The New Quantum Era

Sebastian Hassinger 0:30
Welcome back to our podcast, we're very excited about our guest. Today, we think you'll enjoy this episode, we're joined by Dr. John Preskill. Today, he's a professor of physics at the California Institute of Technology, and quite famous in the quantum computing field for having coined the term NISQ, noisy intermediate scale quantum computers, he wrote a paper in 2018 that really characterize the types of devices, the more sophisticated devices are being produced, but also can't call that a lot of the challenges that those technologies are still trying to overcome. And his work has continued to be really at that intersection of what the art of the possible is, from both a technology and a scientific perspective with the devices that are made, and are made available today should be a fun ride, because he's also one of the ringleaders of the so called do it from qubit intellectual movement, I guess, a steel studying the intersection of quantum gravity theory and quantum information sciences. So it's gonna be a good time. Yeah, what I find really interesting about John is that, in a way, his interest in quantum computing is motivated by the ways in which it can help progress, our understanding of quantum mechanics, specifically, quantum gravity, quantum matter. And as you said, sort of that broader set of topics that are known as it from qubit, we will dig into the information paradox in black holes, for example, and what error correction might be able to tell us about that phenomenon. So it'll be great. So let's jump in. Here we go.

Hello, and thanks for joining us again, we've got a really special guest with us today, we're extremely excited to be speaking with John Preskill. He's the Richard P. Feynman professor of theoretical physics at the California Institute of Technology, or Caltech, where he's also the director of the and I think founder of the Institute of quantum information and matter. He's played an instrumental role in the foundation of the field, the development of quantum information science, famously coining the term NISQ, or noisy intermediate scale quantum computers to describe the devices that we're building today. In a paper that was published in 2018. It's still, I think, critical reading for understanding the field as it is today. So we're very excited to have with us, John Preskill. Welcome, John. Oh, I'm excited to be here with you guys. Welcome, John. Thank you. And we always like to start off just by giving our guests a chance to sort of tell their story a little bit because the the path to quantum Information Science is so varied, you, I think, began as a particle physicist. So you know, how did you sort of make your way into this, this interest in quantum information?

John Preskill 3:49
Well, actually, like a lot of kids of my generation, I got excited about science because of the space program. I remember the day that Yuri Gagarin orbited the Earth. It was an electrifying moment, and then followed very avidly all the missions of Mercury and Gemini and Apollo and so on. And so then I wouldn't learn about rockets and things like that, but amazing days. Yeah, yeah. I got interested in math. And, you know, in high school, I decided the greatest achievement of the human intellect was Godel’s theorem, the idea that we could, that there are things that are true, but which we can never prove, I thought was fascinating. So I thought, okay, I should be a mathematician to set theory and logic. And that's what I had in mind when I went to college, after a year at Princeton, which I had gone to, because somehow I thought that was the right place to do math. I think I kind of wised up that maybe I wasn't going to. I wasn't cut out to be a mathematician. But meanwhile, physics is even better, right? We use math and we can understand nature of by using mathematical ideas and methods.

John Preskill 5:13
Now, this was a time when particle physics was really taking off, I was in grad in the early 70s. And so the standard model was being established and dynamics was discovered and charm was discovered. And there was something very deeply appealing about studying nature at the most fundamental level. And so I, that's what I wanted to do, I went to Harvard for graduate school. Now, the sad thing is, for my generation, we came along just a little bit too late to take part in discovering the standard model and establishing those truths. But it was okay, we were going to do the beyond the standard model physics, right. So I got interested in cosmology because we had all these speculations about physics at really short distances, that we couldn't explore and accelerators. But maybe by studying the early universe, we could find out about physics at very high energies that we can't directly access on Earth. And that was one thing I was interested in. And the other was, you know, ideas about beyond the standard model that we could explore and accelerators. We were going to build the Great Machine, the superconducting supercollider, and they started building it. And, of course, this was later in the 1990s, it got, it got cancelled. But by that time, I had started to get interested in quantum information. Because I was interested in black holes, you know, I had to wait for the SSC to come along. So what to do in the meantime, while there were very fundamental questions, I started to appreciate in the late 80s, about black holes and how they process information, which was a fairly old thing by then, because Stephen Hawking had pointed this out, mid 70s. But I really began to appreciate it in the late 80s. So, as I've often done, I taught a course about Hawking radiation, and how information gets processed by black holes and stuff like that. And that's when I became aware of quantum information as a field of study, several reasons, partly because I thought, well, I'm trying to understand a very fundamental question about information processing. So I should know what people know about that. But also, Charlie Bennett came to Caltech in the early 90s, on sabbatical, for a year. And, you know, he kind of made me aware that people were studying this, like he told me about David Deutsch, his paper. Incidentally, Feynman, of course, had been very interested in quantum. And we overlap Caltech for four and a half years between when I arrived, and when he died. And we used to talk a lot. Never about quantum computing, because I think, right, I had seize that opportunity.

Sebastian Hassinger 8:09
Charlie is kind of the the Johnny Appleseed of quantum computing.

