Depending on who you listen to, quantum computers are either the biggest technological change coming down the road or just another overhyped bubble. Today we’re talking with a good person to listen to: John Preskill, one of the leaders in modern quantum information science. We talk about what a quantum computer is and promising technologies for actually building them. John emphasizes that quantum computers are tailor-made for simulating the behavior of quantum systems like molecules and materials; whether they will lead to breakthroughs in cryptography or optimization problems is less clear. Then we relate the idea of quantum information back to gravity and the emergence of spacetime. (If you want to build and run your own quantum algorithm, try the IBM Quantum Experience.)
Support Mindscape on Patreon.
John Preskill received his Ph.D. in physics from Harvard University. He is currently the Richard P. Feynman Professor of Theoretical Physics at Caltech and the Davis Leadership Chair at the Institute for Quantum Information and Matter, as well as an Amazon Scholar at Amazon Web Services. Before moving into quantum information, he was a leading researcher in quantum field theory and black holes. He is the winner of multiple bets with Stephen Hawking.
- Caltech Theory Group web page
- Caltech IQIM web page
- Google Scholar publications
- Blog posts at Quantum Frontiers
- Wikipedia
[accordion clicktoclose=”true”][accordion-item tag=”p” state=closed title=”Click to Show Episode Transcript”]Click above to close.
0:00:00.2 Sean Carroll: Hello, everyone, welcome to the Mindscape podcast. I’m your host, Sean Carroll. If you have not been sleeping under a rock or whatever it is in recent years and you have some interest in physics and science more broadly, you will have heard of quantum computers. You all know about quantum mechanics, you all know about computers. There seems to be a new thing on the horizon where we’re going to use quantum mechanics to build a new powerful, different kind of computer. In fact, I sometimes speak in the future tense, it’s happened, we have working quantum computers right now, but so far, they’re not big enough and they’re not efficient enough, they’re not accurate enough to really beat their classical counterparts at any particular kind of calculation that you might want to do.
0:00:43.0 SC: However, the technology is improving, the theoretical understanding is improving, we’re very hopeful that pretty soon quantum computers will be doing things more efficiently for certain kinds of things than classical computers do, so it’s a very obvious fun topic for us to discuss here on Mindscape. We’ve brought in one of the world’s top experts, my Caltech colleague, John Preskill, and John like me, started out as a theoretical particle physicist, cosmologist, gravity person, and he made a very explicit choice to move into quantum information science and quantum computing, and he’s since become one of the world’s leaders in that project.
0:01:20.2 SC: He’s also the best person to talk to about the big picture for quantum computing, because as we’ll discuss in the podcast in the conversation, though he is certainly personally extremely excited about both the intellectual adventure of quantum computing and the technological prospects for making it work, he is also relentlessly careful not to over-hype. So he’s very, very specific about what he thinks quantum computers will be good for, and especially the sort of most obvious thing in the world, which is simulating quantum mechanical systems, like chemical reactions or materials or something like that.
0:02:00.1 SC: It’s less clear that quantum computers are going to be very helpful for giant optimization problems, the traveling salesman problem or something like that, but maybe they will. So we talk about What quantum computers are, what the best technologies are to make them happen, and the prospects for using them, where they will really be useful, what kinds of problems we’ll be addressing with them, why people are so excited about it. At the end of the day, probably will not have a quantum computer in your smartphone, but they probably will be useful for various things we can’t even get anticipate. It’s clear that this is a very exciting time to be thinking about this stuff, we not only talk about quantum computers, but how the ideas from quantum information theory have radically changed our views of quantum gravity and black holes and emergent spacetime, and some of my favorite topics along those lines. This is also an area where John is one of the world’s leaders, so he’s a great person to talk to about it.
0:02:54.6 SC: I’ll mention, as I sometimes do, that we have resources out there for Mindscape listeners, we have a whole website, preposterousuniverse.com/podcast. You can get not only the audio files, obviously, for the different podcast episodes, but we have show notes, so there’ll be links to people’s web pages and their Twitter bios and things like that, and also we have full transcripts for every podcast. So if you want to find something, there is a search engine right there, if you say, well, someone was talking about that on Mindscape, we have very, very good ways of searching the entire Mindscape archive or whatever topic for whatever keywords you want to look for, so check that out, preposterousuniverse.com/podcast. And let’s go.
[music]
0:03:54.7 SC: John Preskill, welcome to the Mindscape podcast.
0:03:56.7 John Preskill: Great to be here, Sean.
0:03:58.8 SC: So you’re someone who did something interesting. I mean, you’ve done many interesting things, but among the interesting things is you had a good career going as a pretty straightforward quantum field theory, particle physicist, cosmologist, the kind of stuff that I grew up doing myself and thinking about, and somewhere you made a, not very dramatic, but a noticeable shift into quantum information and quantum computing. So I guess, can I mix up two questions together, first is why that shift and secondly, do you have any bigger picture thoughts about shifting one’s research focus later in life? You’re at an age when many physicists are thinking about retiring, and in some sense you’re peaking now in terms of your contributions to a very vibrant and active field.
0:04:46.8 JP: The best is yet to come, that’s right. So the shift you’re talking about occurred about 25 years ago and… Well, you were around then, this was the mid-90s. And what happened, I think, which made me more receptive to going in a new direction was the SSC got cancelled. You know, for my generation of particle physics, that was our great hope. We were a little too late, I’m older than you, but I got my PhD in 1980 and missed by a few years the opportunity to participate in solidifying the core theory, the standard model of particle physics.
0:05:35.9 JP: So our big hope was, what comes after the standard model, the so-called beyond the standard model physics, where we thought there were a lot of big questions that we could answer, and that was what the SSC was supposed to do, the Superconducting Super Collider. They started digging a tunnel in Texas and spent a couple of billion dollars, and then for complex political reasons it got cancelled, and so I started to ask myself, well, what now? We were going to be able to do some of that physics eventually, but it was maybe a couple of decades away, and I think that made me more receptive to looking into new things.
0:06:13.5 JP: Now, meanwhile, while I was waiting for the SSC, I was thinking about crazy stuff like how do black holes process information, and that was my excuse for digging a little bit into quantum information. And I learned about things like quantum teleportation and quantum cryptography, partly for fun and partly because I thought those ideas might be helpful for understanding information processing at a more fundamental level. And then something big happened, Peter Shor discovered the factoring algorithm, that was 1994. And that was an eye-opener for many people, including me. It made me feel that this was a serious business, that the difference between problems we could solve with computers and the problems it can’t solve because they’re too hard, that difference shifts because this is a quantum world not a classical one. I just thought that was a really, really interesting idea.
0:07:20.6 JP: And you know how it is when you’re in the middle of a transition like that, you don’t quite see it happening, it seems a little more adiabatic, but over the course of a couple of years, I realized most of the things that I was excited about and thinking about had to do with quantum computing.
0:07:35.3 SC: So it sounds like there’s a little bit of everything, ’cause I can imagine these shifts happen sometimes, like you say, adiabatically, people just gradually shift, sometimes it’s an intentional reflection on what they’re doing, and sometimes there’s some external event that might shift things, and for you it was all three at once.
0:07:54.9 JP: Yeah. Of course, I imagine there’s a little bit of rational reconstruction of what happened in the description I just gave you. It was a somewhat complicated time, ’cause I had students and they were working on various things and some of them got excited about quantum computing and some of them didn’t, but everybody turned out all right.
0:08:16.9 SC: Well, and do you have then any lessons for this question more generally, in academia, even, forget about physics, we don’t really encourage or set up or provide pathways for people to make dramatic career transformations in the middle of their careers in their 40s or 50s. Should we or is the system pretty good as it is?
0:08:38.1 JP: It helps that I have tenure, I didn’t have to worry about that. And so I had that luxury of taking a year or so to acquaint myself with another field. It was easier than because the field was new and there wasn’t as much to learn.
0:08:55.9 SC: It was smaller, yeah.
0:08:58.9 JP: What was really helpful, and you’ve had this experience too, probably, is I taught a course, and I did that for the first time in 1997, and a lot of the stuff was pretty fresh, and I really enjoy synthesizing materials and distilling them to what I think is important, so that was really fun to do, but it also deepened my understanding and made me feel like I was kind of on the cutting edge of the field by the time I finished teaching that course. It was a year-long course. Now, another thing that was nice is that the community was very receptive. You know, it was a small community, it’s bigger now, but you might have wondered if this interloper coming in from particle physics who thinks he’s such a big shot starts doing quantum computing, whether there would be any resentment.
