Transcript | Microsoft Azure Quantum: With Great Power Comes Great Responsibility

Microsoft’s Azure Quantum is bringing quantum computing to the masses. What does this technology giant think is in store for the industry? Are we ever going to get a topological quantum computer? Get ready for a lot of fascinating information and a relevant Spider-Man quote in this episode of The Post-Quantum World.

I’m your host, Konstantinos Karagiannis. I lead quantum computing services at Protiviti, where we’re helping companies prepare for the benefits and threats of this exploding field. I hope you’ll join each episode as we explore the technology and business impacts of this post-quantum era.

Our guest today is chief technologist for the central region for Microsoft, and I’d like to welcome Paul Edlund to the show. Thanks for coming.

Thanks, Konstantinos. Nice to talk to you in this podcast. It’ll be fun.

Yes. As everyone listening probably has heard, Microsoft is doing quite a lot in the quantum space, so I’m sure we’re going to have lots of talk about today. One of the biggest things everyone is familiar with is your quantum. So, I love this new development — the idea that everyone’s creating these cloud places where we can have classical and quantum come together. I don’t know if you have anything you wanted to talk about there.

There’s a lot of things and elements that we’re working on. I mean, we’re working on operating systems with Q#. We’re working on technology behind the scenes with qubits, where we’re actually trying to stabilise them and stop decoherence. We’re working on how to build algorithms so that they can span across classical computers and quantum computers, because I think that there’s a fallacy out there that when quantum computers ship, classical computers become irrelevant, and that’s never going to be the case. We have to educate people on what that means. There’s developments around ethics and artificial intelligence. I can’t remember if I already said that, but like I said, there are so many different angles to this conversation that I want to go into all of them, frankly.

Yes, that’s great. So, we could start with the big one: For now, when you do actually touch a real quantum computer and it’s your quantum, it’s from somebody else, right?

Yes. Absolutely. So we’re working with a combination of companies like Toshiba and IMQ and Honeywell and others, because we’re also working with them on the enabling technologies within the physical elements of quantum in the backgrounds too. Choosing multiple horses, but because they’re differentiating from themselves as well, we’re also working across them — but also in conjunction with them, I guess is the right way to say it. So, yes, there’s a lot of moving parts there, I suppose.

The future goal is to have it be that you develop, and some pieces will run in more classical hardware, some pieces will run on quantum and it could even one day be seamless to a certain level of developers?

Well, yes, and we’re creating these layers of extraction only now using something like Singular or Python, which is where most machine learning algorithms are written. You might have elements within that ML that are optimised for more of a quantum space versus a classical space. The person won’t have to really worry about that so much — they just want the answer. Not everybody is going to have access to a quantum computer, so, really, what you’re renting is almost like the equivalent of time, so it’s almost like you’re saying, “I need this for 15 seconds,” or an hour or whatever happens to be. The algorithm will just work in the background, and so almost an asynchronous process, where it’ll say, “Well, we’ll wait until the quantum computer is ready.” Then, when it’s there, it’ll throw the calculation at it, get an answer and then return back to whatever the next step is in the algorithm.

Let me start off with this: Since our project is cryptocurrency based, moving forward, a basic-level understanding of bitcoin will be helpful for your audience. The issue really comes around factorisation, so in order to describe how this works, let’s talk about a bitcoin wallet for a moment. Does that work?

Because these machines are so big right now and so complex and hard to maintain, correct me if I’m wrong, but they’re all still being housed by the companies that make them, right?

Yes, absolutely. In that world, there are so many sciences involved too. There are electrical sciences, mechanical sciences, cryogenic sciences, so it’s very much a cross-sectional learning. Everybody’s figuring this out as we go. Each of the companies is developing how they would handle those, the interconnections between those groups of people, as well as just figuring out if they can figure out ways to increase stability, for example, and make them less decoherent, if you will.

This is, of course, the first time that you’re using technology that doesn’t live inside of the Azure hardware world, right?

Not really — there are scenarios today in Azure, for example, like under certain SAP, where we have these very large instances of SAP that were managed. But those are not the norm. This will be very much like that. Eventually, those will be in our data centers – we don’t have any doubt that they’ll be in our data centers. It’s just that today, the people who are developing them, they’re inventing them, and so we’re using their inventions and then adding our technology on top. They might provide the underlying qubit, but we’re providing the operating system, or they might be providing the underlying cryogenics, but we’re providing the CMOS that sits on top, which is something that we’re doing that’s relatively new, for example.

