Screaming in the Cloud - The Inevitability of Quantum Computing with Dr. Sarah Kaiser

Episode Date: February 4, 2021

About Sarah KaiserI use lasers to melt acrylic and the cisheteropatriarchy alike. Quantum Computing technologist/consultant by day, author and dog mom the rest of the time.Links:Unitary Fund: https://unitary.fund/Sarah’s Twitch: https://www.twitch.tv/crazy4pi314/Learning Quantum Computing with Python and Q#: https://www.amazon.com/Learn-Quantum-Computing-Python-hands/dp/1617296139Twitter: https://twitter.com/crazy4pi314GitHub: https://github.com/crazy4pi314Personal Website: https://www.sckaiser.com/

Hello, and welcome to Screaming in the Cloud, with your host, cloud economist Corey Quinn. This weekly show features conversations with people doing interesting work in the world of cloud, thoughtful commentary on the state of the technical world, and ridiculous titles for which Corey refuses to apologize. This is Screaming in the Cloud. needed, so they built one. Sounds easy enough. No one's ever tried that before, except they're good at it. Their platform allows teams to create consistency for the entire incident response lifecycle so that your team can focus on fighting fires faster, from alert handoff to retrospectives and everything in between. Things like, you know, tracking, communicating, reporting, all the stuff

no one cares about. FireHydrant will automate processes for you so you can focus on resolution. Visit firehydrant.io to get your team started today and tell them I sent you because I love watching people wince in pain. This episode is sponsored in part by LaunchDarkly. Take a look at what it takes to get your code into production. I'm going to just guess that it's awful because it's always awful. No one loves their deployment process. What if launching new features didn't require you to do a full-on code and possibly infrastructure deploy? What if you could test on a small subset of users and then roll it back immediately if results aren't what you expect? LaunchDarkly does exactly this.

To learn more, visit launchdarkly.com and tell them Corey sent you, and watch for the wince. Welcome to Screaming in the Cloud. I'm Corey Quinn. I'm joined today by Dr. Sarah Kaiser, who is very recently a technical staff member and quantum community lead at Unitary Fund. Sarah, welcome to the show. Hi, Corey. Thanks for having me. So there's a lot to unpack. Let's start with the easy stuff. What is Unitary Fund? Yeah, so we're actually a 501c3 nonprofit that is invested in trying to grow the quantum open source software community. So we kind of do two main things. We give out micro grants to help support maintainers on projects or other sorts of community groups or educational projects that kind of help grow the community of quantum developers. development team, and we work on building up software projects in gaps that we see, you know, that maybe are not as interesting or kind of boring, but actually are needed to overall help

the ecosystem grow. So let's start, I guess, with a big meaty topic, quantum computing. It feels like it's something that we've been hearing about for 20 years, usually in almost the same sense as we have cold fusion, which is a practical X is always 20 years away. And over time, recently, we've started seeing actual products and services come out from companies that are talking about exciting breakthroughs from a quantum perspective. But the challenge I've always had is, first, what is quantum computing? Every time I've tried to delve into it, it seems that, oh, just go through the hello world example. And the challenge, of course, is that hello world in quantum computing is basically a field. And in this case, in a literal product sense, it's a hardware accelerator for computation. So in the same way that we have like GPUs and

FPGAs that are, you know, bespoke custom design hardware that can accelerate different, you know, whether it's machine learning or graphics processing, quantum computers are, I kind of loathe that they are called computers. They really should just be called quantum hardware accelerators. Right. You start calling them computers. I start wondering, okay. And I scroll them and, huh, they're not for sale at Best Buy.

What's the deal here? Exactly. You're not going to check your email. They're not going to, importantly, they're not going to replace all computers. Well, not with that attitude. Again, not to sound cynical here, but I can look at some of these electronic stores and they're selling, you know, $200 audio cables because it's better than the $50 one and it's passing nothing but digital signal anyway. So if people will buy anything that's hyped well enough, I mean, there's certainly been enough hype poured into quantum computing to the point where, at least from where I sit, it's occluding what it actually is and what

it's capable of. Yeah, I think I'd entirely agree with that. I think what is important, you know, as someone who has spent over 10 years of their life researching and working in this field, I still do think there still is really interesting and cool stuff here. It's just maybe not exactly the features or things that are hyped for venture capital funding. Yeah, part of the challenge with raising VC for something like quantum is that the immediate short-term returns are hard to demonstrate,