0:09:48.5 JP: And I didn’t encounter anything like that. In fact, people were happy, it sort of validated what they were doing was interesting and valuable that I was so fascinated by it. So it’s a great community, and I learned a lot from people at conferences and summer schools those first couple of years, and that was really fun too.
0:10:11.9 SC: So we’ve already used the words etcetera, but what is a quantum computer? I mean, should we start with what is quantum mechanics? Do you have a favorite way of explaining quantum mechanics, or do we just dive right into what quantum information is?
0:10:25.4 JP: Well, I’ll tell you what the quantum computer is not, it’s not just a computer like the ones we have now, but much, much faster. It really processes information in a very different way. And to sort of jump past to what is quantum mechanics part… Well, not really. I mean, in quantum mechanics, when you have a lot of particles, the description of how those particles behave, that quantum mechanics uses, has great extravagance. If I wanted to describe in terms of ordinary information, classical bits, they way we usually store and process information in the ordinary life. If I wanted to describe what a quantum system with many particles is doing, the number of bits I would need is just inconceivably vast.
0:11:21.9 JP: Now, there’s a catch, though, which is, of course, if I want to get information out of a quantum system, I want ordinary classical information, bits I can write down or put in my head, and to get information out of a quantum system, well, qubits are what we call the fundamental units, the analogue of bits, you have to measure them, and when you measure a qubit, you just get one bit of information. So you know, if you have a few hundred qubits, a description of some typical state of a few hundred qubits in terms of classical information would be something you could never write down because it involves more bits than the number of atoms in the visible universe, but if you want to get information out, you can’t get more than 300 bits out of 300 qubits.
0:12:13.6 JP: So this is the tension in quantum computing that somehow this extravagant quantum world, which is largely hidden from us as a resource, if we can figure out how to use it, but it’s not a simple thing, because we are… Our access to all that rich quantum information is very limited. And so it turns out we can’t use quantum computing to just speed up any problem we want to solve, the problem has to have the right structure and we have to be sufficiently clever to figure out how to use all that rich quantum information that’s behind the barrier that we only have limited ability to penetrate.
0:12:57.0 SC: So when you say qubit, the classical bit is 0 or 1, the qubit is in some superposition of 0 or 1. Probably most Mindscape listeners have heard me say words like that before. And then, of course, when you have multiple qubits, they’re entangled with each other, and it would seem to me like just one qubit in principle, there’s probably any way of thinking about why quantum computers are better, which is entirely wrong, which is that even one qubit has a huge amount of information, because it requires a real number or even in fact a complex number to specify where you are in that qubit land, whereas a classical bit is just 0 or 1, but that’s actually not the important thing. The important thing is the entanglement between the qubits, right?
0:13:40.3 JP: Yeah, and that’s why we seem to be forced, if we want to describe in terms of classical… There’s a mismatch between information in the classical world and in the quantum world because of entanglement. Entanglement of when I have a system with many particles which are highly entangled, that’s the word we use for the characteristic correlations among the parts of a quantum system that has many parts. The correlations among those are just different from correlations among parts of the classical system, and what’s so interesting about entanglement, from my point of view, is that there’s no succinct way of describing it in terms of classical information.
0:14:26.9 SC: And I think that some… One aspect here that maybe the non-experts don’t appreciate as much is that entanglement really wasn’t that big a deal for many, many decades after we invented quantum mechanics, right? I mean, Einstein, Podolsky and Rosen tried to emphasize it, and we had these puzzles, but probably you, just like me, when we took our quantum mechanics classes, there’s a lot of tunneling through barriers and simple harmonic oscillators and solving the Schrödinger equation. Entanglement was kind of there in the background, but not emphasized. Is that an accurate way of putting it?
0:15:02.3 JP: We didn’t learn Bell inequalities. So I’m saying Podolsky and Rosen, that was 1935. And there were some thoughtful responses to that paper which talked about how these correlations in quantum systems are different from classical ones, from Schrödinger for example, who understood pretty well what it meant. But not much happened for 30 years, and then Bell in the mid-60s came up with this idea of Bell inequalities, which in a very visceral way indicated how these correlations are different from the ones we can realize with ordinary information like we encounter in everyday life.
0:15:49.3 JP: And so I could have learned that when I took quantum mechanics, ’cause I took quantum mechanics in the early ’70s and Bell did that in the mid-60s, but it just… It wasn’t part of the quantum curriculum, I guess it wasn’t for you either. Now, it happened though, something… And maybe this also planted a seed, which partially explains the answer to your first question about how I made this transition. When I was an undergraduate at Princeton, we were supposed to do a research project when, as juniors, and who knows how to do that, so you have to go around and knock on doors and see if anybody has any idea, and somebody suggested to me, ’cause I said, as undergraduates so often do, “Oh, I’m interested in the interpretation of quantum theory,” to one of the professors.
0:16:40.6 SC: He said, “Well, you should look into the EPR paradox,” which I had never heard of, but it turned out there was an instructor who had just arrived at Princeton that fall, who had just done an experiment for his PhD With John Clauser, that was Stuart Freedman. Freeman and Klauser did this experiment in the early ’70s, which was one of… Well, the first to sort of convincingly show that these Bell inequalities are violated in a quantum system, and so there really is a difference between quantum and classical information. And so I pumped him a little bit and I wrote a little paper about EPR, so I thought about that some, and so that probably also made me more receptive to some of this quantum information stuff when 20 years ago I realized it was pretty interesting.
0:17:36.7 SC: Well, that’s interesting because, like you say, every undergraduate wants to think about the interpretation of quantum mechanics, and usually that instinct is just beaten out of them, but for you at least it turned into something productive. Now, years later, I take it that what one’s stance is about the interpretation of quantum mechanics doesn’t really enter into your everyday life as a quantum information theorist, right? I talk about the interpretation of quantum mechanics a lot, but let’s just be clear to the audience, the practitioners in the field kind of don’t spend their time worrying about that. Is that also accurate?
0:18:11.3 JP: It’s less accurate than it used to be, I think, at least in the quantum information community, some young people, I think, are attracted to quantum information because they think it is revealing concerning the foundations of quantum theory. But even so, I accept your statement, that it’s not what… It’s not our bread and butter even in quantum information, there are other things that we’re usually thinking about, unless we’re listening to you or reading one of your books or something, of course.
0:18:48.0 SC: Having said that, do you have a favorite formulation of quantum mechanics at the fundamental level?
0:18:56.1 JP: I’m an Everettian, so I think you and I are pretty aligned, actually.
0:19:01.5 SC: Yeah.
0:19:03.3 JP: I’m comfortable with nothing happening in the world besides unitary evolution, if your audience knows what I mean by that, that measurement isn’t something fundamentally different. And you know, what I… And people say, well, it’s… But it requires all these worlds and stuff, so that’s another place where the word extravagant seems appropriate, so we need to have this very, very rich description of experience because we include all the alternatives. But I guess my question would be then, well, suppose it were true, how would the world be different than we actually experience and observe it to be. And so I just find that appealing, it seems minimal, there’s nothing happening but the Schrödinger equation and things are evolving. And if we can reconcile that with what we observed about physics, then I think that’s… I mean, if you have a firmer interpretation, that’s fine with me, I’m not going to get dogmatic about it or anything, but I’m pretty happy with that.
0:20:12.6 SC: That sounds good to me. I’m not going to, I’m not… Let’s just declare victory right there and move on to the computer.
0:20:17.2 JP: We got nothing to argue about.
0:20:18.8 SC: Yeah, exactly. So we have these qubits, they can be superpositions of 0 or 1. If you measure them, you’re going to get either 0 or 1, or whatever product, whatever actual physical property you’re measuring of them, and they’re entangled. So can we just be a little bit more specific to help the audience out there visualize what exactly does that mean that they’re entangled and how does that mean that there’s more oomph to the computational power?
0:20:45.9 JP: Well, what it means is that even if we have a complete description of the full system of all the particles, and by complete, I mean the most complete description that we think the laws of quantum mechanics will allow, when we look at part of the system, we can’t predict what we’re going to see. We can have a complete description of the whole thing, but still be very uncertain about the parts, that’s what entanglement is. So you could imagine like you have a book and it’s 100 pages long, and if it’s an ordinary book, you can read the pages one at a time, or you can even give each page to a different friend and they all read the pages, then they get together and talk, and they can reconstruct everything that’s in the book, but if it’s written in qubits and the pages are highly entangled with one another, you can look at the pages one at a time, but you just see random gibberish.
