What Is Qiskit?

Blake Johnson sees a gap between the circuits classical computers can still simulate and the size needed to prove quantum does better. There are some people that believe that there's nothing in this chasm. I think these people are crazy.

Thinking On Paper

Clips and outtakes on Instagram ›

In this episode · 14 chapters

  1. What the quantum stack is
  2. Making it familiar
  3. Why Python
  4. Is it useful for someone like me?
  5. Spin glasses
  6. Error rates
  7. Qiskit as a toolkit
  8. Simplifying the numbers
  9. Driven by empiricism
  10. Width, and depth of the circuit
  11. Error correction codes
  12. The function library
  13. When we will have applications
  14. Should they be worried?
The problem with the surface code, though, is that it’s very inefficient in terms of information encoding. In fact, you need just an extraordinary number of qubits in order to achieve computationally relevant error rates.
Blake Johnson, IBM Quantum

Blake Johnson works on the quantum engine at IBM Quantum, the part of the software stack that runs a query on IBM's quantum computers. He explains why Qiskit began as a pure Python library and now runs on a Rust core behind the same interface.

He compares error rates in classical and quantum machines, explains circuit width and depth, error correction codes and IBM's roadmap to 2033, and says quantum could be unequivocally the best solution for some problems within five or six years.

Blake Johnson, quantum engine leader at IBM Quantum, explains Qiskit, IBM's software stack for building and running quantum circuits, on the Thinking On Paper podcast. He argues that useful quantum applications will be found by experiment rather than mathematical proof, and that IBM's job is to give scientists tools native to their own fields to run those experiments.

What is Qiskit and how has it changed?

Qiskit, IBM's software stack for building and running quantum circuits, started in 2017 as a pure Python library and is being rebuilt around a high-performance Rust core that keeps the same Python interface. Johnson says the tools can no longer be allowed to get in the way as problems grow. Anyone can try it through a free cloud plan, which he says over 600,000 people have joined.

Are there useful problems before quantum advantage?

Johnson believes there are useful problems in the gap between circuits classical computers can still simulate and circuits large enough to prove quantum advantage. “There are some people that believe that there's nothing in this chasm. I think these people are crazy.”

How does IBM plan to reach 100 million quantum gates?

IBM measures progress by how many two-qubit gates a circuit can run reliably. Johnson says the plan is to reach 15,000 gates by 2028 using error mitigation, then move in 2029 to error correction, which spreads information across several physical qubits and fixes errors in real time, to reach 100 million gates.

What are IBM's LDPC error-correcting codes?

IBM's development plan relies on a new family of low-density parity check (LDPC) error-correcting codes found by its team. Johnson says they need 14 times fewer physical qubits than the surface code, the approach the field has studied for decades.

How is IBM helping scientists who avoid writing quantum circuits?

IBM is building tools for scientists who do not want to write quantum circuits. They include a version of its Granite AI model trained on Qiskit to help write code, and a Qiskit Functions Catalog where partners offer functions driven by a plain problem description; in one optimisation function, Johnson says, “nowhere in that process do you see a quantum circuit”.

Where Mark and Johnson disagree

Mark says he has seen many roadmaps that “never, ever seem to follow the steps” and asks what IBM will actually deliver by 2033. Johnson answers that applications open up as the size of reliable circuits grows, and hopes that by then uses for quantum computing will be “broad and easy to find”.

Questions

What is Qiskit?

Qiskit, IBM's software stack for building and running quantum circuits, started in 2017 as a pure Python library and is being rebuilt around a high-performance Rust core that keeps the same Python interface. Johnson says the tools can no longer be allowed to get in the way as problems grow.

Who is Blake Johnson?

Blake Johnson, quantum engine leader at IBM Quantum, explains Qiskit, IBM's software stack for building and running quantum circuits, on the Thinking On Paper podcast.

How will useful quantum applications be found?

He argues that useful quantum applications will be found by experiment rather than mathematical proof, and that IBM's job is to give scientists tools native to their own fields to run those experiments.

How can you try Qiskit for free?

Anyone can try it through a free cloud plan, which he says over 600,000 people have joined.

Transcript

Jeremy Gilbertson Disruptors and curious minds. Welcome to another episode of thinking on paper. My name's Jeremy. This is Mark. Each week we get to talk to the founders, the builders, the creators that are architecting and building and hopefully delivering new versions of systems of the world that we will get to experience hopefully at one time in the near future. Um, we have a great guest today. We've been talking about quantum. It's been quantum season for a really long time. We are going to take it from this ethereal in the air discussion. We're going to pull it down. Like my grandfather used to say, land the damn plane, right? We're going to land the plane for a bit today.

Jeremy Gilbertson Get a little tactical and figure out, um, how people are accessing quantum, what they're doing with it and what, uh, some folks like our guests, uh, what direction they're headed in. So Mark, what do you got for me?

Mark Fielding A very special guest. Um, can I just say, if you listen to this, please like subscribe, share, so we can get more incredible guests like we have today with teasing who we have today. But if you're interested in quantum, this is going to be it. Please join our book club as well. We're reading Nexus next, um, currently reading quantum supremacy by Michio Kaku. So, um, the Dunning-Kruger effect in full of steam, we've learned some words. We're dangerous. We know some terminology, but hopefully today's guest, uh, Blake Johnson, quantum engine leader, IBM quantum is going to help us give a more tangible, um, use case of quantum.

Mark Fielding And the main talking point today is going to be Qiskit, the SDK, uh, Python framework for building and optimizing quantum circuits, which for you and me, Jeremy, basically means it's going to be easier for classical computer developers to create code for quantum computers, because, um, there's no point having a spaceship if there's no one to fly. And there's no point having incredible quantum computers if there's no software to run on it. So we're going to be learning about that. Um, what else? Um, the IBM roadmap from Heron and 132 qubits in 2024 to 2033, a thousand qubits and beyond, we're going to be talking about hopefully Watson X and AI using generative code developers in quantum, all of it, Jeremy. It's a big day.