and I'm being very charitable with that. Not from a success or a breakthrough story, but from a position of economic viability. It seems that very often one of the challenges with quantum computing and explaining what it is, is even articulating the type of problem that a quantum computer is built to solve. Is that a fair assessment or am I radically misunderstanding something? Probably both. No, I think you're actually pretty spot on there. One of the ways I like to think about it is,

you know, we have a pretty clear description or box that we can say, these are the types of problems that say a GPU is good at. We know that any problem that's highly parallelizable, you know, if we can make the solution to our problem fit in those sort of constraints, yep, throw it at a GPU, we're good. For quantum computing, where we're at is we basically have examples of things that might be in that box. Like we know we can speed up this specific problem, we can speed up this specific problem, but we haven't really worked out what the generalization of, you know, the class of problems or the types of problems that we might be able to solve here. So it makes it hard to say like, well, yes,

here's your arbitrary problem. I actually have to sit down and work through and try to figure out a specific solution to that problem, as opposed to being able to have a general framework like parallelization or something like that to break your problem down into something that I can use on the device. Please don't take this as the deadly insult that it probably comes across as, but it feels similar in some respects to machine learning, where there was a lot of excitement about it, but every time that someone tried to articulate the real world business value, it was either aimed at an incredibly specific business use case that is only going to work within one or maybe two companies, or alternately, it came across as completely ridiculous. The example that springs to mind for that is when WeWork, back when that

was a thing, talked about using their machine learning algorithms to determine that there was congestion in their lobbies at certain times. So they wound up bringing in a second barista during that time window. And it's, let me get this straight. You spent how much on data science to figure out that people like to drink coffee in the morning? Yeah. Have you ever talked to a barista and figured out maybe there's a pattern here that a human could discern way faster, it feels like it either lends itself to mockery or to extreme niche use cases. And I know that's wrong, but that is the impression a lot of folks are left with. Is quantum in the same boat?

So I think actually in some senses we currently are, in that kind of like I was saying, we don't really have a good way of generalizing what types of problems are suitable to speed ups on our quantum devices. Because like, having a GPU, that doesn't mean it universally speeds up everything on my computer. Like if I'm IO bound or something like that, it doesn't help. Same with a quantum computer, I can have one, like even if you literally gave me one today, descended from the heavens, it was perfect, error corrected, I honestly wouldn't know what to do with it. Because we really haven't had a chance to try out larger applications. But I don't, from my perspective, and you know, maybe it's just in that I've been doing research on this for a long time, but I think that doesn't necessarily

preclude that there won't be. There are no no-go proofs that we won't be able to find other interesting things. And what I think, to me, at like a almost romantic level, what is really beautiful about quantum computing is that it's an entirely different physical resource that we're actually using to do the computation. FPGAs, GPUs, they're all like at some level the same sort of transistors on silicon that, you know, at some level function the same ish way. Maybe we put them together differently, but they're the same Legos here. This is now we got connects. So, you know, we haven't been thinking with connect brains for a long time because we've been building with Legos. So, you know, it is kind of hard to actually find what maybe we can build with Kinects. And that's what I'm really personally excited about exploring and using, honestly, the

we need to build up the hardware, of course, but that's where I actually see quantum software as being a really exciting kind of emerging discipline here where we can actually start exploring Kinect type solutions, but all in software. That on some level almost seems to lend itself to another comparison, which is something you happen to be renowned for. Specifically, whenever we look at the industry and what they're doing with AI and machine learning, people are starting to find actual use cases for it. And that's exciting and that's great. And that use case is invariably some form of bias laundering where I put my biases in, the algorithm does this thing as if that somehow absolves me of all responsibility, and then it spits my own biases back to me. But now I'm considered to be somewhat blameless.