0:21:51.4 JP: You don’t get any information from it, or essentially none, and that’s true even if you look at all the page one at a time, you’re still missing almost all the information that’s in that quantum book. There’s information there, you just can’t read it that way, ’cause the information isn’t printed on the individual pages, it’s encoded almost entirely in how these pages are correlated with one another. That’s the entanglement, and describing those correlations in terms of ordinary bits, that’s what’s so expensive. To describe that the way we usually describe and think about information is completely unreasonably costly. And so that’s the way I would look at the reason why we can’t just use the computers we have today and simulate what’s going on in some complex quantum system, it’s just too hard to do.
0:22:50.3 SC: And the way that that helps us with the computer is if I think about what a computer is, it has memory, it has processors, it has algorithms, it’s really sort of the processors that are being improved here because they can sort of… Well, let’s ask it this way. There’s a couple of wrong ways of saying it, and let’s get those out of our audience’s mind. One way that I know Scott Aaronson always gets annoyed at is that it’s not really like there’s a whole bunch of universes doing separate computations and eventually finding the right one, right? Do you agree with him that that is a bad way of thinking about quantum computing?
0:23:27.4 JP: It’s not completely wrong, but it’s very misleading, because it would lead you to expect that we can speed up anything on a quantum computer, any problem we want to solve. And the catch is what I said earlier, that in the end, you’re going to have to measure something, and you can only get a limited amount of information out, even if all these computations in some fanciful description of what’s going on are occurring in parallel, a vast number of computations, we’ve got to eventually bring them together and read something on it. And so the art of making a quantum algorithm that’s powerful, is that when you combine them all together, you get something useful and it’s not so obvious that you’ll be able to do that.
0:24:23.8 SC: No, no, no.
0:24:24.5 JP: The entanglement’s very complicated to describe classically, I just claimed, but does that make it useful? It’s not so obvious.
0:24:31.3 SC: So said in another way, what’s obvious is that it requires a lot more information to specify what’s going on in an entangled quantum state than in a bunch of classical qubits, but it is not at all obvious that you can put that to useful work.
0:24:46.3 JP: Yeah, and I think the best reason we have to think that quantum computers are powerful is that we can’t simulate them with ordinary classical computers. Now, that doesn’t tell us exactly what they’re good for, but it does kind of point us in the right direction. This was Feynman’s idea of 40 years ago, almost exactly 40 years ago, it was in May of 1981 that he proposed that if you want to simulate quantum systems, complex quantum systems with many particles on a computer, you shouldn’t use an ordinary computer ’cause that’ll be just too hard, it should be a quantum computer. And that’s still the best idea we have about how to apply quantum computation to something that people will care about, because simulating quantum systems is important.
0:25:43.6 JP: If we could, we would simulate complex materials and complex molecules on our computers now and predict how they’ll behave, and that would point us presumably towards new types of materials with properties that we find useful or new types of catalysts or pharmaceuticals and so on. And to a certain extent, people do that, of course, in quantum chemistry and in computational quantum physics, but it’s very limited what we can do with ordinary computers, just because it’s so hard to simulate highly entangled quantum systems, and that’s what we have.
0:26:19.6 JP: Like if you have many particles and they’re interacting strongly quantum mechanically with one another, it’s really hard to simulate. With a quantum computer, we think we’d be able to do that efficiently, and that’s… From what we can currently foresee, that’s the application which is most likely to have a big impact on the world. Now, if quantum computing is so different from the information processing we’ve been familiar with up to now, then we probably have a very limited ability to forecast what its most important applications are going to be, but based on what we can currently project and have some understanding of, it’s those applications to simulating complex matter, many particles interacting quantum mechanically, which will be the most important application.
0:27:11.1 JP: Now, it came as a bit of a shock, and that’s what caught my attention in 1994, that you could use quantum computers to speed up solutions to other problems which don’t on the face of it seem to have anything to do with quantum computing, and Shor discovered that quantum computers would be very good at breaking codes, that they could do things like finding prime factors of very large numbers. And of course, that caused interest in the field to ramp up sharply, because people care about breaking codes, some people do, but in the long run, I don’t think that’s likely to be the application that’s going to be of broad interest and might affect people’s everyday lives.
0:28:00.5 JP: Well, it will in the short run, it’s disruptive that quantum computers will be able to break the codes that we use, and which we do use in our everyday lives. Whenever you’re going to a website that has that little padlock next to the URL, you’re using some public key crypto system, and the ones that are in most widespread use today, quantum computers will be able to break. Those crypto systems are based on the presumption that although, by doing a hard computation, you can break the codes, that those computations are just too hard. So if you wanted to break RSA, which is one of the most widely-used crypto systems now, you’d have to be able to factor numbers that are 2000 bits long, and that’s still pretty far out of reach with the most powerful super computers we have now.
0:28:57.7 JP: With quantum computers, that won’t be a hard problem, and then we’ll have to protect our privacy a different way. And what most likely is going to happen is people will switch to new public key crypto systems based on problems which we don’t think quantum computers can solve efficiently. And that will mean we’ll all have to change our browser software or something like that, but for most people, maybe it won’t have such a big impact on their everyday lives.
0:29:34.7 SC: It’s always fun listening to you talk about this stuff, because you’re clearly super excited about quantum computers, but also just so extremely careful about not over-hyping them that I can always see the battle going back and forth between you, in your own mind, like saying how interesting it is, but you can see how people who only know a little bit about it could get carried away and you want to protect against that.
0:29:56.3 JP: Well, there’s a lot of the hype. It’s an issue. What’s happening in the last few years seems kind of extraordinary to those of us who’ve been working in the field for 20-plus years, ’cause it started out… Well, from the start part of what was exciting about it for me is that it had an experimental side and it had a theoretical side, and you might say kind a big gap between the two, but people were really trying to do quantum computing already in the ’90s, and it came along at an opportune time that when this factoring algorithm was discovered and people started to think seriously about what other applications there would be, the experimental physicists had just in the past decade or so, acquired the tools to manipulate single atoms and to manipulate single electrons and things like that, with a motivation which initially didn’t have much to do with computing.
0:31:04.4 JP: One motivation was we wanted better clocks, and the technology for making really good atomic clocks is not that different from the technology we need to operate a quantum computer. So it was fun that there was interaction between the theory and the experiment, but it wasn’t like we all were going to launch startups 15 or 20 years ago…
0:31:29.7 SC: Some of you did.
0:31:30.8 JP: Now everybody’s launching startups, and there’s a lot of money sloshing around, and there are also big companies that are making hefty investments, like Microsoft and IBM and Google and Intel and actually Amazon, which I have some affiliation with now. And people are excited, and the excitement to a certain extent is good, optimism propels things forward. It creates opportunities for young people, the investment accelerates progress, but if expectations are unrealistic, we’ll get burned eventually. And I think… I do expect that quantum computing is going to have a big impact on society. We don’t know exactly how far off that is, it’s not going to happen in five years. I don’t think. Some people think it will happen in five years or ten, I think that’s unlikely, but that doesn’t mean we shouldn’t be pushing ahead and developing the technology.
0:32:44.8 JP: And you know, there are related technologies which might have practical impact on a shorter time scale, like sensing technologies. Quantum sensing has applications that are developing, that better quantum sensors, they have applications to fundamental physics that you and I might be interested in, but also everybody needs to measure things more accurately and with higher resolution and less invasively, that’s important in materials, it’s important in medicine. And the technology for quantum sensing has a lot in common with the technology for quantum computing. I think the excitement to a certain extent is warranted, but we shouldn’t expect everything to happen at once. It’s going to take sustained investment, it’s going to take a lot of effort to make quantum computing a big thing in the sense that it really benefits humankind.
0:33:58.3 SC: Well, I get it, because we all know the technology is changing the world and it’s advancing rapidly, and everyone wants to know the next big thing, and when you take computing, which everyone knows is important, and marry it with quantum, which everyone knows is mysterious, it sounds like it could change the world very dramatically. So I’m sure there’s sort of a feedback mechanism between people who do and do not understand quantum computing getting excited about the very near-term prospects.
