental building blocks are not classical bits that can take the value of zero or one, but they are so-called qubits, or quantum bits, that can be in all kinds of superpositions of states. In other words, in layman's terms, it's somewhere between zero and one, and that allows of course for a much richer set of gate operations, which means that we can do some computations more effectively.
Matt Kiernander I'm really pulling back to my university days here and some of the operating papers that I did. The way that I understand it is very much down on a deep level, you have the zero and the one, and then when you combine those you can have things like an AND gate, an OR gate, or you can do very simple things like add and subtract and it basically kind of builds from those base on and off states into more and more complex and wonderful things. So how does that translate from just the zero and one and what possibility do these basically four bits, as I understand it, open up in terms of possibilities?
HK So in quantum computing we have similar operators like you would have on classical bits, like ANDs, ORs, et cetera, et cetera. But because we have these more rich computational variables, we have certain operators that don't exist classically. And from a physics point of view, that means that we can exploit two important effects. One of them is superposition, that a system can be in a combination of states, and there is this beautiful thought experiment from Erwin Schrödinger known as Schrödinger’s Cat. I can get into the details of it, it's a little creepy, but it's kind of cool. And the other one is entanglement, which means that a quantum particle over here can basically be in the same state as another quantum system at a distance. And because of that, you can do these very interesting and weird operations that frankly are very hard to understand. I have to admit that when I was an undergrad, yeah, I learned it, I passed the test, but it took years to really internalize what these things mean. And so because of these additional effects that we have that we don't have on classical bits, we can do some more interesting operations that allow us to create algorithms that potentially are exponentially faster than classical algorithms for similar workloads.
BP Yeah. I have to raise my hand and say that I was one of the people who took quantum mechanics as a freshman in college just so I could get the hard sciences credit. And the enjoyable part of it is that you're reading more about the theory and you're playing around with ideas like Schrödinger’s cat and you're not actually doing any math which I'm not capable of. So I remember really enjoying that course. I don't think I got a terrible grade even, and I managed to get one of my science credits finished. So Helmut, one of the things that really interested me was talking about this idea that there's a lot of interesting research here, there's a lot of potential. I know companies have talked about quantum supremacy, meaning they've unlocked some big breakthroughs recently in what they can do maybe in the lab and in terms of speed. But one of the things that interested me especially about having you on was talking about what it might look like in practice. So I know there was a competition where companies could sort of submit ideas and the winner for one of them was BMW, and they had a couple of cool ideas about material construction for cars, or how they might place sensors, or how they might design things to be crash resistant. So in what ways does quantum computing enable those kinds of use cases?
HK So I'd like to go back to one of the things that you said, and that is demonstrating some sort of quantum advantage. And as you duly said, this has been demonstrated in the lab for I would say a very contrived problem that has exactly zero value to society today. And so if you ask me, chasing after these milestones is a bit of a flawed race, because the question everybody should be asking is, what are the problems that we have today that we cannot solve and can quantum computing tackle those? And not just say, “Well mine is faster, and that one's faster,” et cetera, et cetera. And so where I'm going with this is that it's important to look at real world problems. And this is, by the way, the mission that my team has at AWS– to work with customers and to actually try to find applications of value. And if not, to raise the bar using classical technologies. Now coming back to the BMW challenge, correct, Amazon and the BMW Group organized a challenge together actually, where we selected four use cases out of BMW's collection of many, many problems that they face during production. For example, sensor placement for self-driving cars, or material deformation with machine learning, et cetera, et cetera. And the beauty of this was that it was open, meaning that we used a crowd science approach. Everybody could submit a potential solution, and it was fascinating to see not only the breadth of solutions, but the creativity that was used. And so for us evaluating the submissions, it was a treasure trove of ideas and a lot of aha moments of techniques, methodologies, and applications of algorithms that we would never have thought of before. And this is exactly what is needed to basically advance quantum computing, to have more people, more minds, and even non-experts that come up with a crazy idea to pitch in, because this is really how we're going to unlock the next step in innovation, in my opinion.