The ethics of quantum computing feel like they're still far away as the actual underlying technology gets built out. But you've been talking about it a fair bit. Tell me more. Yeah, machine learning is a really apt comparison, I think, here, because it is exactly a form of bias laundering. And as someone who's excited about technology and always is exciting about building it, I always try to keep in the forefront of my mind and in forefront of our discussions, like there's, can we do it? And then there is the other question is, should we do it? And so I think, you know, quantum computing is a technology that does have the possibility of drastically changing our society. The computational power for at least the

problems that we've seen speed ups on is incredible. And so I have to really sit with myself and think about, you know, okay, this is great if I have it or essentially good people have it, but what could an adversary or what could a malicious agent do with this technology? And that's why I think it's really important to make sure as we're building out this technology, this community, this field as a whole, that we really try to involve as many people as possible and get as many people as developers. Literally full stack quantum computing is kind of a thing now. So we really need everybody at the table when we're making these decisions so it doesn't just kind of turn into a bunch of white guys at a table making a choice for everyone.

Sorry, the thing you just said about full-stack quantum computing is objectively terrifying on some level. It's one of those, oh great, you think boot camps are hard now because JavaScript doesn't make sense because it doesn't to me? Great, wait until we introduce the rest of this, and it becomes almost this, I guess, surreal vision of a possible future. Do you believe that quantum computing is a technical inevitability? I do. I think we will eventually. It's going to be a long pull.

I know even when I started grad school over 10 years ago, they told me it was 20 years away at that point. I think they still say it's 20 years away, as you said. I have no good idea about hardware timelines, but what I think is here and present, and actually we in the year of our Lord 2021 have a chance to actually influence and change, is how we're developing the stack around the hardware. Like I said before, if someone showed up and gave us a working hardware device right now, we wouldn't have the networking, the classical kind of dispatch. And that's basically where we're trying to build up.

How does quantum computing integrate with the rest of the stack? And honestly, really, the best model for it is as a cloud computing resource. So it's not going to be a device that you have in your house or as you put into your gaming PC build, but it'll be a thing that, you know, is offered and is currently offered actually from a lot of the major cloud vendors like AWS and Azure and whatnot. So I think that trying to figure out like what that looks like from a consumer standpoint is a really exciting and really cool place to actually make a difference. Whenever I start looking into quantum computing and understanding the various approaches to it, I know AWS launched their bracket service last year, which was interesting in that it, oh,

it's finally contextualizing this through the lens of something that I spend a lot of time working with. And I pulled it up. And honestly, I don't know if you're familiar with a subreddit, VXJunkies or not, but, and I'm telling a bit of a secret here, so I will deny this. Fortunately, it's just you and me and no one will ever listen to this, but the entire subreddit is built upon technobabble of explaining things back and forth that aren't actually real and people making up technical words. And it's incredibly convincing. No one is entirely sure when they first discover this whether it's real or not.

And it was a remarkable parallel for looking at what these things were. The terminology behind quantum computing is unreal. The entire methodology by which these things get addressed, the concept of qubits, and different types of quantum computers that require different aspects. And some of them apparently are, I don't know, liquid-fueled or something like that.

And at this point, it's one of these, is this just a giant attempt to have fun at my expense? Because honestly, if so, you won, good work. Why is it so radical a departure from the world that most of us are used to? You highlight something that is kind of uniquely challenging about quantum computing. And I think it really comes from the fact that it is a really interdisciplinary field. At a minimum, if you were to sit down and hire a bunch of people to, in a closed room, build a quantum computer, you'd probably need a chemist, you'd need electrical engineers, you'd need mechanical engineers, you'd need physicists, you'd need computer theorists, you'd need mathematicians.

And something that I really struggled with through grad school is almost every textbook or resource you look at is a view of quantum computing from that field. All you're missing at that point is a bartender for the punchline. Basically. But yeah, like what you're seeing there is basically this amalgamation of jargon from four or five different distinct research areas and fields. And frankly, I feel like even the researchers in the fields, we name things magic states. A lot of our analogies and papers are about King Arthur. It's like the quantum Merlin-Arthur problem. There's at least some fun had with the terminology, but also, yeah, it's kind of a mess.