Yes, and when companies are going to rely on these machines to do real workloads, they’re going to obviously have to be already involved in having their own VPC setup inside of Azure, right? Are they expressing any concern about this reaching out at that last moment to this unknown environment where the machines live?

I don’t know that I’ve actually uncovered that part of the conversation. I haven’t talked about those personally. I’m sure that there are concerns around security, because quantum computers are very much like Spider-Man: “With great power comes great responsibility.” I’m not personally familiar with how those things are being pulled off on a networking perspective, for example, or how they transition between the cryogenic world of a quantum computer and the classical computer that we would use to export the data and get it back into something that’s readable.

Yes. I’m waiting to hear one day that there’s going to be a quantum hacking talk at DefCon because someone found a way to go between some provider and the hardware or something. I mean, we hope not, but —

Well, yes, but we can use quantum computers to protect quantum computers in a sense too. This will be an interesting thing: We have this idea of qubits that are entangled. So, we could use the idea of entanglement to do error correcting, if you will, or to make sure that it hasn’t been tampered. It’s going to be interesting to see how this whole world plays out. We’re in the infancy stage. A lot of the work Microsoft has been doing behind the scenes in our research and development has been around things like “How do we create tamperproof encryption algorithms within quantum — using combinations of classical computers and quantum mechanics to actually create new forms of encryption?” Because if you have access to all this horsepower, the very first thing that anybody’s going to want to do is decrypt all the current RSA 2048 keys and just say, “Well what can we read?” I mean, the recipe for Coca-Cola — how long is that valid for? Decades, I would assume.

Konstantinos

Yes.

So, if somebody put it on the internet once, we might just grab a recording of that transaction and play it into this quantum computer. So we have to invent new algorithms that help us protect that stuff too.

Yes, and we’ll know you did it when there’s a Microsoft-Ola out there or something. I don’t know.

Perhaps — that’s right.

A strange side product. So, are you finding, when you’re talking to companies about getting involved in these phases, like customers considering moving some workloads over this way in the future, what kinds of concerns do they have now? Are they expecting a lot of handholding from Microsoft to get there?

We do handholding through entities within Microsoft, like consulting services or even the technology centers where I work out of, but a lot of our handholding comes from partners like Protiviti, and others who focus in on these spaces, know them more intimately and operate them for some of the larger customers that we have. We rely on a massive partner ecosystem to be the enabling arm for us in many ways. We don’t scale to try to do services for all people at all times. It’s not the way Microsoft is built. We’re very much a partner to an organisation.

Yes, because it’s such a new field to just about everyone that that’s what we’re seeing. Companies want, first, to be educated on what they can do with it, and then, it becomes a matter of drilling down and seeing what you want to do in your company. Do you want to hire people? Do you want to — what kind of experimentation?

When I first got started with computing in the ’90s, it was still relatively new, and everybody was just figuring out we’re going from NetBIOS and NetBEUI and things like that to DIP switches on servers and figuring out interrupts. Then, a few decades later, we’re in this completely new place that is completely different. But the point is that for those people who are there at the inception of real corporate networking, like myself, we had a leg up. Everything was new to everybody.

In the same vein, this is a flat playing field now for kids coming out of college, or even high school people, where people are just getting started in business to have a niche that’s not known by many other people, so this is really a great opportunity. I’m hoping that more colleges pick this up as a curriculum — this idea of quantum programming and quantum algorithms and stuff like that — because it’s a whole net new industry, essentially.

Speaking about education, do you guys have regular visibility into what kinds of percentages of activity you’re finding in Azure Quantum? Are you seeing that education is growing and becoming a major part of the use time?

No. Well, this is relatively nascent anyway. I would say when you first deploy something, there’s always this hockey stick growth, because you’re starting from zero. So, I would say that the Q# language has been in market now for a year and a half, maybe two years, and the interesting outcome of it is, we’ve actually improved some of the classical algorithms because of quantum algorithms. They made the existing ones better, and sometimes magnitudes of order better. That was an interesting outcome — I don’t think anybody foresaw that coming.

Yes. I’m always excited when I see a new environment start to stimulate, like research papers and things like that that help the entire field. I get nervous when people too quickly jump to the IP route, and they’re like, “OK, everything we do from now on, this is just for us. It’s supersensitive.” I feel that stifles growth right now in the industry.

Yes. Having an open operating system like Q# is really an important endeavor where we don’t look at those things. We’ve learned our lessons at Microsoft from being closed. It doesn’t always benefit us. As a matter of fact, in many ways, it harmed us. The Microsoft of now is very much open standards, open adoption — more open, as opposed to close-based algorithms and hoarding IP. Microsoft has a different perspective than other companies. Our perspective is that we help our customers be more productive. It’s not about us trying to hoard all the data. When you’re successful, we’re successful. I’m sure other companies think of it that way too, but the point is that our mission statement is to make every individual and every company on the planet more productive. The idea there is, in order to do that, it means that we have to play in the sandbox really nicely too.