0:34:25.8 JP: Well, here’s how it looks to me, that right now, quantum computers are at a pretty interesting stage. What’s interesting? Well, we can build systems with 50 to 100 qubits. That’s interesting because that’s already large enough so that it’s quite hard just by brute force to simulate with our most powerful classical computers what the quantum computer is doing. So in that sense, it’s natural to start thinking seriously about what could the applications be for these near-term quantum devices, what kind of problems can they solve.
0:35:14.7 JP: But they’re really noisy. In other words, the hardware isn’t very reliable, so when you do a quantum computation, just like with the classical one, you have to put together a lot of elementary operations, and then at the end you read out a result. But with the best quantum hardware we have now, in devices that have lots of qubits, when you do the fundamental two qubit operations, the entangled operations on pairs of qubits, you make an error like once every couple hundred times. So you can’t put together a circuit with more than 1000 or so gates and read out a result is useful. And that’s a real limitation.
0:36:02.1 JP: Now, we understand that even if the hardware is noisy, as long as it’s not really, really noisy, there’s an idea called quantum error correction that we can use to control the errors, but that’s very, very expensive in terms of overhead, the number of physical qubits that you need, particularly with the kinds of error rates we have now, or with the physical hardware, or even if it’s a little bit better than what we have now. So I think it’s really, really important to make better hardware to have operations with error rates which aren’t 1% or half a percent, but 10 to the minus 5 or something like that.
0:36:48.9 JP: Now, that’s really hard, but I’d like to see the focus there because that’ll take us more rapidly to the regime where quantum computers can really solve things that we can’t solve classically, including these applications to physics and chemistry that at least for people in the physical sciences are exciting, and something we’re impatient to realize. If the focus is instead on, well, can we find something interesting to do with these very limited quantum devices that we have now. Well…
[automated voice]
0:37:36.4 JP: You see, even Siri doesn’t get it. Then I’m afraid we might wind up with nothing ’cause it’s possible… Well, look, it’s good to try things. You’ve got to experiment, you’ve got this cool thing and you want to see what it can do, and I think as a physics discovery tool it’s already interesting, ’cause we can study the behavior of a quantum system that’s strongly interacting and becomes highly entangled, which we can’t simulate, and understanding the things they can do just from a point of view of understanding physics better, there I do see progress happening in five years. But as far as a practical application that customers will want to pay for because they’ll be able to improve their financial portfolio or solve some optimization problem, I think we’ll have to be pretty lucky to see applications like that that are really impactful in, say, the next five years. But that’s where a lot of the focus is and…
0:38:44.0 SC: I will confess, because…
0:38:45.6 JP: I think we should be thinking about the longer term. If we’re going to make this into a real thing, then now we have to focus on the really hard problems of advancing the technology.
0:38:54.5 SC: I’ve said this before loud on the podcast, but not to you, which is a confession that one of the many dumb big mistakes I’ve made in my physics career was, I saw a paper by you, I think it was, in the 1990s about quantum error correction, and I instantly thought myself like that is the most boring thing anybody could be thinking about. I mean, obviously important if you want to build something, but it doesn’t seem like the sort of intellectual fun that we all like to have as theoretical physicists. But I was utterly wrong, right, maybe you can let the audience know a little bit that there really is a fascinating set of ideas associated with quantum error correction, not just an engineering project to make the computers better.
0:39:38.5 JP: Well, I’m not an engineer, I’m a physicist. And even back in the ’90s, what I thought was most exciting for me about quantum computing is that it gives us a different way of looking at problems in basic physics, including in condensed matter physics, but also in quantum field theory and quantum gravity, the kind of stuff that you and I like. I think quantum error correction is as important as contribution to science as the discovery that there are fast quantum algorithms, that there’s some problems that we can solve. ‘Cause it’s really an amazing thing.
0:40:31.3 JP: We normally, for good reasons, think, well, quantum mechanics, that’s about little things, atoms, single electrons or whatever, and big things are classical, and we think that for a good reason, because of the phenomenon we call decoherence. When you scale a system up it’s very hard to isolate it from the outside world, it inevitably interacts with its environment, and those interactions with the environment sort of cause the quantum information to leak to the outside, and it doesn’t behave like a quantum system anymore, it behaves like something we can describe classically, essentially because information about what the system is doing is redundantly stored in everything all around it, and that’s not quantum anymore, that becomes classical.
0:41:25.4 JP: And what quantum error correction is about is that in principle, we can scale a system up many particles doing very complicated quantum mechanical things and keep it quantum. And we don’t do that by perfectly isolating it from the environment, because that’s just too hard. But what we do is we take the information that the quantum system is processing and we make that invisible to the environment, and you know how we do it with entanglement? We take the information that we want to protect and we store it in the form of some very highly entangled state shared by many particles, and then the environment comes along and it’s kicking the system and peering into it, and it’s looking at the particles a few at a time, but because the system is so entangled, it can’t see the protected information when it interacts with just a few particles at a time, just like that 100-page book I was talking about.
0:42:26.9 JP: When you look at one page, you don’t see any of the information that’s in the book. The information that we’ve cleverly encoded it in such a way that when you look at just parts of the system one at a time, you don’t see the encoded information at all, and therefore, it can be robust. And we’ve also figured out how you can process it even though it’s encoded in that very highly entangled form, and that’s a tremendous insight that we can make really complex, large quantum things behave quantumly despite the effects of decoherence, which we have correctly believed explains why big things are classical and the little things are quantum. You can outsmart decoherence, that’s the lesson.
0:43:16.4 JP: And this is was controversial for a while, you know, back in the mid-90s, not surprisingly, people started talking about quantum computing and you could break codes and all that, and good physicists said that’ll never happen, because the physicists who were very experienced with decoherence, including in the lab like Serge Haroche, who won a Nobel Prize for figuring out how decoherence works. And he said it was the theorists’ dream but the experimentalists’ nightmare, because you can never really control the quantum system, it’s really hard. And we’re not all the way there yet. We’re making progress and we understand conceptually how to scale things up and still make a complex quantum system work, and I think that’s just an amazing scientific breakthrough.
0:44:05.7 SC: Let me see if I can rephrase it just so I know that I understand what’s going on here. As you said, there’s a lot of richness and complexity in the quantum system itself, but when we look at it, when we measure it, in some sense, we only extract a classical result with some probability. The qubit can be up or down, but we see only one or the other, and likewise for many qubits. And decoherence in a sense is making a measurement of the system, it’s the environment, the outside world becoming entangled and then sort of ruining the quantumness, like you said. So the quantum error correction game is putting the information into the quantum system in such a way that it is not being measured away by the environment. Is that fair?
0:44:48.9 JP: You said it much better than I do. You should be a podcaster.
0:44:52.5 SC: I should write books about this, yeah.
0:44:56.7 JP: That’s exactly what I meant.
0:44:58.4 SC: Good, very, very good. Then let’s make it real. I know that neither one of us is an engineer, but what are your favorite ways to make qubits and entangle them? What are your favorite physical manifestations here that we could build into a computer?
0:45:14.2 JP: Well, you mean what kind of hardware do I like for quantum…
0:45:18.3 SC: Yeah, what kind of qubits? What’s your favorite qubit?
0:45:19.0 JP: Well, first of all, I think it’s important and great that there are a lot of different hardware approaches that people are pursuing and trying to develop, because we don’t have a clear understanding of what type of quantum hardware is going to have the best long-term prospects for scaling up to large systems. Right now, the two leading technologies are superconducting circuits and atomic qubits, usually ions, electrically charged qubits which can trap with electromagnetic fields. So what do you need? If you’re going to have quantum hardware, what do you need?
0:46:03.9 JP: Well, you need qubits. You would like to have a system where you can scale it up to many qubits. You want them to have long coherence times, as we say, in other words, for it to take a long time for the environment to measure the qubit. And then you want to be able to manipulate them quickly compared to that time scale, you want to be able to get them to do something before the environment spoils it all, and of course, you want to be able to read them out in the end, you want to be able to measure them accurately.
0:46:35.5 JP: So it’s tricky to come up with a system that meets all of those desiderata. But atoms are pretty good qubits. You can take an atom and you can control it with lasers, for example, and it could be in either its ground state, its lowest energy state, or some excited state, you’d like that to be an excited state that doesn’t decay back to the ground state very quickly, that’s how you get the long coherence time. So you can actually create with a laser a state which is a superposition of the ground state and the excited state, you can manipulate that. That’s good.