John Preskill 8:14
You could say that, and another thing that Joe was an influence was Seth Lloyd was at Caltech at the time as postdoc, and he was interested in quantum computing. So I, you know, I read Deutsch paper, like the one in from 1985. And now what Deutsch did something really important. He formalized the subject, you know, he defined what a quantum computer is in a way that a computer scientists could understand and potentially leap into answering questions about it. But I was very unimpressed by his paper, I have to confess, I missed the point, I thought that he really he was just talking in very fancy language, about probabilistic computing that, you know, flipping coins to decide the path of a computation. Of course, quantum physics is good for providing a source of randomness. And I didn't think there was anything more to it than that. Just completely wrong. And But meanwhile, I was learning about cool stuff like quantum teleportation, you know, quantum key distribution, and I realized that, you know, the information theorists had figured out interesting things about how much information you can gain if you make a measurement.

Sebastian Hassinger 9:33
When I guess Deutsch also laid out sort of the the case for information being a physical process, right being being mean, there being a physics of information, or the information is understood through physics.

John Preskill 9:48
I think computer scientists, at least many of them, the best ones, had an appreciation about already but Deutsch was making a really important point that If we build quantum machines, he was arguing that they might be able to efficiently solve problems that we can't solve with conventional machines. It was a fascinating speculation, which at the time, I didn't think was very well backed up by arguments. But of course, what was a pivotal moment was the discovery of Shor's algorithm, or me. So short paper, as a preprint was first started to be distributed in April 1994. But you had to be in the in crowd to get it on the archive. And it was being sent around by email, and I was not. But our director was, and he happened to visit Caltech around that time, to give it actually to give a talk about quantum key distribution, but he I heard from him about Shor's algorithm, and I was able to get the paper from him. And I was immediately fascinated because, you know, I, I never knew very much about the theory of computation and computer science, apart from my childhood fascination with girdles theorem, and but the idea that you can solve problems efficiently that you'd never be able to solve, because it's a quantum world, and not a world governed by classical physics. I thought that was one of the coolest ideas I'd ever encountered. So I wanted to learn about that.

Sebastian Hassinger 11:30
In your in your course notes to AI. You describe Deutsch problem, or the solution relying on unknown nonlocality, I think right is it was sort of that the insight that Shor was using that mechanism in his algorithm that sort of made you go back and reassess Deutsch paper or doshas work?

John Preskill 11:54
Well, sure, I mean, after Shor, clearly,

Kevin Rowney 12:01
cats out of the bag.

John Preskill 12:03
Shor built on a paper by Dan Simon, which came up, you know, just months earlier, described a problem of no apparent practical interest, for which you could argue there was an exponential speed up. And that actually was what inspired Peter to come up with his algorithm. Yeah, so I taught that course, because that's the best way to learn something, right? I think at the time, which was 1997. There, maybe Caltech was the only place in the world where we were offering, oh, a one year class on quantum information and computation. And that solidified my, my understanding of the subject. Right away, I got interested in the question of whether you can really build these things. I'm not an engineer, but as a fundamental question, pretty fascinating. There were naysayers, you know, and good physicists, like Bill Monroe was one. And Rolf Landauer argued very vigorously that you build these things, they understood decoherence, you know, learning about so of course, the key thing about decoherence is, there's something different about quantum information than ordinary information, you can't look at it without disturbing it. It's not true of classical bits. And it's true of qubits. And the environment is always looking. And decoherence is a very fast process. That's why in practice, it's very hard to make a cat which is simultaneously dead and alive. Because it immediately interacts with the environment and becomes either completely dead or completely alive. And because decoherence is so fast, both Landauer and Unruh argued, you know, we'd never be able to do a computation of any interest fast enough before decoherence killed it.

Sebastian Hassinger 13:54
I can't remember whose slides it was, but I copied it down because I thought it was so hilarious. It was a Landauer quote that may have been Charlie's slides, or might have been yours actually. But it was the quote, this scheme like all other schemes for quantum computation relies on speculative technology that does not in its current form, take into account all possible sources of noise, unreliable and manufacturing error, and probably will not work. That's Landauer’s last word,

John Preskill 14:22
I talked to Landauer at the time. And he had a very forceful way of expressing himself. And he wasn't just worried about decoherence. To be fair, he was worried about the fact and others were too, that quantum information forms a continuum. It's not discrete. It's not, you know, definitely on or definitely off. So how are you going to control things well enough, there isn't an accumulation of small errors that rotate quantum states a little bit Landauer was at least as worried about that, as he was about decoherence.

Sebastian Hassinger 14:58
So the precision of control rather than just the noise of the inherent system or the isolation of the system.

Kevin Rowney 15:05
And this is interesting, it's a fair critique, because a lot of people showed concern around at the dawn of the analog computing era, that same, that same critique, right,

John Preskill 15:15
Of course, Landauer was very well aware. Discussions about it.

Kevin Rowney 15:18
Same track record. Yes.