Yes, definitely. Do you guys have any idea how it’s going in the topological computing space?

Yes. Well, we’ve been talking about it for quite a while, and we still believe that the topological qubit is the only answer that gets us to true scalability. For those not familiar, qubits operate on an electron flying around a nucleus. When you go to measure the state of that electron, you’re sharing — you have two nuclei and one electron flying between them, and you might occasionally just say, “Freeze!” and measure the weight of the two nuclei. You would expect the one with the electron bound to it to be heavier. That’s one way that we can hold two states in quantum computers at the exact same time, as opposed to a regular binary world where we can only hold one state in our mind, like a light switch.

When you have those two particles together and then another particle runs into it, and you go to look at them and measure them, and you’re like, “Wait a second — I don’t even see the two particles I was expecting,” that’s called decoherence. The topological qubit is more like if you could imagine a bunch of these particles where they’re locked arms around each other’s shoulders and they’re sitting in a tube, and if you could imagine that there were three qubits, but there were four chairs, and you would still get this idea that they’re being in an on and off state at the exact same time, but when you would, say, “Freeze!” you would look at the two ends and say which chair is open. Then, if it was open on the other side, you might call it off, but in the interim, it could have been on or off.

By doing that, by putting them in a tube, you don’t get the idea of running into each other. Decoherence goes down greatly, and what we’ve then done is put error corrections in the qubits. In the interim, what’s happening is, because there is no error correction today, people just throw more qubits at the problem. If you want a thousand stable qubits, and you have a 99% error ratio — where 99% of the time, the qubits are decohering — you need, if you want a thousand stable ones, a million qubits. We’re proposing that topological qubits will solve this problem of decoherence and therefore make them more stable.

Therefore, when you add five qubits, maybe only 5% of the time, they will fail. That’s a huge jump. We don’t need machines that are the size of football fields now. We might only need something that’s the size of a conference room, and if it’s a conference room, maybe we only need it to be the size of a refrigerator, but that’s the world that we’re heading toward.

Yes, and that’s a good goal, because if you just think about the machines we have today, if they were topological, we’d already be cracking encryption. OK, so maybe that’s a bad thing but —

In a way. it’s good that we’re not there yet, because encryption algorithm takes 20 years to ratify and to adopt. If we’re able to crack all of the RSAPs today, we’d be in a world of hurt.

Yes. So maybe it’s better that we’re not quite there. The first thing that’ll happen though, I do believe, is these really practical use cases, once we reach that quantum advantage, which is 2023, that’s my opinion. I think that’s when we’ll have just enough hardware somewhere to be able to really dazzle the world with something practical.

I’m all down.

Yes. You did a great job explaining topological qubits, because that’s one of the things that, when I think of it, it’s like, “Oh, this might be the way that a company like Microsoft could just win the whole thing.” If you get this right, well, then, you’ve got a computer that’ll be hard to compete with.

It’s not really a win thing.

Yes, you know what I mean. It’ll be like —

I have a reason for saying that, but keep going.

Yes. It’s like an overnight leap to first place in performance if you have something like that.

Yes. What I wanted to elaborate on is, I said it earlier: With great power comes great responsibility, and I think that the reason why we’re in such a race is because we realise that the first person there, if they’re not doing things in an ethical way, it could be pretty disastrous. The race is on, mostly to ensure that a malicious nation-state doesn’t use it for really terrible things. I think that that would be a more valid reason. I think the world needs to learn together on this problem. We can’t hoard the security secrets — that’s the big point. As we learn how to secure these things and how to make them operational, we have to share that with the world, because otherwise, it gets to be a pretty creepy place pretty fast.

Yes, well put. In the interim, I know you guys are working with the University of Sydney on CMOS.

Actually, I forgot on that one, who we’re working with, but we’re working on a cryogenic CMOS, which is really important. Do you want me to describe it?

Yes, absolutely. Thanks.

Today, if you look at one of these quantum computers, there’s a great one that Google posted, and I’m not picking on it — it’s a beautiful picture — but if you look at the cryogenic plane where all the computation is happening, what you see is hundreds of wires that are being fed into that quantum computer in order to feed it to the classical computer where we output the data. Anytime we introduce any change in temperature, we have higher decoherence.