Jeremy Gilbertson It's a big day. Let's bring on Blake Johnson. Hello. Welcome.

Blake Johnson Hey, how's it going guys?

Jeremy Gilbertson It's going great. It's going great. We're, we're glad you're here. And we're, we're excited because we want to, we want to get a little bit more tangible with quantum. I think, you know, we've, we've laid some groundwork on, you know, just generally what quantum physics is from a wicked high level, right? What quantum computing is from a very high level, very high level. We're not, there's no whiteboards behind us. We're not doing any equations, but we want to take it, take it and get tangible and what your team is doing and what IBM is doing to make quantum computing accessible to researchers, to people that are trying to improve quantum computing. So maybe if you could just give us a short little background of what you're involved with today, and we can kind of go into some questions.

What the quantum stack is

Blake Johnson My piece of the pie is, I mean, this thing that we call the quantum engine is basically the part of our, our software stack that's kind of responsible for actually executing a quantum, quantum computing query on one of our quantum computers, either in our fleet that's in our data center. We actually have two data centers, one now in New York and another that just opened in Germany, or dedicated on- premises systems at various sort of partners and customer sites around the world.

Blake Johnson So, you know, when you, when you submit a query to a quantum computer, as you guys have just said, you have to construct a quantum circuit and some sort of question of, regarded to relate to the output of your, of your computation, but it's not yet in a form which actually executable in sort of the native language of, of, of the controller that, that, that operates the quantum computer. And so we have to package it up in the right way so that it runs efficiently. And that gives you an answer that would, would hopefully the best sort of performance and accuracy that we can.

Jeremy Gilbertson Got it. So we have, so we have these, IBM has these two data center facilities, quantum computing infrastructure in both of those that is accessible externally through the, through the cloud, I guess. Correct. And people can kind of come into, you have, tell us about this quantum computing network where you have some of these external organizations too, that are kind of in this giant quantum experiment together with you, is that correct?

Blake Johnson That's right. So, I mean, as you said, like the quantum systems are available over the cloud. There's a sort of a free access plan that anyone can sign up for and we've had lots of people sign up for it. I think it's now over 600,000 people, if I have my number right. And so, I mean, it's, it's, it's touched a lot of people already at this point, but yeah, we, we, yeah, go ahead.

Mark Fielding So 600,000 people have signed up to, to use a quantum computer. So if I go and do this tomorrow, because now you've, you've piqued my interest, what will I experience and what will, if a company's watching this, what will they experience?

Making it familiar

Blake Johnson So, I mean, the, the experience we've tried to make familiar, right? Like people are already using a lot of computing technologies and cloud computing technologies, right? So whether or not they're computing locally on their laptops or using some virtual machine on AWS, Azure, or IBM cloud, we've tried to build an experience, which is very much like that, right? So you have some sort of Python, we have a Python framework. We build a Qiskit, which is a tool that, that helps you in this process of constructing quantum circuits and packaging them up to, to, to send to our service, but it feels, you know, the kinds of things you're constructing are novel and different, but like the experience is somewhat, we hope, familiar.

Blake Johnson And we build a lot of content learning material that, that helps you so that you're, we're not just throwing you into the deep end, right? There's a, we have a course material from sort of beginner all the way up to sort of graduate student, student level, grad school level, material on quantum computing that can take you through this experience and also different modalities, whether you like to watch YouTube videos or you like things which look more like a traditional textbook or problem examples. So there's really a wealth of resources.

Jeremy Gilbertson And also to touch on too, I, we did a little bit of background research that, you know, Python, anyone who kind of understands Python or Rust on the other side, cause I know there's some things you're, you're doing incorporating some of the speed of Rust, but some of the ease of use of Python, right, to, to kind of make this whole engine work. Is that the intent?

Why Python

Blake Johnson Uh, so the, yeah, absolutely. Yes. We've been incorporating Rust into, into the stack. Uh, so, uh, Qiskit is, is our label for our entire software stack. And one of the topics I hope we can touch on, uh, today is the way as much Qiskit has evolved and grown. But the kind of at the core of Qiskit is, is our SDK. Um, and that, uh, is a project that we started in 2017. We released the 0.1 version of Qiskit. Um, and it started out as a pure Python library.

Blake Johnson Uh, and that choice of like Python was, you know, partly driven by the popularity of that as a programming language and particularly around, among, um, the audience we were trying to touch on, uh, first, right, like, uh, our, our first audience were sort of other researchers, other people that physicists, computer scientists, quantum information theorists, uh, that had a background in quantum already, but wanted, uh, you know, to be able to touch hardware for the first time. Um, but the things that were kind of prized in that initial phase was, uh, kind of the, you know, the flexibility and adaptability, right. It is a tool for learning. Um, and while obviously we can't leave the flexibility and adaptability aside, we're still, uh, learning is a hugely important part of these experiences.

Blake Johnson You get people familiar with a new paradigm, a different way of computing. Um, but at the same time, you know, we're no longer working with five cubic quantum computers. These are machines that are capable of computations, um, that we call sort of utility scale computations, which ones are, that are not, um, amenable to brute force classical simulation. We're now comparing instead to alternative classical means of calculating, uh, the response to some of some, some of some system. Um, so we're not, you're no longer, uh, you're no longer testing what happens by trying to simulate the quantum computer. Cause you can't, um, right. So it's, it's this, this new regime. Uh, and so that the problems that we're working on are larger, right.