John Preskill 15:21
Yeah. You know, we had learned the lesson that you can't control analog information well enough, so you need to digitize it, and then land are used to like to open the door and then slam it shut, you know, to make it make it clear to whoever was listening, that there's a big difference between open and shut. quantum door. There's not around Yeah. And, yeah, so. So I've got interested in error correction, which I knew very little. And while you know, Peter Shor had this amazing run in the mid 90s. He wrote the paper in 94, describing Shor's algorithm, he described quantum error correcting codes in 1995. Andy Steen did that independently, late 1995. But then there was still a big challenge, which was, how are you going to do quantum error correction, when the hardware is imperfect? You know, you can measure things. So with perfect accuracy, how are you going to keep the quantum computation on track? There was the question of fault tolerance, where everything's noisy, how can you do reliable computation and Shor wrote the first clear paper about how to do that in 1996. So, you know, there, I think there are few better examples in the history of science of somebody having a streak like that, making such fundamental discoveries and rapid succession

Kevin Rowney 16:54
so profound and it drew so many people into the field as well. I mean, it was just such an inspiration, right?

John Preskill 16:59
Oh, and lots of skepticism still in the 1990s. About whether this could ever work. And for good reason. Here we are. It's storming on it. Still struggling that way, though? We can we certainly feel like we're getting somewhere. So, big challenge, but a very fundamental discovery. I mean, this isn't just engineering. Actually. It was interesting. My colleagues who were in high energy physics didn't quite get -- at least this is true of most of them -- why would be interested in quantum error correction? Isn't that just like an engineering thing? But no, I think, well, it is that but but it's a fairly fundamental discovery, that we can control quantum systems, very complex, very highly entangled quantum systems and really get them to do what we want them to do with high accuracy. That's an amazing theoretical discovery. Absolutely, yeah. Now we're in the process of turning it into reality with actual hardware,

Kevin Rowney 18:01
and just the sheer force of that Shor discovery of pushing the Church-Turing thesis back on its heels. Right. I mean, that's, that's profound. Right.

John Preskill 18:09
Right. Because as your remark correctly indicates, the question of whether such machines are really allowed by nature? Yes, was an open question until we had quantum error correction. So, you know, the idea which Shor seem to have pointed towards, that we can solve problems with quantum machines that we couldn't solve with classical machines efficiently. That wasn't going to work unless we could do error correction. And while we figured it out, it was great.

Sebastian Hassinger 18:42
Well, we figured out theoretically, error correction and fault tolerance, in reality are proving to be elusive. Progress was still a long way to progress every year. It's true.

John Preskill 18:53
It's a hard problem. Well, you know, I was very interested in the foundational questions in those days where there were a lot of them to think about, you know, what kind of noise models were encompassed by our ideas about quantum error correction, what arguments can we give that those noise models are physically realistic? How can we exploit the structure of the noise to do a better job with quantum error correction? Those are all questions we were thinking about, you know, 20 years ago, and they become increasingly relevant as the hardware advances..

Kevin Rowney 19:33
And is it fair to assume that that whole avenue of interest for you eventually led to the so called it from qubit movement? Is that a -- is that a fair speculation on my part?

John Preskill 19:46
Well, of course that involved the work of many people, but as I said, I had gotten interested in quantum information, initially in the context of black hole physics and thought about that had a lot and then in the 1990s. And I recognize, for example, the importance of the no cloning principle in that setting. If, you know the big question, which I guess I haven't explicitly mentioned, was if information falls into a black hole, and then the black hole evaporates completely and disappears, what happens to the information. And Stephen Hawking in those days was arguing it's gone forever information is destroyed. And that was very upsetting for the culture. I came from the high energy physicists, we thought it was a fundamental principle of quantum physics that although information can get all scrambled up and become exceedingly hard to read, it never really gets completely destroyed. And trying to understand how that works, has steered a lot of the thinking about how gravitation and and quantum physics fit together. Since really Hawking 50 years and 1974 he did, he discovered Hawking radiation that black holes evaporate. And were I honestly think we're making a lot of progress on those questions, but it's still not, it's still not completely resolved. So you asked about it from qubit. It from qubit was sort of an update of one of John Wheeler’s, aphorisms. He said it from bit, he -- actually pretty interesting insight back 1980s and early 90s, he was saying that progress in fundamental physics is going to hinge on bringing in ideas from information theory. And he tried to capture that by saying it from bit. He was good at aphorisms, you know, the name black holes, said they have no hair and things like that, incidentally, I, we there was one of my teachers when I was a Princeton undergrad. And so he taught a course that I took when my second year as an undergraduate, which covered all of physics in a year, very idiosyncratic. But you know, he wanted he wanted to give us the whole the whole story, from classical mechanics to quantum physics, statistical physics, thermodynamics, and then throw in things like fluid mechanics. And so after that year, we know everything.

Kevin Rowney 22:34
It's quite a breath, walk in the park. Really? Yeah, I

John Preskill 22:38
think teachers, you know, you'd always come into class, immaculately dressed in a suit and tie. And he was a master of the colored chalk of the illustration, he could draw in real time. But I'll tell you one story about Wheeler, which kind of gives you an idea of his teaching style and his sort of visionary nature, the kinds of questions he would be attracted to. So one day, you know, this class is pretty far along, we've learned all kinds of stuff. And he comes in and he says, Alright, take out a piece of paper. And I want you to write down all the equations of physics, don't leave any of them out. And let me know when you're done. A lot of equations, you know, you could write down Newton's law, the grunge equation, shirtings equation and laws of thermodynamics very easily, and dutifully wrote them all down. And then he collected the papers. And he brought them to the front of the room, and he put them on a table. And then he stood back, was quiet for a moment and he gestured towards the stack of papers. And he said, fly nothing happens. Fly, nothing happens. Then he looks puzzled, you know, and then he turns toward it. I have it on very good authority that these are all the equations of physics. But they won't fly. The Universe flies. Something must be missing. So exactly.