RD We had a blog post about quantum computing a while back, and it explained it using this concept of drunken walks where it's randomly traversing a graph. And my impression was that because of the superposition, it could sort of traverse all nodes simultaneously. Is that accurate or is there a better way to describe that?
HK Oh, it's going to be really hard to describe to be very honest with you. And it's a picture that people often like to use to kind of describe in a simple way what is actually happening. In reality of course it is a little bit more complex. Just to give an idea, if I have a maze and I fill it with gas, just whatever gas you want, methane, and I light it on fire, you will see that of course everything will flash. But if you take a very quick photograph you will see that nature immediately finds the shortest path. It's a really cool experiment to do. And if you ask a human, we'll be like, “Uhh, wrong way. Uhh, wrong way.” And a quantum computer in some sense is something similar but of course at a much more complex level that it can more effectively explore that space. Instead of just trial and error, it much, much faster gets to the solution of the problem.
MK You've talked about getting more and more people involved with quantum computing and one of the things that I was having a look through this week was quantum algorithms and how they're different from classical computing where you're dealing with just two bits basically, whereas in quantum computing you're dealing with four. From an article that I was reading, they were saying that only around 20 to 50 people worldwide are working on quantum algorithms and I was curious as to why this is the case. Is there a special education that's needed to be able to even have the capability of working on such problems? What are some of the challenges that we're facing here?
HK The two versus four I think is a little bit of a misconception in that if you have two quantum bits you can basically do 2ⁿ classical data bits encoded, and if you had three it would be eight and so on. That's one of the advantages. But leaving that aside, you raise a very good point, and the point is that our educational system needs to catch up. We are very good at teaching foundational courses especially in the United States, but we're not very good at bringing people up to speed in the deep technical expertise needed, number one, to develop quantum algorithms, but number two, to also have the sense of curiosity to apply this to real world problems. And so let me give you just a simple example. You can look at many quantum algorithms that physicists have created and they're pretty straightforward, just a few lines of pseudo code. Super easy, right? Oh, look, you can even prove it scales exponentially faster in a classical method. But then when you sit down and try to implement it, you will soon realize that we don't have advanced compilers like we have today for classical computers, so it's basically by hand doing assembly language in many, many of these cases and that's a daunting task. And so for example, the Quantum Solutions Lab is currently finalizing a project with a customer where we did what is known as a resource estimation of how many qubits, how many operations will you need to actually solve a particular problem in finance using a quantum device? And so it starts with simple things. First you need to take classical data and somehow ingest it and convert it to quantum data. Then you need to take that, massage it into a particular shape so you can actually leverage a quantum algorithm, and then you need to read out the result. And when you sit down and actually do the math to all these individual steps, you will soon find out that you get such a large overhead from bringing this data into the quantum system that the speedup that you originally had goes from here to here. And so in other words, the overhead really, really messes with whatever the theory was telling you. And this is this complexity that you only gain after years and years of scientific research, people that have PhDs and have worked on this and made careers, and these are these very deep concepts that we need to be better at bringing more towards a student body. It takes time to learn this stuff, but it's not impossible to learn. And it shows also very nicely this big issue that we have that there's just not enough education in a field that is really exploding exponentially right now.
RD Yeah. You mentioned it needs different compilers. Does it also maybe need its own programming language to take advantage of the new paradigms?
HK Absolutely. And there's many proposals for types of programming languages. Different companies have created different ways and there's open source approaches. There's many different ways of doing this. But this ties straight back to something that is very dear to us at Amazon, and that is that we are in an era where not only that we don’t know which language will be the best, but we don't even know which technology we're building these machines out of. They're very, very delicate. They're very difficult to build. If I were to elbow a quantum computer that would not be good for the measurements. You laugh, but it's true. They have to be in giant isolation chambers for that reason. And so where I'm going with this is that as long as we have not figured out what technology is going to be the best one to scale and build machines with thousands of qubits, it's very hard to build the right set of compiler tools that are really honed into that particular technology. And this is why at Amazon we have a service called Amazon Braket that allows you to play with five different quantum devices, very different architectures, and a sixth one that is coming online soon, so that you can run your experiments for the problem you're interested in and then compare. Because you see, we're assuming, like in a classical computer, if you run it on an Intel versus an AMD or some other processor, you should get the same answer. In this case, we don't know. It could be that trapped ions work better than superconducting systems for this one particular problem. And so as long as we don't figure out these challenges, we still have a long way to go before we can really develop solid software platforms that will help us develop solutions at scale.