Blake Johnson Uh, and it's no longer sort of acceptable for the, the tools we use to interact with the quantum computer to get in the way to present new barriers to doing useful work. Uh, and so that's what's caused us to, to shift our attention, to really sort of dig into the performance of, of Qiskit itself and make sure that it really excels. Uh, and so that's where we started this project, uh, two and a half years ago now, almost three years to kind of reinvent Qiskit from its core, you know, started out as a pure Python framework, but now to have it basically be a high performance Rust core that maintains, uh, the same Python interface to it. So our users, uh, still interact with it the same ways, but they're interacting now with a much higher performance engine under the hood because it's, uh, in Rust.

Is it useful for someone like me?

Mark Fielding Is it useful for somebody like me or somebody like Jeremy or our audience to know what sort of problems you're solving or working through with this? Is that useful in any way to know that?

Blake Johnson Uh, in terms of, in terms of the conversion of Qiskit to Rust.

Mark Fielding Well, you said like, so you're, you're now modeling problems and working out problems that you can't do on a, with a classical simulation. Is it, is it useful to know what that might be like, that kind of problem?

Blake Johnson Uh, sure. Um, I mean, there's, uh, I mean, a lot of the problems that we're studying now, um, have to do with, uh, simulating physical systems, right? So, uh, uh, we're looking at a lot of condensed matter systems, uh, chemical systems, uh, where, uh, uh, like spin glass systems, uh, where, uh, we're trying to answer some, some questions about, uh, how those systems, uh, evolve in time. Um, so they're, they're kind of dynamics, um, or you might ask, be asking questions about kind of like the spectrum, like the energy spectrum, the vibrational spectrum, and so on of, of a chemical system.

Spin glasses

Mark Fielding So spin glass pertains to, spin glass pertaining to quarks. Is that where we're headed?

Blake Johnson Uh, I mean, spin glasses appear throughout all condensed matter physics, uh, where we're talking about the behavior of, of solids and materials. Um, and you can get, you can get rather, uh, unusual, uh, systems, ones that we don't, uh, know actually exist in nature, uh, but that exists mathematically and that have even sort of more bizarre properties, right? And these are, uh, the kinds of things that, um, uh, people would love to study, but often, particularly in the realm of these synthetic materials, like you can't make them in the lab because they're kind of mathematical, uh, abstractions or mathematical concepts. But now we can kind of create, uh, miniature versions of them on a quantum computer and study and study their properties.

Jeremy Gilbertson Is that, is that like a dark, like a dark matter, dark energy thing? Like we know it exists mathematically, but like no one's ever been like, Oh, there it is over here. Right. Is that what you're uh,

Blake Johnson that's an interesting analogy. Um, I mean, certainly like, uh, so one of the things which is, uh, I guess one similarity might be the fact that, you know, you can create, create a model of something of how it would work, how it should work mathematically. Uh, but, and then have it kind of be able to study, uh, it's, it's properties kind of in practice. Um, and, and, uh, I'm not so in touch with, uh, the latest models that people are considering with dark energy and dark matter, but, um,

Jeremy Gilbertson no, it's okay. I, it's, it's interesting to kind of see where these little tests are, are, are going to be pointing everybody towards you got what, 600,000 people who have actually signed up to access. Uh, your quantum systems through Qiskit. And how is the, what what's, and that's over two years, the last two years.

Blake Johnson Uh, I mean, so the, we launched the first quantum, uh, our first quantum computer on the cloud in 2016. Uh, so it's eight years ago now. Uh, so that's 600,000, uh, people that have joined in that period of time. Um, uh, but where there's actually even larger audiences that we touched that haven't, uh, that are starting to learn, but haven't yet signed up to, to run their first circuit. So I think we've had an over 9 million people, uh, view, for instance, our, our digital learning content. Um, and we also have, um, uh, so we have this, this large audience that we can, can speak to. We also know, uh, about 250 organizations that have joined our IBM quantum network, where those are sort of, uh, enterprises, startups, uh, research institutions that are making sort of a more, a larger commitment. They're putting more resources in, in sort of, um, um, really become developing expertise, uh, and how to, to take quantum computing into sort of the next it's next era.

Jeremy Gilbertson Yeah. I was thinking too, like a lot of it, a lot of, and you alluded to this a little bit, uh, before talking about like chemical processes and natural processes that are kind of inherently quantum that we don't understand because we don't have the computing mechanism, or we. Traditionally didn't have the computing mechanism to kind of process and understand that information. So let's say you had these researchers that are, you know, maybe chemists, um, or, you know, focusing on, you know, proteins or something like that. Are you, you're finding them kind of get interested, but they don't have necessarily quantum computing or computer science knowledge. Right. So how are you bridging that gap in this whole thing?

Error rates

Blake Johnson Uh, so that's, I mean, that's really important, right? Like the, um, the, you know, if the first phase of quantum computing was about, you know, uh, learning about the technology itself, like letting people answer about, you know, how does quantum mechanics work, you know, in practice or, uh, what are qubits? What is entanglement? What is superposition? These kinds of basic properties that we're, we're, we're harnessing to do a new kind of computing. Um, that kind of the next phase, uh, you know, sort of in that same phase, we were really focused on kind of the fundamental performance of, of, of the hardware systems themselves. Right. Uh, because quantum information is inherently more fragile than quantum, than classical information, right?

Blake Johnson Like if you think about error rates at a classical computer, uh, you maybe observe, uh, an error rate sort of in one out of 10 to the 15 or even 10 to the 18 operations, um, which are kind of scales, which are so low that it's only in sort of the largest, uh, classical computer systems that you even observe, you know, a few errors per day. Whereas even in a high performance quantum computing system, error rates might be sort of, uh, one in every thousand, uh, operations that you have an error, so like protecting information, uh, and sort of coping with errors and in quantum computing systems is a much more, uh, prevalent, it's, it's a much more, uh, larger part of the process of building a high performing quantum system is, is sort of dealing with these errors, um, through various technologies.