Kevin Rowney 24:10
dramatic flair. That's right, in a suit and a tie to that's great. Something to ponder

John Preskill 24:15
I don't really know. That was typical. We try to make you think about things in a non standard way.

Kevin Rowney 24:27
Some fun, fun, colorful personalities, no doubt. Yeah.

John Preskill 24:30
So actually, it's pretty cool the way ideas which were being developed without fundamental physics necessarily in mind, like quantum error correction have turned out to be very relevant in other areas of physics, not just gravitational physics, but condensed matter too, because you know, different types of quantum error correcting codes we now realize can be thought of as different phases of matter with different entanglement and stuff like that. But the idea that quantum error correction had something to do with quantum gravity. It was first hinted at, in 1997, Juan Maldacena made this amazing discovery that at least in the in a sort of toy model, quantum gravity on a negatively curved spacetime, there is a exact correspondence between quantum gravity in a three dimensional space and a conventional quantum theory in two dimensions, that somehow that extra dimension is emergent. From the physics of the two dimensional theory, that's a realization of what we call the holographic principle that came out of thinking about black holes, you know, it goes back to Einstein and Hawking who said, hey, the entropy of a black hole goes like the area of its event horizon. And that seemed very puzzling, because the amount of information you can stuff into a region of space, it should go like the volume, right? If, in fact, if you tried to put too much information in a black hole forms, and so there's a limitation on how much information you can record, which really goes like the area of the boundary of the region. And so that's true, not just black holes, but but more fundamentally. And so I'm all the same as model was very nice, fairly explicit realization of this principle, that the world, at least in this model is a hologram. It's really two dimensional, fundamentally, but it behaves for all practical purposes, as though it's three dimensional, you can live there and experience three dimensions, just like we're doing now. And so how are you supposed to think about that there's some kind of dictionary that relates that three dimensional space to its two dimensional hologram. And what we eventually came to realize is that that's a kind of quantum error correcting code to take this hologram, and you can cut pieces out of it, you know, and remove them. But there's a redundant encoding of the information. So deep inside the three dimensional space, it's very robust. And it's really a kind of quantum error correcting code. To understand work, you need to know something about quantum gravity and also about quantum error correction. There weren't many people who knew both.

Sebastian Hassinger 27:30
Is there any parallel that you can draw between Gottesman Kitaev Preskill error codes or error correcting codes and the and the holographic projection of three dimensions that you were just describing?

John Preskill 27:46
Well, I don't see a direct, parallel, but they're both surprising in a way. And so look, just to give the background, you're asking about a coding scheme, which we suggested over 20 years ago, and I guess, at the time to experimentalist seem kind of futuristic, but is now taken more seriously, people are doing it with devices. The idea was, instead of using qubits encoded in a two level system, well, that's not quite the right way of saying it, you can take advantage of a continuous variable system and store information in it, which is digitized, which is what makes it robust. But then you can take advantage of our ability to manipulate a continuous variable system like our harmonic oscillator, it could be for example, a microwave resonator and superconducting circuit, which behaves like such an oscillator, it could be the motional state of an ion in a trap, which is another type of oscillator and the code that we proposed, I guess it was in 2000 is a way of encoding cubed or anyways, some finite dimensional system, which is very robust in that continuous variable system. And you know, it has some potential advantages because we can make really good cavities, which have very good coherence. And as an approach to error correction, it's, well, it has advantages and disadvantages compared to others. So the thing that's surprising, or one thing that's surprising, is everybody knows about the uncertainty principle. You can't unknow with precision, both position and the moment of a particle not with arbitrarily good precision. But what this code takes advantage of is that if you consider some quantum state of an oscillator, and then you move it a little bit, you can move it by either increasing its moment or increasing its position. And you can simultaneously measure both the shift in position and the shift in momentum, with arbitrary precision as long as promised that those shifts are small. What determines small Well, it's Planck's constant, what else that's all physics comes into the, into the uncertainty principle. And, and this is really how the robustness of of the code works. Because you can prepare one of these states, which kind of looks like a comb grid, where, you know, there, it's a superposition of different possible positions, say, of a particle. And then if that gets shifted a little bit, you can make a measurement of how much it was shifted. And that's true when you shifted in position or when you shifted momentum. So you can make a measurement that diagnosis that error. As long as the shift is small enough, you can unambiguously correct

Sebastian Hassinger 31:00
it without collapsing the state. Yeah, that's

John Preskill 31:03
right. And so it's protected, it's protected, in particular against the form of error, that is our biggest enemy in say, a microwave resonator, which is the loss of a photon. Okay, so that loss of a photon you can think of as kind of being a little shift in space. And it's something these codes can correct. And that's one of the reasons that they're advantageous in that setting, they seem to be very well suited for dealing with the type of error which is, is the most common and in that type of hardware.