0:47:13.6 JP: You can read it out pretty easily, I mean, you can do it with lasers again. You can shine a laser with the right frequency on an atom, and if it’s in one state, like the ground state, it will absorb the light and readmit it very rapidly, so it glows, but if it’s in another state, then the light doesn’t see it, so it stays dark, so you shine a laser and some of them shine, some of them don’t, those are the zeros and ones. So that all works great. As with other quantum computing technologies, the hardest part is you’ve got to get qubits to interact, you have to be able to perform entangling operations on pairs of qubits, and that’s why it’s useful for them to be electrically charged.
0:48:00.6 JP: Then they have Coulomb interactions, the propulsive interactions, and you can leverage that to couple together the internal states of the qubits and their motional states as they’re vibrating in a trap, and that’s how you can get them to interact in a pretty well-controlled way and do pretty good entangling two qubit gates. So that’s a good technology. Now, superconducting circuits is something really completely different, but it has… It satisfies all the desiderata. Now, what’s being manipulated is the collective motion of billions of pairs of electrons, okay, it’s a circuit.
0:48:43.0 SC: Or a single qubit.
0:48:44.8 JP: But it’s very, very cold, it’s at 10 or 20 millikelvin or something like that. So it’s a material that conducts electricity without resistance, that’s what we mean by a superconductor, but you make it much colder than it needs to be to be a superconductor because you don’t want to have a lot of stray… Well, essentially, light, except it’s microwave light at those low temperatures, which could be interacting with the electron motion, and now you can have… This isn’t the way you really do it in practice for practical reasons, but you can imagine you’ve got a loop of wire, it’s conducting electricity without resistance, but it could be going around either clockwise or counterclockwise, and those are like the two states of the qubit, and those can also be in superposition, it can be a little bit of going clockwise, a little bit of going counterclockwise, and you can read out these circuits.
0:49:53.2 JP: The key thing is… Well, actually, the way you do it is you couple to another thing, like a microwave resonator, just like the thing that… A wave guide, just a photon wave guide, and the frequency of that wave guide depends on whether the qubit’s in one state or another so you can read it out. So that’s all great too. Now, these are about the best technologies we have, and in both cases, you can build devices with tens of qubits, but again, those entangling two qubit gates have pretty high error rates, like at the 1% level, and we’ve got to do much better than that.
0:50:33.9 JP: Now, I looking forward, I do have a preference for the superconducting qubits because there are… There’s a lot of opportunity to sort of fool around with the electrical engineering of the superconducting devices. Well, I wouldn’t even put it that way, ’cause it’s more fundamental than that. You have a lot of… There are a lot of tricks you can play and there are a few ideas we’re pursuing that might result in having much, much better performance for the gates, and that will make a huge difference, ’cause it means in the near term, we can execute larger circuits and still get reasonable signals and noise when we read out, and in the longer term, it means we’ll be able to do error correction with a lower cost, and that’ll make it less expensive to scale things up further.
0:51:25.5 SC: That’s interesting…
0:51:26.3 JP: There could be big surprises coming from other technologies. A few years ago, I wouldn’t have mentioned what is now one of the most promising approaches, what we call Rydberg atoms. It just means an atom now is either in its ground state or it’s really, really highly excited with some electron in a very large orbit, and those guys have big dipole moments, so they can interact strongly with one another, that makes it… Well, that way they have interesting interactions, then they can interact quickly, you can read them out, it’s got a lot of great features, and that’s really come along rapidly in just the last couple of years. And there could be other things like that which sort of take a leap forward, ’cause we’re really in early days, I think as far as the technology goes.
0:52:18.6 SC: It’s interesting that your favorite is the superconducting circuits, ’cause when you mentioned the atoms, there seems to be an obvious advantage of atoms, namely they are small, so we could pack a lot of them close together. But maybe our ambitions are not to have 10 to the 10 qubits, but only thousands of qubits, and so we don’t need to make them that small, but what is the… Do you need space for the atoms?
0:52:44.3 JP: I’d probably want to have millions… Millions of physical qubits, yeah, and that might be a big thing, but it’s not so easy with the atoms either. So if you use this trapped ion technology, remember the way you get the atoms to interact with one another is through their Coulomb interactions, the way they’re vibrating in the trap, and if you put in more and more ions, you get more and more normal modes of vibration, and then you’re going to try to drive them, they correspond to different microwave frequencies, and you’re driving it with lasers, which are optical, but you can have differences between frequencies of different lasers, which can excite different modes, but it gets to be a pretty thick forest of normal modes if you put in too many ions.
0:53:36.7 JP: So it gets very difficult once you have more than, I don’t know, 32, 64 ions, something like that. And so then you’ve got to… It will have to be some kind of modular design, so you’ll have a lot of processors, maybe each one has tens of ions, and you’ve got to get them to talk to one another so they can share entanglement, and so you need some way of designing an interconnect between traps. People have ideas about how to do that, but it’s hard. Or otherwise, you can have a trap with lots of zones where you have a few ions in one zone and a few ions in another, but then you have to be able to move the ions between zones so you can transfer entanglement from one zone to another, if you’re going to get a highly entangled state with many ions, and people are working on that too. But it’s hard. Everything’s hard.
0:54:30.9 SC: Yeah, everything’s hard.
0:54:31.0 JP: Nobody said it would be easy. Quantum computing is hard, that’s why we haven’t made more progress. Well, we made a lot of progress, but it’s only… You know, it’s been decades of improvement in the design of the qubits and the control and the fabrication and all that, but we still have a long way to go.
0:54:49.9 SC: And how do you get the superconducting loops to entangle with each other and interact?
0:54:55.7 JP: Well, the superconducting circuits, they have magnetic fields and electric fields, so they can interact either capacititively, that means the electric field or inductively, that means the magnetic field, and of course, the art is to control those things well.
0:55:15.8 SC: Yeah. What should I be visualizing in terms of the geometry, are these arranged in a line or a plane or a 3D grid?
0:55:27.4 JP: Well, 3D is hard, but there are two-dimensional level layouts. Pretty much the state of the art still in superconducting quantum computers is this device that Google built a couple of years ago now, which they call Sycamore. It has 53 working qubits. They’re in a two-dimensional array, you can execute entangling operations on pairs of qubits in that array which are neighbors to one another, and they are able to do a sequence of layers of such entangling gates. In an experiment they did in 2019, they did up to 20 layers and something like a thousand operations altogether, and because of the noise that I keep talking about, then when they measured all 53 qubits at the end, they get the right answer about one time out of 500, so not such good signal noise.
0:56:35.3 JP: But then they can run it millions of times in just a few minutes, the same circuit over and over again, and then they get a good enough sample of the output to have something that’s statistically useful, and already that… Quantum computation is something that’s really hard to simulate, 53 qubits is a lot. And with the best classical methods that we have to simulate what Sycamore is doing in just a few minutes on a classical supercomputer takes at least days, so in that sense… You know, sometimes I say it’s quantum David versus classical Goliath. Because the classical computer is this huge system, it covers to the equivalent of two tennis courts and it’s burning megawatts of power.
0:57:32.2 JP: The Sycamore, it’s just this little chip, you know? Just doing its thing inside a dilution refrigerator. And yet, in a sense, it’s doing something that Goliath has a very hard time keeping up with. Now, it’s not doing anything useful.
0:57:50.8 SC: As yet, yeah.
0:57:52.2 JP: The computation that it’s doing is useful for only one purpose, to demonstrate that the quantum device can do something that’s really hard for a classical device to do.
0:58:04.4 SC: You mentioned that it’s in a dilution refrigerator, even though it’s tiny. I mean, do we imagine that there will ever be a day when we will have iPhones with quantum computers inside?
0:58:17.0 JP: Well, if we’re carrying them in our pockets, they probably won’t be at 20 millik, but it could be a different technology. There are quantum technologies that work at room temperature. And why would you want to have it in your iPhone, that’s another question. You might want to have it as sort of a smart card, you know, if we haven’t talked about quantum cryptography, but which is another intellectually interesting idea where exactly what the best use case is is still a little bit murky. But the great thing about quantum information from the point of view of security, is that you can eavesdrop on quantum systems without disturbing them in some detectable way. This is what’s really different, or one of the things that’s really different.