[music plays]
BP Do you want to save dozens of hours on developer interviewing and hiring? Well Turing's talent cloud combines tens of thousands of machine learning data signals across tech skills, soft skills, and more to get you the best match every time fast. Less interviewing, more building. Start your two-week risk-free trial at Turing.com.
[music plays]
BP One of the things you just mentioned which had interested me as I was reading about this was that for the BMW challenge for example, people could submit their ideas and their experiments and essentially quantum computing in the cloud was available to them. That is just such a breathtaking thing to think about because access to these machines are so limited. And I mentioned this to you before we got on the show, but my impression of it from an old New York Times story is like a giant hanging chandelier suspended in liquid nitrogen or something like that, a very delicate sci-fi looking thing. Obviously being able to access it in the cloud is great. Do you think there's also progress that we're going to see in the near future to make the machines themselves a little more durable, a little more accessible, something maybe you still shouldn't elbow, but could have in a slightly less expensive and safeguarded setting?
HK I have to tread lightly on the answer to this, because depending on who you ask you're going to get a different answer. But the fact is that quantum states are very fragile– very, very fragile to noise. And noise here is a broad term. This could be vibrations, thermal noise, any kind of external influence. Just imagine having to manipulate something extremely delicate without touching it. It sounds like the impossible thing to do, well this is what it means dealing with a quantum bit. And so, are there tabletop devices, if we forget about control electronics and classical compute to drive these? Yeah, there are. Trapped ion systems are not very large machines. They don't require special cooling either. But they have other disadvantages, like for example, very slow gate operations. So you see, each of these devices has pros and cons. There is no question that some of them are more coherent than others, in other words, they are more robust to noise. Others, like a superconducting system, use traditional type fabrication approaches, meaning that we can very easily scale up because we have 50+ years experience in that. But at the same time, they come up with other drawbacks, like connectivity. You cannot have every qubit talk to every qubit, which might be desirable for particular algorithms. Whereas in a trapped ion system, you can do whatever you want with the arrangement, but then you are confined by the size of an ion trap. And so these are all big challenges that we don't know yet and they will definitely influence what these machines in the future will look like. But very likely in my opinion, it will be superconducting systems at least for the foreseeable future, given the vast experience that we have and the big advances in both architecture but also error correction techniques that we've had recently. And by the way, this is something that we're working on at our Center for Quantum Computing in Pasadena that we embedded at CalTech, which is exactly focused on developing better materials, better fab techniques, but then also error correction schemes to be able to build scalable machines down the road.
MK I'm curious as to what personally you’re invested in changing or furthering within the realm of quantum computing over the next 5 to 10 years. What is it that really kind of interests you and gets you up in the morning around quantum computing that you are wanting to make an impact on throughout your career?