Kevin Rowney 31:41
It's such a cool algorithm. And I'd love to hear a little bit more about the the background of the thinking and the intellectual history of that time. Because we I think, for our audience, they're kind of like really vibing a lot on the idea that, you know, quantum Information Science somehow informs, you know, the way the surface of a black hole processes information that's just so fascinating. But is it? I mean, are we correct, and speculating that some of that research then came back the other way and informed the thinking around quantum error correction? And there's almost a two way flow of influence on those two very interesting abstract ideas.

John Preskill 32:19
It's a good question. And in principle, it could have happened. I will say this, that thinking about quantum error correction in the context of gravitation led us to construct new types of codes, which weren't previously known. Now, whether those codes are actually useful for other purposes, like in quantum computing, that hadn't really been established. But, you know, I think having new ideas for how to protect quantum information is something that we'd like to have more of, where eventually, we may find applications, you know, as far as quantum error correction, and in particular, this idea of encoding the information in an ordinary harmonic oscillator, we might not have proposed that, if not for interactions with experimentalist at the time, in particular. When I found out about Shor's algorithm, there was somebody else at Caltech, who was very interested in quantum information in the mid 90s. That was Jeff Kimball, who is an experimentalist. And he's the master of trapping atoms in optical cavities. And had, you know, shown that you can entangle atoms by manipulating the electromagnetic field in the cavity, and you can entangle an atom with a photon in the cavity unfolds like that. And so in the mid 90s, Jeff was saying, you know, I should be able up to some, you know, cut off some limitation, be able to make any quantum state of the electromagnetic field that you want. So what do you want me to make? And so that provided some of the inspiration for thinking about is there a promising way of encoding information in the electromagnetic field in a cavity? Jeff was considering optical light in a cavity. So what happened later on was Rob Schoelkopf in particular, and his proposed the idea they call circuit quantum electrodynamics. Where the cavity was was a microwave resonator. So there, they were manipulating the quantum state of microwave light. So I can remember about 20 years ago, hearing Rob talk about this idea of circuit QED it was at a Gordon conference, where he still in published on this idea, and Jeff and I were both in the audience. And we talked about it afterward because we were very impressed by what Rob was proposing. And Jeff said to me, you know, if I were a graduate student, now, I'd work for Rob Schoelkopf. Because what he saw was the things he had been working very hard on doing, you know, for 10 or more years, you could do better with microwaves with than with optical photons,

Sebastian Hassinger 35:33
Right. It's interesting how, in a sense, like, we've come back to AMO with like, the neutral atom platforms that are proliferating now, right? I mean, there's now these optical tweezers with atoms in an array that can be manipulated for computation. It seems like the the engineering has caught up in a sense with with the superconducting regime,

John Preskill 35:57
Right. Well, I think the case of neutral atoms and tweezers is quite instructive for several reasons. One is that maybe five years ago, nobody was talking about it. You know, and so I think that illustrates the potential for new technologies to emerge, which can really take off and, and lead further developments in the field. And I think we can already do pretty interesting things with those systems, there are systems of hundreds of atoms in tweezers, where you have some, which in principle, you could operate as a circuit based universal quantum computer.

Kevin Rowney 36:39
And these are, these are Rydberg arrays you're referring to?

John Preskill 36:43
That's right. So that means they're highly excited atoms, you know, it's pretty cool, you can, a laser beam can hold on to a single atom, and then you can highly excite that atom. And when two such Rydberg atoms are in proximity to one another, they actually interact quite strongly because of their dipole moments. And you can exploit those interactions, potentially to do circuit based quantum computation. But also to do interesting analog simulations of quantum. Even if you don't have universal control, you can create new types of phases of matter. And we've already seen some, some promising illustrations of how you that can lead to to new discoveries, you know, the condensed matter physicists have been hampered, because they have all kinds of exotic ideas about new phases of matter. But it's really hard to make these things with electrons in material. And the AMO physicists come along, and they have a different bag of tricks for realizing new quantum phases of matter with greater flexibility than we could before where you had to synthesize a material and

Sebastian Hassinger 38:00
Right and Misha Lukins group, has that result where they made the quantum spin fluid, I believe on their neutral MRA and and ColdQuanta has a device that can produce Bose Einstein condensates on command that you can play around with, which is pretty impressive.

John Preskill 38:19
Well, that Lukin experiment is I think it's kind of a milestone, because the theorists have been talking about quantum spin liquids, you know, for over 20 years. And there's been controversy about whether there are any real materials,

Kevin Rowney 38:37
like quantum spin liquid, just theory, but now we know Yeah,

John Preskill 38:40
well, yeah, I mean, so you could say, Okay, well, you know, no big surprise, if you could put together atoms that interact in a certain way, and the theorists can predict that you'll get a quantum spin liquid. But things are more subtle than that for a number of reasons. You have to prepare the state. And they do it by a kind of adiabatic method where, you know, slowly change the way the atoms interact with one another, to get into the interesting quantum phase. And it didn't turn out exactly the way the simulations predicted it would. And the reason was, there were different timescales that the theorists hadn't yet taken into account that you have to worry about when you're doing this adiabatic change. And so you might not wind up creating the actual ground state of the system, but some excited state and then they had to go back and the theorists had to figure out what what had happened. That's an example of the, you know, back and forth between theory and experiment in quantum matter, which is now possible,