0:59:21.6 JP: We’ve talked about entanglement, but one thing that’s really different about quantum information from classical information is you can look at classical information or you can copy it and you don’t have to change it, but if it’s quantum information and you observe it, you disturb it.
0:59:37.8 SC: It’s gone, yeah.
0:59:39.9 JP: And that actually can be the basis of a kind of cryptographic scheme where you’re sending quantum states and if somebody tries to eavesdrop, you can do a test to see that the state has been disturbed, and you can base security then not on some assumption about… The computation you would need to break the protocol is too hard to do in any reasonable amount of time, it’s really based on the laws of physics that you can collect information without creating a disturbance. So maybe you’ll want to have a little quantum, I don’t know, smart card that you can carry around, so when you go to the ATM, I don’t know if people will be going to the ATM by then, but wherever you go, you can have someone verify your card but nobody would be able to copy it.
1:00:28.2 SC: I have no idea what good a quantum iPhone would be or a quantum computer inside the iPhone or smartphone would be, but I am 99.99% sure that it would be incredibly useful just because even I’m old enough that I remember a day when people said, why would you ever need more than 64K of RAM in your personal computer? I don’t see what the usefulness of that would be, and then I remember when people said, why do we need web browsers? We have the phone book and we have other ways to do things. And so even if I can’t foresee what the applications are going to be, there are different kinds of things, as you’ve been saying, that you can do with quantum algorithms and computers, and so someone will come up with ways to put them to use is my conviction.
1:01:08.5 JP: Well, you remember when nobody could imagine why anyone would want a computer at home, usually the explanation was that you can organize your recipes.
1:01:18.8 SC: Yeah, my explanation why you need a web browser at the time was, well, I can order a pizza on a web page. And people said, well, we have ways, we have ways of doing that, other ways, but… Okay, let’s get more into what exactly we can do with the quantum computers. One thing that slipped by there, which ’cause I know some people are not completely convinced or haven’t heard of yet, we have quantum computers, they are working, right? There’s the IBM quantum experience where you can go on to a web page and run a little quantum algorithm yourself, so it’s not like the existence of the computers is the issue, it’s all these scalings and error correction and decoherence is the issue, yeah?
1:01:58.2 JP: There are quantum computers, you can use them, and there are several different hardware providers where you can have cloud access. Actually, AWS has the system they call Braket, where you can access an ion trap quantum computer and a superconducting quantum computer. But they’re not useful yet, you’re not able to solve problems you can’t solve other ways. Now, that doesn’t mean people… It doesn’t make sense for people to be playing around with them, it probably does make sense, because they are going to be useful, and if you have a company for which solving hard computational problems is of value, then you might want to have some quantum expertise on board to have some people who are familiar with what quantum computers can potentially do and explore the potential applications to the problems they’re interested in.
1:03:09.7 JP: But how long it will be before quantum computers can solve problems that people really care about, say for business reasons, and solve them with a speed up relative to what the best classical computers running the best algorithms for solving the same problems, how long that’s going to be, I don’t think anybody really knows that.
1:03:34.9 SC: That sounds perfectly fair.
1:03:35.8 JP: There’s a lot of interest in optimization, just ’cause everybody wants to optimize things, and there are heuristic ideas about how you can use a quantum computer to skip up… Sorry, to speed up optimization, but we don’t really have very convincing theoretical arguments that quantum computers will give big speed-ups for optimization, they might. We know there is, we know they can speed up exhaustive search, but that’s not a spectacular speed-up. If you’re looking for the solution to the traveling salesman problem or something like that, you want to find a route that visits a whole bunch of cities which is as short as possible, you can always solve that problem by trying all the routes and finding the one that’s shortest, but that’s completely inefficient. Nobody can solve the problem that way for more than a few cities.
1:04:30.8 JP: And exhaustive search isn’t a good way to solve it, with the quantum computing can speed up exhaustive search, but it doesn’t speed it up by a spectacular amount the way it speeds up factoring. The time… That’s assuming you had quantum hardware with the same clock speed, the same number of operations per second, as your classical computer. I can solve the problem in the square root of the time it would take to solve it classically. That’s something, but it’s probably going to be a long time before quantum computers are of high enough quality for us to be able to take advantage of that speed-up.
1:05:10.8 JP: So I think the speed-ups that are going to have an impact first will be the ones where we think it really is an exponential speed-up, a spectacular speed-up. Solving those problems in material science and chemistry is an example, but those require lots of qubits and that’s probably going to be a while. Now, just because we… Theorists can’t guarantee that quantum computers are going to speed up certain problems, it doesn’t mean that it’s not going to happen. It’s important to experiment.
1:05:52.5 JP: Classically, we have lots of examples where there are heuristic algorithms which turn out to have a lot of practical value, even though the theorists can’t very well explain why they’re so valuable. Machine learning is the most obvious example now. AI is having a big impact. You can drive autonomous vehicles, you can win at Go, you can recognize faces using deep learning networks, and we have very limited theoretical understanding of why for those applications we can train those networks efficiently. People just tried it and found out that it worked. So if you’re an optimist, you can say, well, let’s try a lot of things on these near-term quantum devices and see if something works. And that’s a worthwhile thing to do, and that’s one way in which these devices that are now available on the cloud will be used and are being used.
1:06:52.3 JP: But in the near term, you could say maybe the most important application for these quantum computers we have now is they can help us learn to make bigger quantum computers, and this quantum error correction idea I keep harking on, we’ve gotten to the point now where we can start to try out quantum error correction on the devices we have now and try to improve it and get it to run more efficiently at lower overhead costs, come up with new ideas. We have to see whether the noise in the devices we can really build has the right properties for quantum error correction to work effectively. And that’s interesting scientifically and I think also interesting from the point of view of moving the technology to the next level. So that’s another thing we should be doing with these computers that we have now.
1:07:48.2 SC: Maybe we can be a little bit more specific about what our theoretical knowledge is about the amount of quantum speed-up we can either expect or hope for. So my understanding is there definitely is speed-up in certain well-understood cases. This exponential speed-up, the speed-up that we would really like to see, what’s the status there? Is it that we… It’s possible that every problem can be exponentially sped up, or do we know that there are any problems that could be exponentially sped up? What do we think about that?
1:08:18.2 JP: Well, I guess it depends partly on what your standard for rigor is.
1:08:23.1 SC: It’s pretty low.
1:08:24.8 JP: I would say we have pretty convincing arguments that quantum computers will not be able to efficiently solve exactly the NP hard problems. You know, the hard instances of the so-called NP hard problems. NP means the problems where when we find a solution, we can efficiently verify that the solution is correct. The traveling salesman problem that I mentioned is an example of a so-called NP hard problem, that means it’s an NP, once you find the route, you can check that the route has a total mileage less than some specified number. It’s hard to find the route, and that problem has the property that if you can solve that efficiently, then you can solve a ton of other optimization problems efficiently, and that would be of great interest for many applications.
1:09:32.3 JP: We don’t think quantum computers can find exact solutions to those problems efficiently. Now, they may be able to find approximate solutions with some advantage over classical computers, we don’t know that for sure, that’s really more of a speculation, but that’s something we should continue to strive for and try to clarify. And that’s part of what is motivating the effort to use these near-term quantum devices to see if we can speed up optimization.
1:10:15.3 JP: Well, I mentioned factoring a couple of times, so what’s the deal with factoring? The truth is, we can’t prove that factoring is hard classically. It’s not one of these NP hard problems.
1:10:28.4 SC: It’s pretty hard. But not that hard.
1:10:31.2 JP: Well, we think… No, why do we think it’s hard? Because really, really smart people have tried very, very hard to come up with a better factoring algorithm, and they’ve been doing it for decades and they haven’t succeeded. But some brilliant student may come along tomorrow and come up with factoring algorithm, that’s possible, but we do know, theoretically, if we have a quantum computer that operates reliably, it will be able to solve factoring efficiently. So that’s… It’s not exactly exponential, it’s super polynomial, it’s almost exponential, this speed-up. It’s a very spectacular speed-up. You can solve factoring problems if you have a quantum computer with a clock speed that seems physically reasonable, you can solve it in hours, and even if it would take the age of the universe to solve it with the best classical computer, so that’s a real speed-up and that’s the kind of thing that gets our pulses quickening.