Blake Johnson And so like, even today, like a large part of our efforts are focused on driving down the physical error rates and then developing methods, error mitigation methods, and eventually error correction methods, uh, that can, can squeeze more performance out of, uh, uh, out of, out of the quantum hardware. But, uh, as we start to then, uh, sort of get to a, a space where the system can actually do computations, which are relevant to a more traditional, um, uh, computational scientist, someone that, you know, is used to studying chemicals, used to studying higher energy physics problems, is used to studying, uh, material systems using high performance, traditional high performance computing as their tool for which they do those, you know, they enable those investigations.

Blake Johnson Like you, like you say that, that, you know, we want to, to make quantum computing, uh, impactful to them using terminology that they already know. Right. So like we're, uh, actually trying to engage in sort of this, this next phase of, of trying to, you know, capture the interest and actually deliver value to computational scientists, trying to deliver tools that, um, that are kind of native to their, their domains. Uh, we really, uh, are not, we're trying to do it. We're not trying to do it alone. We're really trying to engage a community and building up, uh, these new tools. This is something, go ahead.

Mark Fielding No, no, it's open source. And the, the, the more we learn about quantum computing, the more we learn about the importance of the software and how it has to be designed for classical developers, because there simply isn't that deep knowledge in enough minds to execute, to create these quantum circuits. So that has to be done. I just, I wonder, so from what you've been speaking, if we could pan out onto the, the IBM roadmap that leads from where you are now to 2033 and then beyond the thousand qubits, um, where does Qiskit fit into the roadmap and how you mentioned earlier, you wanted to speak about the evolution of it. How is it evolving in sync with the roadmap and how, how do you foresee the next, like, is it Qkit, Q, Qiskit 2 next year you expect to come out? Is everything on track and how's it evolving?

Qiskit as a toolkit

Blake Johnson Uh, interesting question. So, um, in some ways Qiskit is kind of, uh, um, one of the foundational technologies that enables the entire roadmap, right? Kind of Qiskit, we view our hardware and software stack, uh, as a co-design problem, um, to, to power sort of the rest of the developments, uh, in terms of an ecosystem that builds on top of this kind of foundational technology. Um, and so, like I said, you know, having entered now the era of utility and quantum computing, uh, we, we, uh, sort of spurred our development in terms of the performance of Qiskit itself. But then, um, you know, uh, what Qiskit, uh, and its related technologies have to do, uh, to enable, uh, larger and larger computations and to enable computations in different problem domains, it has to evolve.

Blake Johnson So, um, one of the things that we've been doing now is in a, um, is in a, uh, part of the Qiskit stack called the Qiskit runtime, which is a, uh, a layer that is, uh, uh, really part of the quantum computer itself. It, it kind of, it, it's, uh, maybe if you, if you like, it's part of like the operating system, um, that managed, uh, one of its critical functions is to, to manage the presence of errors by, um, by technologies, by error mitigation technologies. And these are ones that execute, uh, not single circuits, but collections, uh, of related circuits, uh, and then introduces post-processing, uh, from the outcome of the, of executing, uh, that large, larger family of circuits to get, uh, uh, an answer, which is more accurate than, uh, than executing single circuits on their own.

Blake Johnson So it allows us to get more performance, more accuracy out of our computations by sort of, uh, by looking at the behavior of, of a larger families of circuits where we can kind of, uh, manipulate, uh, quantum noise, uh, to, to figure it, to, to basically emulate the performance of a, of a, of a quantum computer that had fewer errors. So that's an important technology that we're, we're doing, we're, uh, we've been developing and releasing in the last, uh, two years. Um, and, and that's a technology which will continue to evolve together with our hardware, um, uh, towards, uh, larger, uh, enabling users to kind of confidently execute circuits of a larger and larger size. So this year, our, our, our goal is to deliver circuits with up to 5,000, two cubic gates as a reliable computational tool.

Blake Johnson And leading up to, uh, 2028, uh, we will be continuing to drive combinations of improvements in the hardware and the software in this, in this kind of Qiskit runtime, uh, uh, technology that enable computations up to, uh, 15,000 gates. Um, but after that, uh, we intend to take a, a real step change, uh, in 2029 to an error corrected technology stack, um, which will, um, use a different method for managing errors.

Blake Johnson Uh, we're really using instead of encoding information in, uh, or trying to deal with errors by running large collections of circuits, uh, and, and post-processing, uh, the outcome to emulate the effect of lower errors, kind of after the fact, uh, in error correction, you instead, you can code information redundantly in multiple physical qubits, uh, and use real-time, uh, detection and correction, uh, of, of errors as they occur just to, uh, suppress the presence of errors sort of in each execution. And these technologies aren't mutually exclusive. They can work together.

Blake Johnson Um, uh, which we think may be an important part of how we sort of bridge, uh, between a realm of error mitigation and error correction, but that error correction, uh, technology is, is we anticipate being extremely powerful because it will allow us to go from circuits of like 15,000 two qubit gates, the circuits of a hundred million, uh, two qubit gates in 2029. And we anticipate even going, uh, uh, even further beyond that with our ultimate goal in 2023 to be able to execute, uh, circuits with a billion, um, two qubit gates.

Simplifying the numbers

Mark Fielding Could you just simplify that, that those numbers and those gates, what, what for a layman, what does that mean? Going jumping to that many million gates on it?

Blake Johnson Um, so the, I mean, here's where, if we had a picture, it would help us a little bit. Um, but the, uh, the different applications there, there are people that have had made various estimates of how large of a circuit you need to be able to execute in order to encode problems in different application domains, whether they're, uh, chemistry, uh, material science, uh, biology and life sciences, uh, optimization, uh, right. There's, uh, or even sort of, you know, cryptographically relevant problem scales, right. And, uh, for all of these problems, uh, you know, our answers that we come to up with now are based on sort of our current knowledge of, of, of algorithms.