Sebastian Hassinger 39:38
you know, you wrote the paper where you, you create the acronym NISQ, noisy intermediate scale quantum machines, quantum devices in 2018. And you're often sort of called on to sort of update your view on where we are, how far you know, advanced are how much progress are we making towards, you know, universal fault tolerant quantum devices, but I also, you know, you gave a version of that at the Solvay Conference, which also included your interest in quantum matter and quantum gravity. And I just get the impression that building these devices brings the theorists and experimentalists together for a practical, you know, a project that has all of these added benefits. That may not even be apparent when when you start to try to build your first neutral atom array that that contributes back into the basic science of understanding quantum, as is the quantum matter in quantum gravity and other topics of quantum mechanics? Is that do you see those as two distinct efforts? Or do you think that's all sort of one one in the same mission to try to drive the the field forward?

John Preskill 40:51
Well, in a sense, it's one on the same mission. But you know, I'm a physicist, I don't need to apologize for that. Here, I think it's really great to build quantum computers that can solve practical problems that will benefit the world or so we hope, eventually. But I think we can be excited right now about the potential for the quantum technology for scientific discovery. And that's where I see things happening, you know, on, say, a five or 10 year timescale.

Kevin Rowney 41:27
This is great. This is exactly one of the topics, we wanted to talk to you about this. I mean, if you could perhaps speculate a bit, what area of experimental physics do you think would be most powerfully explored by NISQ machines now available? Or soon available? Do you feel like there's insight on a new area of breakthrough which is on the verge of of emerging?

John Preskill 41:49
Well, when all always has to be a little cautious about predicting what's going to happen? But where I see a lot of potential? Is that one thing we're so of course, a big part of the question, a big part of the issue is always Can you do something with a quantum machine, you can't do with classical machine. So so one thing we can say in that respect, is that the methods that we know of, for simulating on a conventional computer how a many qubit quantum system behaves, they start to fail, when the systems become very highly entangled. You can ask about how and you know, how entangled are the low energy states of a system like a complex molecule or material. And that has a lot to do with how hard those are to simulate on a conventional computer. But there's some uncertainty about how hard those things really are classically, because we've got pretty good classical methods, we know they have limitations, they keep getting better. And for the things that you care about, if you're a chemist, or a material science scientists are they are those systems really so highly entangled, that the classical methods will fail, not so clear, but when it comes to simulating the dynamics of highly excited matter, classical methods are not good at all, because those systems become very highly entangled. And the classical methods of break down very rapidly. So I think one potential area where we can expect discoveries is in very highly chaotic quantum systems, which become very strongly entangled quickly. What types of new phenomena? Can we find what types of knew far from equilibri phases of matter, you know, most of the study of phases of matter, up until recently, has focused on equilibri phases. Because partly because that's what we often encounter in the lab. But with quantum computers and quantum simulators, we can start to investigate new types of math or new phases, which are far from equilibrium.

Kevin Rowney 44:11
And it's interesting, I need to go out of my way here, we have had previous guests answer the same question. And if I'm not mistaken, they they have been in agreement with you, it feels like there's a, at least a rough consensus that this particular areas is ripe for breakthrough,

John Preskill 44:26
right. And I'm looking at it from the perspective of what's really hard to do with conventional computer, because that's where you're carrying quantum computing has the potential to show us something new.

Kevin Rowney 44:36
So So these these odd, odd new phases of matter. I mean, I'm just an amateur I study this in my my part time, but I mean, are you referring to things like I don't know, spin glasses and time crystals or is there some other domain of of the world that that this is pertinent to?

John Preskill 44:55
Yeah, things like that. Time crystals are interesting. One thing about time crystals is there. So we call them Floquet phases, what that means is you drive them in a periodic way. And that's something you can do with a circuit based quantum computer pretty well, maybe more easily with a circuit based quantum computer than with an analog simulator, which doesn't have universal control. And so I think there are such Floquet phases of matter, which are periodically driven, which we had the opportunity to discover, experimentally, and then theorists can go to work, understanding them, I think we also get some guidance from quantum error correction, that if you consider a quantum error correcting code, but then you introduce noise, or if you, you know, change the code a little bit by changing the Hamiltonian for which the ground state or something like that the way that system behave behaves, When we drive it away from equilibri has a lot to do or is related to, can we correct errors or not? And so I think we can leverage things that we've learned about quantum error correction, and when it works, and when it doesn't work, to discover or look for new types of faces matter. And that's also going to be interesting. sounds so

Kevin Rowney 46:19
cool. Wow. Just amazing.

Sebastian Hassinger 46:22
I wanted to make sure we made time for touching on the work that you've been doing that your your student Robert Huang has been doing, because I think it's it's quite interesting. I've heard you talk about a little bit in the past. And if I'm not mistaken, it he started by sort of looking at classical shadows. And and, and how they may relate to machine learning applications. Is that right? Or is there there's something? It's a fairly broad area that that he's working with, it seems.