1:11:35.6 JP: Now, as far as these applications to material science and chemistry and so on, again, I would say the best argument we have that those problems are really hard classically is nobody can figure out how to solve them. Again, it’s not for lack of trying. Physicists and chemists have been trying for decades to come up with better algorithms that can run on classical computers for simulating the behavior of quantum systems with many particles, but still the best algorithms that we have have a runtime which rises exponentially with the number of qubits. Quantum computers will be able to solve those problems efficiently. Problems like simulating the dynamics of a quantum system, that’s evolving with time and… Well, with some caveats, they’d be able to solve problems like finding the low energy states, finding the ground state of a molecule and things like that.
1:12:46.6 SC: So we’re thinking of things like chemistry, materials, superconductivity, stuff like that?
1:12:52.4 JP: Yeah, so those are the applications where I feel pretty confident the problems really are classically hard, and quantum computers will be able to solve them efficiently, but efficiently is… Computer scientists mean something technical by that… By those words, they mean that the time it takes to solve the problem scales in a reasonable way with the size of the problem, like the number of particles in the material, or the number of electrons in the molecule or something like that. That doesn’t necessarily mean it’s really practical, and so we’re going to have to scale up quantum computers quite a bit from where they are now, before we can do simulations of materials that are too hard to do with our classical computers and might teach us something interesting.
1:13:55.1 JP: I mean, one of the Holy Grails in computational quantum materials for decades has been trying to understand high temperature superconductors better and try to point us in the direction of discovering new high temperature superconductors, potentially of great practical interest, if we can get them to room temperature in particular. And that’s a really hard computational problem, and already you can consider model systems with a number of particles which is of order 100 or so, where that’s too hard to solve with a classical computer. You should be able to solve with a quantum computer.
1:14:44.4 JP: But then the real cost of running that quantum computation comes back to this issue that the quantum computers aren’t perfect, and we’ll probably have to use quantum error correction to run those algorithms, and how many physical qubits we need, then, is going to depend on how reliable the hardware is, what the error rates are. And right now the error rates are pretty bad, so it would probably mean millions of physical qubit. And what do we have now, maybe 100 physical qubits, not quite that. So it’s a big, big gap to cross from where we are now to where we’ll be confident quantum computers can solve problems that material scientists are interested in, which seemed to be too hard to solve classically. I hope it will happen soon.
1:15:39.3 SC: And these problems you mention, these kinds of problems, it’s easy to see why they’re a natural fit for quantum computers because you’re using entangled qubits to simulate literally entangled materials. Does that mean that we shouldn’t necessarily hold out any hope for quantum computers being especially good for simulating other complex but classical systems like economies or societies or technologies or anything like that?
1:16:05.3 JP: Well, I wouldn’t put it as strongly as we shouldn’t hold out hope, but I would say I don’t see any very persuasive argument that quantum computers will be good at problems like that.
1:16:18.0 SC: Good.
1:16:21.6 JP: Yeah, I think as best we currently understand, like I said, we don’t expect quantum computers to dramatically speed up everything, and so the class of problems for which very substantial speed-ups are possible is something that’s probably limited and exactly what are those problems, what are the things that are too hard to solve classically that we can solve quantumly, I don’t think we really know so much about that yet. Our understanding of it has advanced some over the last 25 years or so, but still, like I said, the best idea we have… I keep coming back to it, is simulating quantum systems with quantum computers, that they have these applications to cryptology came as kind of a surprise, and there could well be other surprises. There probably will be.
1:17:19.6 SC: And for the young people in the audience, just so they know, there’s no proof that there isn’t a genius algorithm that uses quantum entanglement to help us solve these big simulation problems, right?
1:17:33.0 JP: Well, it’s really, really hard to show from first principles that a problem is hard. In fact, we’re very bad at that. There’s the whole field of complexity theory, the study of hardness of problems, which has many beautiful results, but it’s largely founded on unproven conjectures, the most fundamental and famous of which is the P not equals NP conjecture that there… I alluded to it in passing before, that there are problems for which we can verify the solution easily once we find the solution, but it’s really hard to find the solution. We don’t know for sure that you can’t solve the traveling salesman problem lickety-split if you find the right algorithm, but most people who have thought seriously about it think that’s really a hard problem.
1:18:31.5 JP: On the other hand, quantum computers can’t solve that efficiently either, so we really have to… It’s quite subtle to find this sweet spot where an important part of the topic is understanding what’s hard classically and define these problems that we can solve efficiently quantum and we can’t solve efficiently classically. Theorists have made only limited progress on that, and once we have the quantum computers, I’m sure we’ll learn a lot more just by experimenting with it.
1:19:08.9 SC: Let’s bring it home, maybe, by going back to field theory and quantum gravity and black holes that you were thinking about before you made this phase transition. And it turns out that you’re still there, like you have not completely abandoned this. These ideas from quantum information theory and quantum computing turn out to be relevant for things like the black hole information problem, is that right?
1:19:31.8 JP: Yeah, and in a way, I sort of came back to it, that’s… I would say there are two big surprises for me from the perspective of the way I thought about the field when I got started 25 years ago. I actually believed early on that quantum information was going to be important for fundamental questions in physics, and it has had an impact on understanding phase transitions and quantum matter, and I thought it would be helpful for understanding black holes and quantum gravity as well, which like I said earlier, is kinda how I got into it. I thought I should study quantum information, ’cause that’ll help me to understand how black holes behave.
1:20:22.5 JP: But there was kind of a transition that I didn’t expect from the people who do quantum gravity for a living, many brilliant people who are friends of both of us, were not too interested in quantum information, and then suddenly they were, and suddenly you could go to a quantum gravity meeting and the talks would be about quantum error correction and computational complexity and entanglement. And I found that quite exciting. And our current perspective is that space itself or spacetime is not truly fundamental, it’s sort of an emergent property from something deeper. And what is the deeper thing?
1:21:18.1 JP: Well, when I talk about space, I mean geometry. When one thing is close to another thing, that’s a property of geometry, where things are relative to one another, that’s geometry. And the underlying explanation for geometry we increasingly believe is quantum entanglement. So you and I, as it happens, we’re on Zoom, so we’re not in the same room, but if we were, we could imagine trying to break the entanglement, we don’t have to be in the same room to imagine this, try to break the entanglement between where I am in space and where you are in space. And we think that if you did that, you would actually break space, that we would be disconnected from one another, and there wouldn’t be any way to travel from where you are to where I am if there weren’t for that entanglement. So in that sense, quantum entanglement seems to be what’s gluing space together.
1:22:24.3 SC: And just to be super clear, ’cause I always get people confused by this, it’s not the entanglement between you and me that gives us the distance, it’s the entanglement between invisible degrees of freedom in the space between us, which is a little bit hard to visualize.
1:22:40.6 JP: Well, part of the problem is we don’t understand it as deeply as we would like to, but we have lots of hints that that’s what’s going on. And these ideas, which emerged largely through the study of quantum computing, like quantum error correction, turn out to be just what we need to understand this at a deeper level. It kinda makes sense that if space is going to be an emergent property, it better be pretty robust, so you shouldn’t be able to tickle it and make it fall apart. And it’s kind of a quantum error correction code that has this property that it’s not so easy to break space up. Once you have quantum error correction, it becomes harder for the environment to attack a quantum computer and make it fail.
1:23:35.7 SC: I’m literally trying to understand this exact feature of emergent spacetime better myself right now for research, writing papers purposes. So we’re at the end of the podcast. While I have you here, can we be a little bit more specific, can you be, about… You say robustness of space, that makes perfect sense, but what is the relationship? And an explicitly, as we can be, between a quantum error correcting code, which seems like something you build into a computer, and the nature of space itself?
1:24:08.9 JP: Well, here’s something that’s fairly concrete, but it requires a bit of a detour…
1:24:20.6 SC: If you’re up for it, I am.
1:24:22.8 JP: Holographic duality, and so there’s a particular setting in which we understand quantum gravity best at present, and it gives us a handle, a tool for understanding properties of black holes that form and evaporate and so on, and it is a duality between a ordinary quantum theory that doesn’t involve gravity at all, but which actually is an alternative way of describing gravity in which black holes form and things fall and so on. There’s a complicated dictionary that relates those two descriptions, there’s one with just ordinary quantum mechanics and the other one that involves gravity, and that dictionary is actually the encoding map of a quantum error correction code.