HK Well if I can candidly answer, the one thing that always motivates me is working with an amazing team. Amazon offers one of the best environments to really create and basically be always amazed by new and great ideas. But if we bring it back to the actual science, probably one of the things that motivates me most is the sheer difficult and endless search for the big value with quantum applications. It is, you could almost say disheartening in some sense, when you look at a problem, and I can give you a simple example in a second, and you're like, “Maybe quantum computers will work here.” And then you realize, “Yeah, that's not going to work.” And so there is this endless search but it's not just a lost cause, because Amazon has this tagline where we’re customer obsessed. And at first it took me some time to internalize that. And what this means is that we just don't have a hammer and try to hammer everything with the same hammer. We have the customer's problem, and we try to find the best way to solve that one problem. And so what I mean by that is, let's take the example of work that we did with the BMW Group on robot motion planning. BMW was interested in assessing if we can optimize the motion of the robots in their plant using a quantum computer. We looked at the problem and found a mathematical formulation. This mathematical formulation then needs to be converted to a mathematical formulation that a quantum device will understand. So we went from an equation like this, to an equation like this, from something that was compact to three or four lines of mathematical formulas. And then we had to embed this in a quantum device, in this case a special purpose quantum annealer– this is a quantum optimization machine. And we found out very quickly that the overhead incurred by trying to solve this problem in this mathematical representation for the quantum device is brutal. And in fact, if you have say a car that you have to seal 20 seams, what you'll find is that you will need at least, in a bare minimum, 20² to 20⁴ variables to tackle the problem with a quantum device. You have at least a quadratic to a quartic overhead. That's bad news because it's just going to blow up for you. And so of course we were like, “Okay, well this was not a good use case,” but then in parallel we took a step back and said, “Can we look at this from a different more creative way?” And we decided to use an algorithm that is used in Amazon's middle mile optimization for package deliveries, basically some vehicle routing, but then we sprinkled some physics know-how on it and used an optimization routine that is purely physics called annealing– the same type of procedure you have in a quantum annealer or a quantum optimizer. And we were able to reduce the time for these robots to move by about 10%. And if you now think about how many cars they make every day that's a huge number. And so you see on one hand, okay, this was not a good application. On the other hand, here's a better solution that came out of a completely different corner of science, where having a bunch of people with very different backgrounds and leveraging their expertise resulted in a very different solution than was previously known. And this I think is what gets me up every morning. The fact that I can innovate both classically but also with quantum devices.
MK Do you find it kind of ironic that these quantum computers can calculate some insanely complex and fundamentally mind blowing things, but on the flip side you're trying to find a problem simple enough that you can feed into a quantum computer? There's kind of this weird battle here where you're solving very complex things but you're having to distill it down into a very simple input.
HK And you see, this emphasizes the infancy of the technology. Because if we want to solve a big data problem on a classical computer, we have libraries, we have the whole nine yards of software packages that allow us to just do the ingestion of the data, manipulation, processing, and back out. In quantum we do not have that, and this is why we have to do this legwork. There are startups and companies that are building high level software layers on top of these devices, where suppose you are an optimization expert that looks at mixed integer problems. Then you just write down the equation and then in the background this gets converted to something that the quantum device can understand. But we're really at the beginning of this journey and we really have to build a lot. And I think that the most important thing right now is to really see where we can really help, and when we look at what my team and many others have seen, the first high value applications will be likely in chemical dynamics. This is important for calculations for certain medications and in material science. And now you might say, “Well, those two obscure things, why would Amazon be interested in that?” Well, it's not just that. Amazon is not interested in making materials. Amazon is interested in building a platform that allows others to come up with new materials the same way that we have EC2 and all the other tools in the cloud. And so where I'm going with this is that the reason why material science and chemistry are going to be the first big value applications is because the lift is going to be the smallest one. Quantum machines speak the language of quantum computing, and those two applications are quantum computing problems, and this is where they’ll excel.
BP That's very cool. It reminds me of another recent breakthrough that I hope will have a net benefit to humanity which was the AlphaFold program over at DeepMind which can now unlock the structure of these proteins in such a short amount of time compared to the years it used to take them. So it’s interesting to talk about computers or AI systems that can in some ways speak the language of nature as you pointed out, and therefore solve some really intractable and interesting problems. Obviously from my perspective as a consumer, I want to be able to crack all the cryptographic ciphers and hashes, take all the money I need from any of the crypto accounts out there. So security is my main concern, obviously.