And so I've written a textbook on teaching quantum computing with Python and Qsharp. And one of the things I've tried really hard to do there is strip as much of that away as possible and use common language terms from programming to refer to what are effectively just names for particular types of matrices and stuff like that. My challenge, too, at least to my mind, is every time I step through this, it talks about things like running a quantum shot, for example, which, again, does not detract from the idea of having a bartender involved somewhere. But even the idea of doing something like that is bizarre to me because my problem is I cannot, in layperson's terms, come up with a reasonable explanation for what kind of problem would I have that this would solve. And I'm not even asking for a real business problem. That still feels like it's years away. I'm talking about things like, well, all right, you're going to learn to write code. So all right, we're going to make the program spit out, hello world. Cool. I can do that. Now we're going to make the thing count from one to 10. Awesome. Yay. I'm programming. And honestly, you are at that point. Sure. It's an elementary level, but that is fundamentally what it's about. A lot of things I've looked at, even their basic hello world equivalents are tremendously confusing. Is that just something I'm missing?

I don't think there's something exactly that you're missing there. What I think has been happening is a lot of the software and tools that we're developing for quantum computing right now is pretty heavily focused on bootstrapping up those initial quantum devices that folks are building right now. So, you know, we have some number of qubits that you can access from IBM or INQ or Honeywell or wherever. And most of the software that's getting written is really geared towards that, which it's kind of like trying to think about writing programs on your computer in machine code, because that's literally, you're thinking about gates. The programs are often called circuits.

Yeah. Assembly is a good analogy here. It feels a lot like the assembly class I took half of once and then immediately stopped attending because, wow, all right, my brain is full time for me to excuse myself. You're right, that feels very similar. Yeah, I'm not a firmware person. I don't really like thinking at that level. I'm much more comfortable in Python where I can just say, please give me a variable and I don't have to think about pointers or memory management or anything. You know, I'm really excited about building and thinking about what the tools for quantum computing look like at that level, at that level of abstraction.

And so the closest we have, there are some different programming languages that are being targeted more at the algorithmic level like that, like Qsharp and stuff like that, some of the Python tools that are out there. And that's where I think is a much better place for people to start with quantum computing because then I feel like that's more commensurate

with like what you would see with a Python or Rust or something like that. Hello world program, as opposed to here's all of the assembly instructions to say hello world on the screen. This episode is sponsored by ExtraHop. ExtraHop provides threat detection and response for the enterprise, not the starship. On-prem security doesn't translate well to cloud or multi-cloud

environments, and that's not even counting IoT. ExtraHop automatically discovers everything inside the perimeter, including your cloud workloads and IoT devices, detects these threats up to 35% faster, and helps you act immediately. Ask for a free trial of detection And that's part of the challenge, is that there needs to be at least some grounding at that level of technical competence. I would still argue as a result, and please feel free to contradict me on this one, that if people wind up coming at this without having some grasp of lower-level computing concepts and principles, they're likely to struggle. Is that a fair assessment? Yeah, just like in classical computing, we have developers at every level in the stack. We have people who are working on, you know, literal CPU instruction optimization sort of things all the way to doing web dev and database and cloud stuff. Right now, most of our tools in quantum computing

and honestly, most of the educational focus is all at that CPU assembly sort of level. And there, yes, it is more necessary to kind of know more about the hardware, to know more about like what the qubits are doing, because you're literally interfacing with it. I think it's a lot easier to start at a higher level. And it's kind of like if you want to learn a new Python package or something like that. I don't usually sit down and read through the API docs.

I will start with their very high-level example, and then as I need to understand things as I use it, I will go and dive in. I think you can take a similar approach to quantum computing where you say, all right, I want to actually start at this really high level. Like, let's talk about algorithms, let's talk about built in function sorts of things, and then drill down and understand kind of once you have that broader picture of what's going on. I think it's kind of like rather than starting zoomed in on a map on Google Maps at Street View to understand where you are in a city, it's much easier to maybe start zoomed out a lot farther. So I opened this episode by joking about the tutorial being a PhD. That is clearly a bit

above and beyond where we actually are in this day and age. But what are the realistic prerequisites? I've never been a fan of gatekeeping, and I refuse to accept the answers, you must have this degree from this university. Cool, then you need to get out of my office, because there's never just one path to anything. And I understand that there are absolutely prerequisites that in many cases are hard to find without very specific academic achievements, credentials, and prerequisite. But I don't know that a PhD is one that I would even accept. So

what is the real world limit of what you should know before diving into this space? First of all, I really do think anyone can actually be a quantum technologist or be a quantum software dev. Really, the most critical skill for working on this stuff is basically linear algebra. So if you can multiply matrices on a computer with whatever programming language, you can already start building quantum software, basically. I actually went into quantum computing. So I was in a regular physics track in undergrad. And then I saw all these like triple integral crap. And then I was like, this is really hard. I don't want to memorize any of this. And then I saw quantum was like, it's just matrices. And I knew how to make my computer. I knew how to make Mathematica