One of the things that we worked on is this idea of a cryogenic CMOS that — if you’re familiar with quantum computers, they operate at a colder temperature than deep space. So, if you’ve seen Guardians of the Galaxy, you saw that when Yondu got shot out of that spaceship, he froze instantly. That’s a thing. That’s a really cold temperature. Quantum computers are an order of magnitude colder than that. They operate at something called 15 millikelvin. Anytime you introduce temperature into that equation, you lose qubits by just decohering.

We have been working on our cryogenic CMOS, which operates somewhere around 100 millikelvin, and by introducing something that operates at that low of a range, we decohere less particles. It’s essentially the result. Now, what that means in the physical sense is that you would go from hundreds of wires connecting to that cryogenic plane maybe down to three. Once again, each of those wires has temperature that’s above the cryogenic plane that’s operating. Something substantially higher. I forgot what the exact number is — I think it’s 400 kelvin or something like that. In doing so, you keep the temperature down. Therefore, you keep the decoherence down, and therefore, you get better results out of your qubit.

How did you anticipate that coming to market? Do you see that as being the first entry into Microsoft-owned hardware inside of the quantum space?

Well, I don’t know about the owned part. I’m not sure, actually, to be honest with you. I’m not sure about it.

Ok.

If it’s something that we planned to say— “We’re going to build these things, and you’re going to buy them from Microsoft” — I don’t know what our plan is there, to be honest with you.

Well, I just meant like keeping it in the cloud — it’s your contribution.

Yes. If you look at the picture of the cryogenic CMOS, literally, it would sit within an inch or two of the actual cryogenic plane, where you would have the topological qubits, and I’ve got an image of this. If you look for Mark Russinovich’s presentation at Ignite — and it’s somewhere around the halfway point — you’ll find this image that he’s got of this. What it means is, it sits on the same silicone, essentially, as the cryogenic plane, so it’s really close — within inches.

Yes. I could put that in the show notes if you sent that along — that would be great.

Yes, I will, for sure.

Thank you. Have you seen anything that worries you right now in this space?

Well, beyond that with great power comes great responsibility?

Yes.

Concerns? I don’t know. I think everybody’s going at this — they’re racing. We’re all racing to this problem. The reasons for racing are multifold. One is, obviously, there’s a race to revenue, but there’s also a race to be able to solve some of the world’s toughest problems, like, could we find a way to create a material that absorbs carbon monoxide from the atmosphere, the CO2? Is there a way that we could create fertilisers faster? Is there a way that we could react to viruses like COVID quicker?

Each of those things are particularly challenging for classical computers because it has to walk through a computations format that called FIFO, which basically means “First in, first out.” With the quantum computer, we don’t have to deal with it — it’s basically all in, all out at the same time. If we have more qubits, we get answers tens of thousands, if not millions, of times faster than with a classical computer in some cases, depending on the number of qubits that are stable. I think that that’s where our heads are at right now —trying to figure out the technological challenges. And then, like I said, different elements of science are focusing on different things.

So, we really do have a great R&D team that’s just focusing on cryptography. We have other ones that are focused on algorithms. We have another one that’s focused on majoron capability — the topological qubit. There are so many elements of science involved here, and this is the benefit of Microsoft, as we’re a pure science company. Pure science— meaning, we don’t just do science to create a phone or create a device. We actually have science to figure out science, so we’re no different than an MIT in that sense. There’s multiple people racing to the problem from different angles.

One thing I noticed recently as this field is exploding: Let’s say the free machines that are available — so, IBM, they have 20 machines, and 10 of them are freely available. The queue times as we record this episode are pretty much blown away now. You’re lucky if you’re getting access to a machine certain days now if you’re in the free tier. Obviously, a lot of people are interested in this demand, and I’m assuming you have to pay to access because they’re other people’s machines. Do you see others coming online soon to meet the demand? I feel like even the pay ones are going to start to get pretty scarce.

Yes. It’s supply and demand. There’s a lot of pent-up demand because of challenges like COVID, for example, or material sciences, for example. I don’t think that the problem is going to go away anytime soon, because there are so many underlying technological problems, and they’re really expensive to build. They take a lot of people. You’re not going to have one of these in your basement anytime soon. I see that getting probably a little bit worse before it gets better. It depends on the amount of demand, frankly. I don’t have a crystal ball into that. If we can solve for this problem of decoherence, what that means is that we can take what was the size of a conference room and make it the size of a refrigerator, and if it’s the size of a refrigerator, we can make it something smaller than that. Those are real benefits — that’s a real bonus.

Are there new partnerships that are going to be announced in the future?