John Preskill 46:54
Right. So Robert came to Caltech as a beginning graduate student in the fall of 2018. And he already had a background in machine learning, not quantum physics. But he had done research as an undergrad, he was an undergrad in Taiwan, in machine learning and knew a lot about it, I knew very little about it. In 2018. As an aside, I used to be a freshman advisor at Caltech. I haven't done that the last few years. But when I meet the freshmen for the first time, you know, I would always ask them, What are you interested in? What are you excited about? And maybe 10 years ago, there would be all kinds of responses a lot, a lot of them being things that I knew something about, like, Oh, I'm excited about string theory, or gravitational waves, or maybe it was something like CRISPR, and genetic engineering and neuroscience. But by five years ago, by far, the most common answer was machine learning. It's they see how it's changing the way we do everything, including how we do science. And so I was lucky to have Robert as a student, because I could, under his tutelage, learn something about machine learning. And he, he already had some big questions on his mind early in his PhD. One being, how far can we go? If we, you know, have access to quantum systems in the lab, and we can measure them? How far can we go at predicting the behavior of other quantum systems, which are different from anything we've encountered in the lab before? How can generalize from that data to make predictions for new systems? And another big question is, and how much better can we do these things? If we process information with quantum machines? And so we've worked on both those questions the last five years.

Kevin Rowney 48:55
That's interesting. I'm sorry. So it's both using classical computers and machine those machine learning algorithms to infer quantum behavior and quantum computing to to infer quantity.

John Preskill 49:05
But yeah, you said it better than I did your kind.

Kevin Rowney 49:10
Note first, so no, fair.

John Preskill 49:12
The first question. We've got a really big challenge, right, because quantum systems of many qubits are very extravagant, you know, they have their you know, the words we use is there's a Hilbert space of unfathomably large dimension of possible quantum states, it seems far beyond our capacity as classical human beings to envision or grasp that extravagant

Kevin Rowney 49:38
Not such an intuitive abstraction. Yeah

John Preskill 49:42
But on the other hand, let's say I want to use classical ML to efficiently make predictions about quantum systems. I got to translate that extravagant quantum system. I've got to convert it into a succinct classical description, which captures physically relevant properties of the system. And that was this idea that we called Classical shadows. It's a way of making measurements that are experimentally feasible today on a quantum platform, you know, which could have, say, hundreds of qubits. And then from those measurement outcomes, we can, first of all, by classical processing, predict many properties of the quantum system. So, you know, it's, there are some properties, we want to predict. What do I mean by properties? Well, you know, physicists often talk about correlation functions, like I'm interested in how correlated are the qubits over here with the qubits over there. And expectation values of operators that sort of act locally on the system, things like that, we we'd like to be able to predict. And of course, one way to do that is just a measure that thing you want to predict over and over again, until you get a good statistical estimate of that quantity, but it's very inefficient. So what we showed is by doing measurements that are experimentally feasible, and are just chosen by sampling from some random ensemble, you can make predictions for a number of properties, which is actually exponential in the number of copies of the system that you measured in the lab. So that was the idea of classical shadows. And so now we have this quantum to classical converter, the classical shadow somehow doesn't tell you everything, right? You a lot about the quantum system, we measure that and the lab for some quantum system of interest. And now we'd like to generalize from that data to predict properties of other systems. So I mean, a kind of futuristic thing you might hope to do, is you've got a big handbook with lots of information about different molecules where you know, you've measured different properties of the molecules. And now you'd like to be able, from that data to generalize and make predictions about other molecules that you'd never synthesized before.

Kevin Rowney 52:04
This is the ON FIRE topic in chemistry right now. Yeah, you know, yeah.

John Preskill 52:08
And so in, you know, a lot of ML is heuristic, which is fine. So, you can't, you don't have guarantees of performance. But we are interested in under what conditions can you actually prove rigorously that you can make accurate predictions? And we succeeded in doing that for a few different cases. And one is that you do have some phase of matter. And, but it can be in a lot of different states, you the way a physicist would describe it is, is there some Hamiltonian which, and this is the ground state is the lowest energy state of this Hamiltonian? And suppose there's some family of Hamiltonians. And we sample from that family and we measure properties, the ground states for those samples. And now we'd like to ask, can I make predictions about properties of the ground states, for other Hamiltonians that are in that same family in that same phase, we found conditions under which it's guaranteed that those predictions are accurate, and that the number of samples that you need, and the amount of classical processing that you need to do to generalize is is efficient?

Kevin Rowney 53:20
Wow, those are cool results. Really amazing. Yeah.

John Preskill 53:24
And I should put in the disclaimer that not only does Robert come up with questions he also answers them -- well not just Robert, we had other collaborators.

Kevin Rowney 53:35
I got, I can't,

John Preskill 53:36
you know, you had to formulate the problem, and then prove theorems. And he does the numerics. And he's really been a joy.

Kevin Rowney 53:48
This is really cool. So and I mean, it's perhaps a little bit rude. Please forgive me if I'm being impolite here, because the theory, the abstractions, beautiful, but I mean, could there be a route towards applications in this domain that would perhaps, I don't know, shed new light on your just ground level performance in classical or quantum ml?

John Preskill 54:08
Yeah, well, you know, I'm a physicist. So I am -- ML is great. I want to understand things, right. I don't want it to just be a black box. That makes predictions. I mean, that's fine. I guess it's a good.