multiply those. So I didn't have to do it by hand. So like, I really think there's a lot of misconceptions about exactly as you say, gatekeeping, or you must be this smart to participate. I regularly now in the open source community work with a ton of folks that have no background in quantum, they were actually like a web front end dev, and they're helping to make contributions to these quantum open source projects and tools. So my personal belief is anyone can be a quantum developer. And I would hope people take me up on that and take a look at, you know, some of these higher level kind of approaches that really, like linear algebra,

a little bit of statistics is nice. But honestly, that's where having software is helpful, because it'll just do that for you. You don't have to think about the details of exactly what's going on there. As someone who basically capped out at pre-calculus, to me it sounds, oh, okay, this is not going to be accessible to me without a whole lot of study and planning. But the reason I bring that up is not for pity or for you to, oh, no, it'll be fine. And then I go in and it is very much not fine. But to point out that I believe this is like almost anything else in technology across the board, which is today it might be beyond my capability of easily getting into and assimilating. But the bar always gets lower, never higher.

Things simplify over time. It used to take three weeks to effectively get a web server up and running. Now it requires basically a passing thought or a checkbox on a website. It gets easier with time. So my question for you is, do you have a ballpark and very general predict the future senses of when this starts becoming more accessible to more people without the either math background or math focus? And I understand that's an incredibly loaded question. Yeah, I straight up generally refuse to answer the question of when are we going to have a quantum computer? But I think about now how easy it is for me to like use PyTorch or something

like that to do machine learning sort of things. I can, you know, in one or two lines with a folder full of pictures of my dog, get it to train on my dog. That's the kind of accessibility that I really hope, kind of as you were describing, that we can get to with quantum computing. And I really do think that in probably the next five to 10 years, we can get the software there. Whether we have hardware necessarily to back it or not, that is sufficiently large for what people want to do. I honestly am never worried about and frankly don't care. They're working on it. They're engineering problems. They'll get there when they get there. You know, it took how long to get transistors from the giant triangles of lead down to what's sitting here in my PC next to me. But the software and kind of like that user experience or what does it mean to actually use this technology is somewhere that we

can make huge strides in the next five to 10 years to have an experience kind of analogous to checking a box or whatnot to add whatever it is, quantum machine learning or whatever, to your projects or whatnot. So you said that you don't ever accept or answer the question of when are we going to have a quantum computer? And that's fair. But let me see if I can sort of do an end run around that. What do we actually need in order to make quantum computers practical? And you can, of course, solve for practical however you'd like. Sure. So where we're at right now is basically we have a bunch of different kind of competing types of technology. Like you were even talking about the liquid run ones that

possibly you were meaning the ones that you have in a bath of liquid helium or nitrogen. But, you know, we have superconducting qubits, we have ion-trapped qubits, we have optical qubits. There's lots of different options. And basically there's five criteria that we need to have for it to be a good scalable type of device. Each of those technologies usually meets three no problem. Then there's one that's an engineering stretch, but we mostly got in hand.

And then there's one that's like kind of an open question. And so basically where it seems like we've landed is superconducting is probably, at least at the moment, superconducting and ion trap technologies are kind of the leading candidates, but mostly we just need time. The nice thing that these devices that they're currently kind of pursuing can leverage is all of our experience building all of the silicon manufacturing infrastructure. Obviously, we're pretty good at that. That's all of classical computing. And so we can leverage that for miniaturization. And really what they're kind of iterating on right now is reducing noise.

So quantum devices in general, to stay quantum, have to be isolated from the environment. And so it's just working on progressively better isolation. Which sounds increasingly hard to do, given that we can't effectively handle isolation even in a conceptual sense. When we look at things like, oh, data security of, oops, did I accidentally turn the database backups into a web server with the wrong mouse click somewhere. It sounds like getting stuff like that separated out but still usable, is that as heavy a lift as it sounds like? Yep, pretty much. And honestly, that is kind of one of the most interesting questions to me. Having done some research on it, like quantum machine learning is a thing.