HK So that's a very good point. Two facts regarding this. Number one, if you want to break RSA-2048 which is our current encryption standard, you would need, give or take, about 4,000 theoretically perfect qubits. With the currently available hardware that is still in this so-called noisy intermediate scale quantum era, you need to use a lot of these currently available physical qubits to build one of these perfect logical qubits, and the overhead is at least one in a thousand, if not more, depending on the technology. So to break RSA-2048, you would need at least 4 to 10 million qubits, if you see where I'm going with this. Right now we have a handful. The other advantage is that we already have post-quantum crypto schemes in place. NIST held a competition, the finalists have been chosen, so we have new ways of protecting the data using classical technologies. And in parallel to that –this is also something that is unique to Amazon– there's also quantum networking, where you basically make your systems immune by just simply leveraging quantum encryption. And by using this, we just created the Center for Quantum Networking, we're able to secure connections between our customers and data centers making them immune to any kind of cryptographic attacks that would otherwise exploit Shor’s algorithm to break current crypto schemes.
RD You talked about the various ways that cryptographic computers are set up and you mentioned a couple of them: the superconducting and the trapped ion. What are the other ways that supercomputers are made?
HK There's many approaches. Probably the most commonly found ones are these trapped ion systems and superconducting devices. Even within superconducting devices, there's different approaches. Some of them exploit topology, others exploit different control electronics, others exploit acoustic baths to basically suck out noise from the chips. Then you have diamond nitrogen vacancies, you have silicon qubits, so there's a plethora of things. Basically anything that is a quantum state can be used for computation. The question is, can you control it? And that's where it gets difficult.
MK I do have a question, because I try to make these conversations as accessible as possible. For any software developers out there who currently aren't 18 years old and thinking about a career in physics and quantum computing, is there anything that your average software developer can do to contribute to quantum computing? Or is that kind of locked off at the university level for them?
HK No, absolutely not. As a matter of fact, pretty much every quantum provider out there today has a ton of documentation online, like libraries, tools, educational material. There's many courses you can take for free as well. But more importantly there's platforms like Amazon Braket that allows you to experiment with different quantum devices in the cloud for free, up to a certain point might you.
MK Not ten million qubits to just run for a little wild cryptographic venture.
HK But you see, the beauty is that we have created a bunch of Jupyter Notebooks with examples that show you some of the simplest quantum algorithms and that explain to you how you can program and target these devices. And if you ask me very honestly, I think that the biggest breakthroughs will come from not quantum physicists but people that look at the problem from a very different perspective and say, “Why would you do it this way? Maybe let's try this.” I think this is where really we're going to unlock some new ideas for the next generation of quantum algorithms, and for this everybody should come and play. This is really the most important message that I'd like to give.
[music plays]
BP All right, everybody. It is that time of the show. We want to shout out a member of the community. Usually we shout out someone who's earned a badge like a lifeboat or a great question, but today I just want to shout out all the members of our Quantum Computing Stack Exchange which has been around for many years. There are questions here about how to solve a maze, how to work on the Kraus representation –don't know what that is– swap gates, padding a quantum circuit to increase the amplitude. And I will link in the show notes to the question, “How does a quantum computer do basic math at the hardware level?” This is one of those classic Stack Overflow Stack Exchange answers that's longer than most essays I know how to read. So quite interesting stuff there, so thanks to our community for contributing. And if you're interested in learning about any of this stuff and following along as Helmut said, it’s a cool community to check out and ask and answer some questions. So as always, I am Ben Popper. I am the Director of Content here at Stack Overflow. If you want to ask me anything you can find me on Twitter @BenPopper. If you want to connect with the show, questions or suggestions, email us at podcast@stackoverflow.com. And if you like what you heard, leave us a rating and a review. It really helps.
RD I'm Ryan Donovan. I edit the blog here at Stack Overflow. You can find the blog at stackoverflow.blog. If you want to find me on Twitter, I'm @RThorDonovan.
MK I'm Matt Kiernander. I'm a Developer Advocate here at Stack Overflow. You can find me online on Twitter and YouTube @MattKander.
HK My name is Helmut Katzgraber, Senior Practice Manager of the Amazon Quantum Solutions Lab. You can find me online on Twitter @Katzgraber.
BP All right, everybody. Thanks for listening and we will see you in the multiverse.
[outro music plays]