Blake Johnson And, and so, uh, you know, there's multiple algorithms that are relevant for each of these, but, you know, they kind of put you in some space of sort of width and depth of the representation when, when written as a quantum circuit, which in sort of more traditional computer technology terminology is like, uh, memory and time, right? Like, uh, it's like how much space, uh, do I need, but it's quantum space rather than classical space to, to write down my problem. Um, you know, how big of a memory do I need on my, on my computer and, and time is, is, you know, equivalent to sort of how many operations and I, uh, uh, execute with that, with that amount of space.

Blake Johnson Um, and so, uh, when we're talking about circuits, you know, uh, you know, current knowledge would suggest that you need to get out to, um, this kind of 100 million or even billion, uh, scale, uh, to finally, uh, have sort of definitive, uh, advantage over the best known classical algorithms. Uh, but that landscape, um, is largely in, you know, I would say today is informed by, by theory. We have sort of mathematics. We can, we can do proofs of, you know, I compare this mathematical thing with that mathematical thing. And I can tell you that this sort of gap exists. Uh, but the early history of classical computing wasn't driven by proofs.

Driven by empiricism

Blake Johnson It was driven by practice, by empiricism, by people sort of developing algorithms, um, because they worked and they were effective in practice, not because they can prove, you know, their computational complexity compared to others.

Jeremy Gilbertson Kind of a little analogy to like Newtonian versus quantum physics, right? Like very first one is very more concrete proven. It's the way it is. And this is more trust us. It's the most proven theorem of all time. Just trust us.

Blake Johnson Um, hmm. Yes. Um, maybe, uh, the, I mean, I would say that it's a little unfair to quantum theory in the sense that people had very, uh, very concrete reasons, uh, for, uh, for quantum theory, particularly in terms of, uh, kind of thinking about the world as, uh, in kind of a probabilistic sense or, you know, the, the, the knowledge that I can possibly have as an observer. Um, and, uh, I think a, a, a really sort of statistical kind of Bayesian like approach the world kind of naturally leads to a kind of a quantum description. Um, but I, I guess maybe the example I was trying to think of more in reference to the early history of, of classical computing, right.

Blake Johnson Is like people were doing all kinds of things, uh, that were practically useful with like sorting algorithms, for instance, uh, quick sort is, uh, one of the most predominant, uh, sorting mechanisms, uh, that we know today. And it's, it has fantastic performance. Um, but, uh, people were, were using it because it was good before they could prove that it was, uh, was effective because they, we actually needed to develop a new, uh, mathematical trick, a new mathematical idea in order to understand the difference between worst case complexity and the average case complexity. Um, and without this sort of math, math, this new mathematical framework, we couldn't prove, uh, that the, the, the quick sort was a, was a good sorting algorithm.

Blake Johnson So, um, what my, my point is really that what we know about algorithms today is, is about, uh, is a little bit of a reflection in my view of, of our kind of our state of, um, mathematical proof apparatus. Um, and, uh, what I'm very excited about is, you know, um, we've, there's, there's kind of two boundaries we know about the kind of the limits of like, if you give me an ideal quantum computer, what could I do with it? Uh, and we can tell you certain boundaries there of like what it could do. And then we have the, the, the other limit of how large does a quantum computer need to be before I can stop really simulating it.

Blake Johnson And right now that the state of the world is like, we've, we've passed this first boundary, uh, where you can, you can today go sign up on an IBM quantum computer and execute computations, uh, that, uh, would take you, um, extraordinary amount of expertise to find an alternative means because you can't do it by brute force, uh, a way to simulate its outcome, its output. Uh, so we now have, we've entered this realm realm where there's this chasm between where we know how big the system needs to be to do something provably better and the, the, the, the regime where I could simulate its outcome, its output. Uh, there are some people that believe that there's nothing in this chasm. I think these people are crazy. I, I, I can't, I can't accept the fact that nature admits such a large gap to exist.

Blake Johnson And so what, what I think is going to happen is that we, uh, by enabling this sort of new era of, uh, empirical study of exploration, uh, with tools that kind of are native to different computational scientists, that, uh, the, the kind of the, our current worldview of algorithms will, will change dramatically. Um, and so that's, I think what makes it really critical then to, to embark on this journey of making tools that, that are, um, easy to consume, but from, from different, uh, uh, sort of computational domains.

Jeremy Gilbertson Makes sense. I want to go back a little bit to the, to the increase in the quantum gates, right. As we, as we project out to potentially a billion or whatever is in, forgive my ignorance, this may be a really silly question, but like, is the, the nature of quantum, like, as I think about things like, you know,

Blake Johnson Really that what we know about algorithms today is a little bit of a reflection in my view of our state of mathematical proof apparatus. And what I'm very excited about is, there's two boundaries we know about. The limits of if you give me an ideal quantum computer, what could I do with it? And we can tell you certain boundaries there of what it could do. And then we have the other limit of how large does a quantum computer need to be before I can stop really simulating it. And right now the state of the world is we've passed this first boundary. You can today go sign up on an IBM quantum computer and execute computations that would take you an extraordinary amount of expertise to find an alternative means because you can't do it by brute force, a way to simulate its outcome, its output.

Blake Johnson So we've entered this realm where there's this chasm between where we know how big the system needs to be to do something provably better and the regime where I could simulate its output. There are some people that believe that there's nothing in this chasm. I think these people are crazy. I can't accept the fact that nature admits such a large gap to exist. And so what I think is going to happen is that by enabling this new era of empirical study of exploration with tools that are native to different computational scientists, that our current worldview of algorithms will change dramatically. And so that's I think what makes it really critical then to embark on this journey of making tools that are easy to consume but from different sort of computational domains.