In my world, people assume that I have a lot more information than I give. No.

OK.

I don’t have that crystal ball, and even if I did, I probably wouldn’t be able to divulge it. I don’t know, but I just will say, you said 2023. I don’t know — we’ll get there when we get there. I’m optimistic. I was optimistic four years ago, when I was telling people that they should pay attention to this technology too. You may have a few steps forward. You might get a few steps back. The big thing that I worry about is that without the guardrails in front of the system, I think it’s going to be a rough go, and there’s got to be a lot of thought around what those guardrails look like. How can people use them? How can we ensure that when somebody does hoard the machine — or because they’re paying a lot — that they’re doing things that we want to put a quantum computer through and not doing it for something malicious that harms us all, or harms a percentage of people.

Yes.

Just because you can rent one doesn’t mean that that it’s being rented for the right purposes. “Is there any altruism in that conversation?” I think is the right way to pose it.

Depending on at what point they’re accessing that backend target, you might not even be able to discern what’s being done. When it gets down to that low level, it’s hard to say what those calculations are really accomplishing.

Right. In Microsoft, you have that whole organisation called Aether, which uses the principle of FATE, which stands for “fairness, accountability, transparency and ethics,” and they look at the problem and say, “For the best interest of the company for the next 20 years, should we do this?” As we create some of these crazier quantum technologies, that Aether committee is really important. (I always forget what the acronym means, because it’s really long. It’s like “Ethics and AI,” and then my brain stops.) The idea there is that they’re a committee that tries to understand fairness, accountability, transparency and ethics, and that question “When will I be able to get them? Is the access equitable? Is it fair? Right?” Regarding accountability, do we know that they’re doing it for some good purpose, or is it for some nefarious purpose? How will we know? Is it transparent, what they’ve done? Then, is it ethical? So, we’re looking at that lens. I think that the question is, “How do we formalise those guardrails inside of our platform?”

Yes. There could be this dangerous period of time — it might not be very long, but where the machine access has gotten cheap enough and they’re powerful enough that someone could figure out a way, with just an hour of time, to do a whole lot of damage before anyone even realizes it.

Perhaps.

I don’t think that’s going to last very long if it happens, but it is feasible.

Well, we’re always ahead of where governments are, but this is one of those things where because we’re in the United States, then, ideally, you would think or hope that we would be a banner for what the right process should be. I think that we need other people to gut-check that. Would Germany agree? Would Ireland agree? Would France agree? Would the U.K. agree? Would Australia, China, all through Asia — would they agree? So, it is a gut-check against the world.

It makes this feel like a new kind of superpower.

That’s right.

Quantum superpower.

Pretty much like Spider-Man, yes.

Yes.

I’m excited. I really do think some amazing things are going to come out of this. It’s just like regular computers: People have used them to do some amazing things, and then your heart breaks because of some of the ways that it’s hurt whole communities like children, like people who don’t have access to the technology. It’s a barrier for them. You would hope that we would not put our hand on that same hot stove twice, like we did with classical computers. Time will tell.

Yes. The cloud access to these machines is great. It’s like the great democratiser at some level. Even if you can’t access one of the pay ones, you can still play around in the environment, and you guys did a really good job with the Quantum Katas, like trying to make it easy to learn through actual practical applications. I was pretty impressed with the way it rolled out.

That’s cool.

I’m hoping to see some new hardware come online soon.

Yes. We’re all kind of hoping, yes, for sure

I know there are lots of companies building machines, so I wouldn’t be surprised if they ended up being backend targets.

I’m more excited about all of them. I think that it’s not about being right — it’s about getting it right. We’re going to make some mistakes. The question is, how long will those mistakes persist? Will we make them over and over again? We have to do it quickly. We have to do it fast. We have to do it quickly. We have to do it together.

Absolutely. Well said. Yes, you guys could check out the show notes to learn more about how to start playing around with Azure Quantum. Once again, thanks so much, Paul, for coming on. I really appreciate it.

Well, thank you, Konstantinos. It was cool. Hopefully, we could do it again.

That does it for this episode. Thanks to Paul Edlund for joining today to discuss Microsoft’s Azure Quantum, and thank you for listening. If you enjoyed the show, please subscribe to Protiviti’s The Post-Quantum World and leave a review to help others find us. Be sure to follow me on Twitter and Instagram @Konstanthacker. That’s “konstant with a k, hacker.” You’ll find links there to what we’re doing in Quantum Computing Services at Protiviti. You can also find information on our quantum services at www.protiviti.com, or follow us on Twitter and LinkedIn. Until next time, be kind, and stay quantum curious.