Kevin Rowney 54:24
That's a that's a valid criticism of all of them. And let me the concern about a big stochastic parrot. Yes. And

John Preskill 54:29
so we're we don't know quite what to say about that yet. But I think example, which harkens back to something we discussed earlier. So we have the capacity now in the lab to make new phases of matter. No one's ever seen before. How do you know it's a new phase of matter? Okay. And so one of our results shows that you can efficiently learn to classify phases of matter, under certain assumptions, but at So even if you, well, this a little stronger than what we prove rigorously, but the Numerix indicates that if you're sampling, you know, states from different phases of matter, you can do unsupervised amount to sort them into different phases. Okay? In doing so you've learned some classifying function, some property of the system that tells me phase A is different from Phase B. And so I think stuff like that, as far as, again, applications to science, you might have been asking about more practical application. No, no, anything's fun. Yeah. But as far as application to science, I think now that we have the hardware to explore new phases of matter, we're going to have to know when we found something really new, instead of boring. And so I think these tools are going to be important.

Kevin Rowney 55:52
You need a rigorous analysis tool, you can't just say it looks different. So it's a new, it's a new face. Yeah.

Sebastian Hassinger 55:58
As the practicality as has come up a number of times, I mean, I feel like applications in science are at least as important as applications to computer science or engineering, let's say. In that sense, it feels to me in a way quantum computing is playing the role that the the space race in the in the NASA mission was playing, as you were saying, sort of inspired you to to get into science in the first place, right? It's something that's motivating all of this collaborative codesign, experimentalists and theorists working together, physicists and computer science and engineers all working together to some outcome. We're not even sure what it is. But it's, it's, it seems all good to us, actually.

John Preskill 56:46
You brought up earlier the Solvay meeting. And for for a physicist that has a certain cachet, you know, because all the conferences, going back to 1913 have sometimes been places where historically important discussions take place. And so I thought it was, and this was not a meeting of engineers, really, it's a meeting of scientists, but it was experimental physicists, people who do quantum gravity, people who do quantum matter, people who do theoretical computer science, you know, who have enough in common, that we can have fruitful discussions and share ideas, and the fact that there was a Solvay Conference, which, mostly in the past has been either about high energy physics or condensed matter physics or astrophysics, where the theme of the meeting was the physics of quantum information is an indicator of where physical science is heading. And I think a lot of the excitement comes from these connections between fields that really stimulates progress.

Sebastian Hassinger 57:50
That's fantastic. That seems like a terrific place to end up. That's really a great sort of feeling for the for the, the enticement of the field and the promise of the field. And thank you so much for being with us, John. It's been

Kevin Rowney 58:07
really Yeah.

John Preskill 58:08
Are you telling me an hour is up? All right, it's

Sebastian Hassinger 58:10
right, exactly. You have a meeting to get to.

Kevin Rowney 58:14
I'm trying to be respectful of your time, but also Yeah, did fly by Yes.

John Preskill 58:17
We were having so much fun.

Kevin Rowney 58:19
We really enjoyed this. Thank you so much. Thank you.

John Preskill 58:22
All right, thank you.

Sebastian Hassinger 59:09
That was a great conversation. Super, super interesting. It's such a privilege to be able to talk with people who've been around from pretty much close to day one. I mean, as John said, he overlapped at Caltech with Feynman for about five years. And he's been tracking and involved in the field since the Deutsch paper in 1985. So his roots go very, very deep. And he's you know, he has such an interesting take on, you know, what the threads are critical to his sort of understanding of how the field is progressing.

Kevin Rowney 59:44
So cool. I think also, I was really thrilled by the work of his graduate student on his team, Robert Huang, in terms of -- you know, the cool applications of machine learning and Quantum Information Science. There's just this rich theme that just keeps coming up with these interviews. So well, I got much more along these lines to come, but I can't wait to see where this guy's work.

Sebastian Hassinger 1:00:09
Absolutely. And it is exciting because you know, when I first encountered quantum and machine learning in the same sense, it was, you know, using quantum computers to somehow accelerate machine learning, which they've got a long way to go in that regard. But using ml and using the enormous classical computing power that we have with those algorithms, to better understand quantum systems to exert classical control and these quantum systems, that's extremely promising from a scientific perspective, but also from an engineering, you know, manipulation and control of these systems. Maybe we get better error correction through ml and other techniques as well. So very, very interesting.

Kevin Rowney 1:00:50
We live in a very interesting time, as we say, most episodes

Sebastian Hassinger 1:00:53
It's the best part of this podcast is how mind blowing these conversations are so cool. Yeah. So thanks for joining us. We hope you enjoyed it. If you did, please subscribe and leave a review on Apple podcasts or Google or whatever other platform you listen to, and we'll catch you again next time.

Kevin Rowney 1:01:13
Looking forward to the next time. Okay, that's it for this episode of The New Quantum Era. podcast by Sebastian Hassinger and Kevin Rowney, our cool theme music was composed and played by Omar Costa Homido. production work was done by our wonderful team over at Podfly. If you're at all like us and enjoy this rich, deep and interesting topic, please subscribe to our podcast on whichever platform you may stream from. And even consider if you'd like what you've heard today, reviewing us on iTunes and or mentioning us on your preferred social media platforms. We're just trying to get the word out on this fascinating topic and would really appreciate your help spreading the word and building community. Thank you so much for your time.