Jeremy Gilbertson Makes sense. I want to go back a little bit to the increase in the quantum gates as we project out to potentially a billion or whatever. And forgive my ignorance, this may be a really silly question, but the nature of quantum, as I think about things like path integral, so you model all the paths and find the best one. Super simple explanation of that, I'm sure. But is it to have enough of these paths that you can really analyze all of these things? Does that have anything to do with it? Or am I off the mark on that one?

Width, and depth of the circuit

Blake Johnson So I mean the space aspect, the sort of the width of the circuits is maybe the easier piece to kind of intuitively grasp, right? I sort of need enough quantum memory to encode my problem. And exactly how much memory you need depends on how you encode, right? There are maybe simple encodings that use up a lot more space than more complex encodings that can be denser. But sort of the size in terms of number of qubits I need really I can, at least needs to be proportional to the problem, the size of the problem I'm trying to solve. In terms of the depth of the circuit that you need, yeah, maybe thinking about it in terms of your analogy is appropriate. Like I mean, again, there's going to be some factors which are related to things which are less problem dependent and more like statements about the hardware.

Blake Johnson Like maybe I need two bits of quantum memory that are far apart physically on the machine to interact. And maybe that takes some amount of computation to get them to connect. But really it's kind of the overall scaling of the depth has to do with something about sort of how many operations I need in order to sort of encode the evolution of the quantum state. And so yes, that's going to be related to, for instance, in the domain of like simulating nature, that's going to be related to sort of, it could be directly related to, for instance, the amount of time that I want to capture in my simulation. If I'm trying to capture very short times, maybe that's a lower depth computation. And if it's a longer time, maybe it's a longer computation.

Blake Johnson And if I'm trying to capture like the properties of a material that depend on interactions between things that are far apart, then I need to sort of naturally have some sort of encoding where enough time evolves to capture the interaction that leads to the behavior that I want to study. So, yeah, I don't know. Maybe that's the best we can do for now in terms of like an intuitive picture.

Jeremy Gilbertson Yeah, no, it makes sense. Thanks for the clarity. That's great.

Error correction codes

Mark Fielding Well, I'll have another one then on the amount of qubits and the error correction. And this is, we're going to demonstrate we've read too many books here. And one of the things we've learned is the increasing amount of qubits gives you more qubits that you can use for the error correction. But we've read about different types of quantum computers. So, I'm not sure now in the IBM world if that still applies to what you're doing. So, if you have a million qubits, then you can use 950,000 of them for the error correction. Can you help me make this work in my head?

Blake Johnson So, absolutely. Yeah, there's a lot of different error correcting codes with very different properties. And particularly with respect to how efficiently they encode information. So, maybe the predominant paradigm that has existed in the quantum computing field for the last couple decades, there's been an extraordinary amount of effort studying something known as the surface code. And the surface code has some beautiful properties in terms of like the constraints it requires the system builder to satisfy. You need error rates that are somewhat achievable. You only need sort of local connectivity qubits only need to be able to have operations with neighboring qubits. Things like this that lead to some practical benefits in terms of being able to realize error correction.

Blake Johnson The problem with the surface code, though, is that it's very inefficient in terms of information encoding. In fact, you need just an extraordinary number of qubits in order to achieve computationally relevant error rates. And so, that's led to effort to find something better. And recently, our team has found a new family of error correcting codes that are they're a kind of low density parity check code, which is LDPC codes also exist in sort of classical communications theory as well. But these codes have some properties which are very attractive. And in particular, they're much more efficient in terms of their encoding density. So, our IBM development plan is based off of realizing systems with these LDPC codes.

Blake Johnson And we believe that we can do so with 14 times fewer physical qubits than you would need in an equivalent system that's based off of a surface code. And finding the order of magnitude is huge in terms of the engineering practicalities of how long it will take us to build such systems.

Mark Fielding Okay. Thanks.

Blake Johnson I was hoping we could return for a bit to some of the things we're building to make these systems accessible to other audiences. You touched on at the very beginning the new pieces involving Watson X and generative AI. And I think that's one piece of this that is particularly cool and also brand new. IBM has put an enormous effort into building up a generative AI technology stack based on open source models known as Granite. And in partnership with the Watson X team, we recently have created a version of the Granite model that has been particularly trained on Qiskit. And so, this is really cool because you now have a tool that can kind of exist within your development environment. Something called the Qiskit Code Assistant. It has a VS Code extension.

Blake Johnson Or if you like Jupyter Labs, Jupyter Notebooks, there's a plugin for that, too. But it allows you to sort of, without leaving your development environment, use this powerful LLM to help you write Qiskit code. And it's been particularly trained on a corpus of modern examples. Which is great because Qiskit has evolved a lot in its seven years of life. And so, it can be more accurate, for instance, than finding a five-year-old answer on Stack Overflow. But we hope this kind of particularly helps the beginner start to engage with the tools and start to get started faster.

Blake Johnson A second thing, which is an entirely different kind of technology altogether and a very different strategy, is that we're starting to build tools with partners that instead of operating at the language of circuits, operate at the language of the prompts in their domain. So, there's two in particular that have been released in something we call our Qiskit Functions Catalog. There's one from QControl, which is an optimization function, and one from QNSSYS, which is a chemistry function. And both of these new tools developed by partners on top of the IBM stack, you really give it a prompt description that has nothing to do with quantum computing at all. In the optimization case, for instance, from QControl, you give it a description of your optimization problem, and it gives you back a solution.

Blake Johnson And nowhere in that process do you see a quantum circuit. And even better, it works. And so, I think this is really powerful in terms of finally being able to touch new audiences with tools that are native to the way that they're used to thinking about problems.

Jeremy Gilbertson I love that. Quantum solutions, but without the complexity of the interacting with the quantum hardware. Yes.