1:25:18.5 JP: So what does that mean, for example? We can think of the dictionary in the following way, that there’s some geometry, it’s dynamical, black holes are forming and they’re moving around and gravitationally attracting one another and so on, but this alternative description that doesn’t involve gravity lives at the boundary of this spacetime, and now you could… There’s all this wonderful stuff happening inside, and now you’re going to start messing around with the boundary, there’s some math that takes the physics on the boundary and maps it to the physics deep inside the spacetime.
1:26:02.2 JP: And now you state, like you start removing little pieces of the geometry or hitting them with a hammer, or… Sorry, removing little pieces of the boundary.
1:26:11.6 SC: Of the boundary, right.
1:26:13.1 JP: Take away some of the qubits from the boundary or flip them or do something awful to them, but not too many of them. But what’s happening deep in the spacetime is still okay, and that’s because this map from the boundary to what we call the bulk, where the interesting gravity is, it’s very non-local. So things that look local in the bulk, like a black hole right here in my room, on the boundary that actually involves lots of degrees of freedom, which are entangled with one another, and you can take away some of those degrees of freedom, but you don’t spoil that entanglement for the most part, and what’s going on with the black hole doesn’t get altered. So that’s an example of what I mean by quantum error correction.
1:27:05.8 SC: It’s a great example, and I think that… I’m tempted to ask, so I’ll just ask it, why not? In that explanation you just gave, which is beautifully clear, the role of holography is absolutely central, and in the sense, we did, by the way, have Netta Engelhardt on the podcast for those who want a longer disquisition about that, and Lenny Susskind before her. That does live in this context of what we call the AdS/CFT correspondence, where we have a spacetime where there is a boundary that we can identify and it’s a spacetime itself, etcetera, etcetera.
1:27:41.0 SC: And as you know as well as I do, our world does not appear to be anti-de Sitter space, does that mean that… Is the holography sort of a way to answer the questions in a very simple context, but we’re going to have to do more work to apply it to the real world, or is there a more direct connection there?
1:28:00.8 JP: It’s a toy model, it’s not the real thing, just as you said, it’s not our universe, it’s something that’s easier to understand than our universe, so we understand it better. And eventually we need to throw away that crutch and… Well, I don’t know if you said these words, but what I was describing applies in what we call anti-de Sitter space, it’s negatively curved space, and that doesn’t seem to be the one we’re living in. It turns out that it’s easier to understand because of this nice relationship between the quantum system living on the boundary and the physics going on in the bulk in anti-de Sitter space.
1:28:42.0 JP: I think the lessons are broader, that the connection between entanglement and quantum error correction and geometry will extend beyond that toy example. But we’ve got a lot of work to do to elucidate that. Now, you had Netta on your podcast, and she told you about some of these spectacular recent developments that she’s been involved in, which actually teaches things about the case that’s more like the real world, understanding how black holes evaporate and what happens to the information in them, and what she and collaborators found, I think also has a width of quantum error correction, ’cause they said things like, what’s going on outside the black hole is related to what’s going on inside the black hole. And I think that actually involves one of these quantum error correcting maps as well, but we need to flesh that out.
1:29:55.5 SC: It sounds like it’s a pretty exciting time when we can bounce back and forth between black holes evaporating and the nature of spacetime and little circuits, superconducting that will help us solve the traveling salesman problem.
1:30:08.4 JP: I’m having a blast, yeah, and so are you, so are you.
1:30:09.9 SC: I am, and I hope that everyone listening is also and they catch some of the spirit that is very much alive here. So John Preskill, thanks so much for being on the Mindscape podcast.
1:30:19.9 JP: It was a pleasure, Sean.
[music][/accordion-item][/accordion]
Wow – this is a wonderful discussion! Thank you so much. A question that is undoubtedly very naive – would superdeterminism, if so, make quantum computers impossible, or be irrelevant to their function? A related question would be – why doesn’t many worlds make it impossible to get a useful result from a quantum computer – or do you just get the result that is correct for your world? Sorry, but I am a biologist.
Pingback: Sean Carroll's Mindscape Podcast: John Preskill on Quantum Computers and What They’re Good For | 3 Quarks Daily
Massive calculations to identify amd model the specific operative conditions that encourage molecules to interact to join with other molecules to form other ‘compound molecules’ has applications for nearly every field of science. Innovations that result will translate directly to commerical use and to accelerate the creation of processes and products that can improve humankind. Obviously Chemistry, Chemical Engineering, Biochemistry, Physics, Materials Science, Energy (creation, capture and storage), Biology, Nutrition, and Agriculture to name a few will benfit. If that is all Quantum computing ever accomplishes than it will have made magnificent impact.
When raw calculation capabilities are applied to true AI, true Machine LEARNING, and modeling and emulating complex systems then we take even further strides.
Imagine harnessing Quantum Computing to study an indvidual’s brain chemistry and its influence on their mental health and behavior. Image when we know with much greater certainty the effects of hormones snd chemical imbalances on depression, anxiety, impluse control, ADD, and OCD. These are very real conditions that affect humanity. Imagine tuned ‘vitamins’ and medicines that can improve or even eliminate these brain conditions.
Massive calculations are needed everywhere. More bright mathemeticians, algorithim creators, data science and data aggregators, and chemists and chemical engineers, neuroscientists, and physicists are needed now.
Humans have created amazing machines and medicines already. Imagine a world where mere massive calculations can take place constantly and where everyone who can write an algorithm can test their ideas. The Quantum era changes and improves life on earth and anywhere else we happen to live.
Its not science fiction take wjll take place a bundreds years from now. True Quantum Computers are coming to a PC or device near you in the next few years. You can connect to the Quantum computers via the Internet to the Cloud server farms across the world. Every student, college student, graduate student and scientists anywhere can learn to tske advantage of the CPU and GPU in their hands connected to the Quantum Computers located in server farms.
Thanks IBM, Amazon, Microsoft, Google, Oracle, Rackspace for the public clouds. Add Honeywell and startups to these same Cloud companies who are busy creating Quantum Compuer science and hardware itself. Thanks Intel, AMD, Nvidea, ARM, for the CPUs and GPUs. Nividea is also building the servers. Thanks to the process tool makers that make it possible to manufacture the CPUs and GPUs.Thank to the Quantum researchers designing and creating Quantum Computers. Thanks Qualcomm and Samsung and Apple for building the systems we hold in our hand that are tbe gateways to the the Quantum Computers in the Cloud.Thank you to educators st every level who teach us. guide us, inspire us.
The future is in good hands.
wow, very informative conversation.
Rob Horsch,
Superdeterminism is the idea that you can reproduce quantum mechanics using classical physics. So, superdeterminism does not make the quantum computers impossible or useless.
It is a relief that there was no conjecture that quantum computers might ultimately solve the hard problem of consciousness. Maybe it is because Roger Penrose’s position that whatever consciousness is, it is not created by computation (paraphrase). In his book, “The Emperor’s New Mind”, he challenges the idea that artificial intelligence is a valid concept. I agree with that assessment, and I take from his reasoning that Intelligence does not emerge from the manipulation of numbers. Intelligence requires the understanding of meaning. Computers are not capable of understanding anything or the meaning of anything. In fact, A.I. is a total misnomer. It most likely was a strategy to raise investment funds to research the very valuable capabilities of computer science. The term should be I.A. for intelligence augmentation. Computers, be they classical or quantum, are tools, and like all tools they serve an important purpose, but they are not intelligent.
If quantum computers become a reality somewhere in the next ten years and they do some of the things they claim, might they find that it is not for the public to know about. Would this be kept in the hands of big tech and Governments.
Fantastic interview (as usual).
As a computer scientist, I hoped to hear more about the potential for breakthrough AI algorithms. Professor Preskill did make passing mention of it, but there was a pregnant moment where I waited for a discussion of quantum mind (or quantum consciousness) and how that might apply to quantum computing. If this technology proves to be a superior simulacrum of real-world quantum processes (a point that was made multiple times during the podcast about chemistry, materials science, etc.)–and if the human brain is ultimately a quantum process–then perhaps this is the key to true machine consciousness? Not just simulated intelligence–a concept stated above in a comment by Mr. Wade, which I whole-heartedly agree with.
The “100 page book” analogy was particularly evocative. I couldn’t help but think of the holographic nature of the human brain, and that all of the pieces to the puzzle of “true” machine AI are staring us in the face…
Also, the quantum error rate that exists in today’s hardware reminds me a lot of our own ability to produce accurate results when we use our minds to calculate or remember something… ;-0
AH