We've found algorithms that can help us speed up certain sort of machine learning tasks. But the problem is only any advantages we find that at an algorithmic level, there are entirely blown away once we use like basically load the data, the data, like transferring classical data into the quantum computer for it to operate on it. So those sorts of protocols that people usually just, let's assume we have that. We need to fill in the homework. We need to fill in those answers

before we can figure out some more applications. At some level, it sounds unsatisfying, but the answer is it's still a work in progress. I will say that it's interesting to see that even as early days as it is, you're still focused squarely on the ethics piece of it. Out of curiosity, is this something different than what we saw with the rise of things like machine learning? Or were there folks early on in that process as well,

talking about the ethics and thinking about the bigger philosophical questions? In other words, do we have a better chance now of avoiding some of the pitfalls that we keep smacking into as a society because we didn't pay enough attention the first time? I would like to hope, but I honestly don't think we are learning fast enough. I mean, it is early days, but what I've, even in the course of my career, seen it shift strongly from being an only academic pursuit to now a very largely industrial, most everyone I know now works for companies working on this stuff. They're not postdocs. They're not professors. And that gives me some hope because, weirdly, in general,

I think companies are better at being ethical than academic institutions just because they have lawyers. But I want to hope that we can do better, but right now I don't think we're on a better track, honestly. Well, I have a serious problem ending an episode on that much of a downer, so let's ask one more question that expands on something a bit more hopeful.

If folks have listened to this episode and don't have the shrieking aversion going back in time 20 years to struggling with math class in high school or whatnot and think, I actually would like to get started with some of this, where would you recommend that they start? a version going back in time 20 years to struggling with math class in high school or whatnot and think, I actually would like to get started with some of this. Where would you recommend that they start? Self-promotion-y, come chat with me on office hours. So I stream a lot on Twitch, both just kind of working on quantum open source projects. And I also do office hours where people

can come just ask me whatever questions you have about quantum computing or just kind of tech stuff in general or crazy stories about what we blew up in the lab. There's lots of good resources like myself on Twitter and to just kind of like actually interact with the people who are currently building this stuff because I think at a personal level we're probably different than who you might expect is actually working on the technology. As I mentioned earlier, I also have a textbook, or it's not a textbook, really. It's more of like a kind of, I think, PowerShell in a month of lunches sort of format. But it's called Learn Quantum Computing with Python and Qsharp. It's basically geared for your average sort of dev. Helps if you know Python, but you don't strictly have to know Python. Just any sort of programming experience is good. There are tons of good open source sorts of resources out there.

There's good awesome lists, stuff like that. The main thing I will caution is guard yourself against the hype. Kind of like how we opened the episode with. There is a lot of hype out there that we you know, we've solved time travel with, you know, quantum healing crystals, whatever. Bring your best rational sort of logical skills when kind of exploring some of that stuff and you will gain the most. I think that is a much more uplifting vision for the future than a dark cloud over the future of humanity. But that seems to be the

season for either one of those now. People get to choose their own adventure on this one. If people want to learn more about what you're up to specifically, where can they find you? You've already mentioned your Twitch stream, but where else? Yep, I'm on Twitter a lot. So my handle there is crazy, the number 4PI314. I did pie memorization contests in high school. So it was my first internet handle. And that's pretty much what I am everywhere on GitHub, on Twitter. You can also find more about what I'm doing on my website, sckaiser.com. And we'll, of course, put links to that in the show notes. Thank you so much for taking the time to speak with me. I appreciate it. Yeah, this has been really fun. And I hope folks are interested to check out some quantum computing stuff

and come make fun of it on Twitter too. Sounds good. Thank you once again for your time. Dr. Sarah Kaiser, Technical Staff Member, Quantum Community Lead at Unitary Fund. I'm cloud economist Corey Quinn, and this is Screaming in the Cloud. If you've enjoyed this podcast,

please leave a five-star review on your podcast platform of choice. Whereas if you didn't like this podcast, please leave a five-star review on your podcast platform of choice, along with an angry, incoherent, misspelled comment telling me why all of this stuff is wrong and you need to have a PhD

in order to approach any of this. This has been this week's episode of Screaming in the Cloud. You can also find more Corey at Screaminginthecloud.com or wherever Fine Snark is sold. This has been a humble pod production stay humble