The function library

Mark Fielding So, in this function library, it's kind of split into, I think the first one is probably a circuit function, right? And the second one is more application function, the Qantasys one, I guess. So, that would be like a researcher could come in who's doing ground state problem work with molecules and kind of just jump in and start to figure out how some of this stuff works. Is that the intention?

Blake Johnson Yeah. I mean, what QNSSYS has done is really cool. I mean, one of the kind of new frontiers in terms of enabling us to tackle larger and larger problems is finding new ways to kind of blend quantum computing and classical computing, right? I think people used to think of these sort of in contention or really competition, right? Like I just had my quantum computer, just my classical computer. But I think it's become more and more obvious to us over time that these technologies really ought to be used together. And that they're more, they really complement each other and kind of we can find ways to augment the capabilities of both when we use them in concert with each other. So, in particular with QNSSYS, what they've built is a technique which can take the problem of computing properties of like ground states of molecules.

Blake Johnson And they run some, they execute some circuits representing something about the configuration of that system. And they'll get back answers that have errors, but they can use, but because they have access to this, these quantum samples from something which is closer to the right answer, they can use a more efficient classical technique to then refine those answers and get a better final outcome. So, they're really using kind of something which is like traditional HPC-like computation, really in concert with the outcomes of a quantum computer to jointly solve a chemistry problem better than you could with just the quantum machine on its own.

Blake Johnson And in this problem of whether or not they can do better than what you can do with just a classical machine on its own is kind of we're in the thick of sort of answering that question, but I think we're very close and that's really exciting.

Jeremy Gilbertson Yeah. And I love what you said about these two working together. It's not like you have one camp that's like quantum or bust and then other camp is classical or bust and we're like, we're going to go separate. It's kind of like, let's work together. But there is, I think, theoretically, you're going to be some point where quantum is going to be like, all right, we got this figured out. It's time to move on and move to this next phase of computing. How far out is that? I mean, I'm not going to hold you to anything, but is that 10 years? Is that 15 years?

When we will have applications

Blake Johnson In terms of when we'll have applications of quantum computing where this is unequivocally the best solution? I mean, I think I'm even more optimistic. I would say probably in the next five or six years, we will get to that phase. It's speculative, but that's where I sit.

Mark Fielding Maybe a more interesting question on that is, so it's 2033. What's driving IBM now and what will you be able to do if you execute on this roadmap? Which is another question is like the roadmap. See, like I've seen a lot of roadmaps in the past and they never, ever seem to follow the steps, but the IBM one seems to be very well thought out. I'm not sure how you've managed that, but 2033, what will you be doing? What will people be doing? Where should we be thinking about the application, the real world application to change lives and save lives in 2033 with this?

Blake Johnson Yeah, I mean, so as I think we've talked about, right? Like the way that we think about the power of these machines is really in terms of the size of the circuits they can execute reliably. And I think it's pretty safe to say that whatever applications there are, we're simply opening up the aperture to more and more applications as the size of circuits we can reliably execute gets bigger. By the time we get to 2033 and our BlueJ systems executing circuits for the billion gates, I think we're touching even the points where we know that applications exist. And so particularly in material science and the other areas of application, whether they're chemistry, optimization, so on, we don't know yet, but we have a lot of time to sort of bring, to improve algorithms in all those spaces.

Blake Johnson Because I think it's, I have a hard time believing that there won't be major advances in those demands as well. Particularly because like the state of the art in terms of optimization, right? Is that the gate model quantum computers are now sort of the best quantum devices for solving optimization problems. They're even better than the machines that were sort of specific special purpose machines built for optimization. And okay, they're not yet at the point of beating the best non-classical methods, but you have to find, you have to actually work really, really hard classically to outperform them. And so if we're already there today, I don't know, I think it's kind of crazy to think that the system's going to improve by another seven, eight orders of magnitude and I won't have found advantage.

Blake Johnson So I really, I don't know how broad the scope is. I guess only time will tell, but my hope is that by the time we're talking about BlueJ in 2033, we'll really reach an era where the applicability of quantum computing is broad and easy to find. So that it's really starting to impact most domains where the kind of the fundamental difficulty of the problem has to do with the fact that we're trying to simulate nature or trying to, we're dependent upon problems that have structured data, which are the kind of the fundamental properties where there's really good evidence that quantum computers should basically be the best tools for solving those problems.

Jeremy Gilbertson Blake, this has been an amazing chat. I think we've definitely succeeded in getting tangible on some of the things and what you guys are doing with over 600,000 people locking in and testing and playing around with quantum computing. And then also the, I think you said like 9 million people investigating and trying to get up to speed through your learning platforms on quantum. Like as we move through these phases, you're bringing in the community, you're activating community. I think it's doing some really cool things. I've got two things that, I got one thought experiment that I always like to throw out there. And then we've got a final question. We always like to ask our guests to leave a question for our next guest. And it can be about any topic. It doesn't have to be quantum.

Jeremy Gilbertson It doesn't have to be computing. It can be literally anything. So let me hit you with this thought experiment first. So we talked about, a lot of systems in nature are quantum mechanical. Mark and I have been geeking out about photosynthesis. We've been geeking out about nitrogen fixing, like all of these things that nature does really well. And nature does quantum processes like at room temperature. And we're trying to do quantum processes in very controlled environments. I always like to think like, bits are easy to deal with. But like you have cubits, they're a little harder to wrangle and a little more finicky. So is there going to be a point you think where we could potentially, or could there be a point where we potentially unlock a secret of nature that could inform how we design quantum computers?

Blake Johnson It certainly seems possible. In fact, I would even say probable. One of the fundamental tensions you find among practitioners or system builders, I think you're probably well aware that people choose vastly different quantum systems to build quantum computers. And they're all successful in different ways. But the fundamental components we're using start out with very different ingredients. From people that are using natural quantum systems, things that look like a single atom or a single ion, to sort of on one extreme to those of us that use kind of synthetic quantum systems, where we are building in IBM quantum computers that they're kind of their heart, our cubits are based off of superconducting quantum circuits that have an energy spectrum, which is like an atomic energy spectrum.

Blake Johnson So we have kind of, we talk about artificial atoms as being kind of the constituent components. And, you know, these two different realms have different trade-offs, right? The natural ions, nature makes quantum systems with, when you pick the right one, that have extraordinary coherence properties, meaning that the quantum formation is very long lived and very well protected on those these quantum systems. But part of the reason that sort of the natural cubits are so good is that their interactions with the outside external world are very weak, right? So that means that the kind of the operation rates on these natural cubits are very slow.

Blake Johnson Whereas with the engineered systems like superconducting circuits, if you don't take great care in their construction, their interaction rates with the external world are very strong, and that leads to very poor coherence properties, so that the rate of, so they sort of decohere very, very quickly. And of course, like the ideal quantum system would be the one that sort of blends the properties of the two, right? Where you have kind of the lifetimes, the coherence times of the natural systems with the operation rates of the synthetic ones. And so maybe there's something still to be learned from nature about how to sort of get access to just those sort of neural channels that we need to communicate with quantum systems, but that still preserve the remarkable coherence properties that you can find in nature.

Jeremy Gilbertson That's why this stuff is so exciting. Like the potential of this has my mind spinning in really fun ways. Blake, we appreciate you coming on and talking about what you guys are doing. I love the accessibility component that you guys are building in with your SDK, with getting people educated, with pulling people into this quantum network that shares learnings, and third- party people are building applications for chemists to figure out ground state molecule stuff. I think it's awesome and amazing. What question would you have to leave for our guests? And our guests are all over the place. I mean, quantum, they're in AI, some of them are creators, some of them are writers, some of our ethicists. I think it would be like just our next guest will be working in crypto and blockchain. So if there's a link there.

Blake Johnson Crypto and blockchain. I guess I would say, I guess I'm curious to see what an expert in crypto and blockchain, when they believe that quantum computers will impact their field. And if they had a BlueJ-scale quantum computer, what would they do with it?

Jeremy Gilbertson Keys to the BlueJ, I love it.

Should they be worried?

Mark Fielding Should they be worried? Should our next guest, we spoke to the IBM quantum team the other day, and you should be worried working in crypto or is there anything to worry about?

Blake Johnson I mean, briefly, we can touch on that, right? Like our team is, so even a BlueJ-scale quantum computer is not sufficiently large to break current day public key crypto systems. You need to be even larger. It gets close. But we're not going into this blind and our team has also been at IBM, has also been at the forefront of developing so-called post-quantum cryptography methods. In fact, IBM participated in the development with partners of both of the protocols that have been selected by NIST as the new standards for post-quantum cryptography, and are the new recommendations for enterprises to use to ensure that our information stays secure in a world where quantum computers of cryptographically relevant sizes exist. So, yeah, you don't have to worry even at the extent of IBM's roadmap.

Blake Johnson But I mean, I would also say like particularly for enterprises that deal with sensitive information, banks, healthcare systems and so on, learning about the new cryptography methods, now's the time, right? Because transitioning those systems can be slow and painful, but we actually have tooling that can already help you make that transition. And so it's not too early to sort of get a handle of the cryptography used in your enterprise and figure out your transition plan.

Mark Fielding Okay, thank you. I've got one observation and one question. So my observation is that there's not enough highly skilled computer developers in the world. Every time we have a guest on, if they're in the blockchain, it's like there's a shortage of people who can do this. If we work in robotics, there's a shortage of people who can do this. Now it's quantum, there's a shortage of people who can do this. And we're making these systems, IBM, you're doing, you're making these systems to try to make it easier so that you don't need to have that. And maybe with AI, you don't need to have that level, so you can combine the two. I think that's very interesting and I'm going to keep thinking about that. And my one question is, Osprey, Flamingo, Kookaburra, Heron, Starling, who's the ornithologist? What's with the birds?

Blake Johnson We have several colleagues that are really into birds. But I guess the final say, particularly in terms of the latest round of birds is Jacob, who particularly insisted in sort of the latest update of our roadmap had some Australian birds because he's from Australia. So Kookaburra in particular.

Jeremy Gilbertson I really enjoyed the Starling in there because Starling points back to a program called Boids back in the 1980s, I think, which was a software development program that tried to map how Starlings were random because they fly in these. So there's basically one Starling that makes a move to the left, and then all the Starlings around it make that additional move to the left. And that's why the whole pack moves and swirls in all these different ways. So I grabbed on the Starling. I was like, hey, that's cool, right?

Mark Fielding Sounds like a complex quantum system there, Jeremy.

Jeremy Gilbertson There it is. There it is. Well, this has been a great chat, Blake. Thanks for entertaining our organized and our ethereal questions. I know we threw a couple at you. We appreciate you running with all that. And Mark put a great show right up together for this with some links of what you guys are doing. We really appreciate you joining us. And Mark, closing thoughts before we get out of here?

Mark Fielding Join our book club, like and subscribe to Thinking on Paper so that we can continue having conversations like this and try and the dots of emerging technologies. And yeah, thank you for... I think we're quantum addicts, Jeremy. I think maybe we need to do something about it. We need to go to Germany. I can go to Germany and look at the new center. Is it open to the public?

Blake Johnson I don't think so. I mean, these systems, right? Like we don't want to see anyone coming up. We don't want the qubits to decohere when you show up, Mark. That's the reason why you can't go. I think I do understand that our partners at Cleveland Clinic actually have their on-premises system in their cafeteria. So maybe if you have any friends that are researchers at the Cleveland Clinic, you can go walk by.

Jeremy Gilbertson Amazing. Well, thanks again for joining. Those of us still listening, thinkingonpaper.xyz. Be curious, stay disruptive.