John Preskill

Notice: We are in the process of migrating Oral History Interview metadata to this new version of our website.

During this migration, the following fields associated with interviews may be incomplete: Institutions, Additional Persons, and Subjects. Our Browse Subjects feature is also affected by this migration.

We encourage researchers to utilize the full-text search on this page to navigate our oral histories or to use our catalog to locate oral history interviews by keyword.

Please contact [email protected] with any feedback.

ORAL HISTORIES

Credit: Lance Hayashida - Caltech

Interviewed by

David Zierler

Interview dates

February 26 and March 12, 2021

Location

Video conference

See catalog record for this interview.

Usage Information and Disclaimer

Disclaimer text

This transcript may not be quoted, reproduced or redistributed in whole or in part by any means except with the written permission of the American Institute of Physics.

This transcript is based on a tape-recorded interview deposited at the Center for History of Physics of the American Institute of Physics. The AIP's interviews have generally been transcribed from tape, edited by the interviewer for clarity, and then further edited by the interviewee. If this interview is important to you, you should consult earlier versions of the transcript or listen to the original tape. For many interviews, the AIP retains substantial files with further information about the interviewee and the interview itself. Please contact us for information about accessing these materials.

Please bear in mind that: 1) This material is a transcript of the spoken word rather than a literary product; 2) An interview must be read with the awareness that different people's memories about an event will often differ, and that memories can change with time for many reasons including subsequent experiences, interactions with others, and one's feelings about an event. Disclaimer: This transcript was scanned from a typescript, introducing occasional spelling errors. The original typescript is available.

Preferred citation

In footnotes or endnotes please cite AIP interviews like this:

Interview of John Preskill by David Zierler on February 26 and March 12, 2021,
Niels Bohr Library & Archives, American Institute of Physics,
College Park, MD USA,
www.aip.org/history-programs/niels-bohr-library/oral-histories/47147

For multiple citations, "AIP" is the preferred abbreviation for the location.

Well, of course, I started out my career doing particle physics and with occasional forays into cosmology. So the backstory of IQIM is, I made a mid-career shift in research interests in the mid-1990s. And that happened around 1994, when I learned about Peter Shor's factoring algorithm. But I was primed, I think, to get interested in the subject of quantum information and quantum computing for a couple of reasons. One was that just the previous year, the SSC had been canceled. And for my generation of particle physicists, this was really a terrible blow because we had come along a little bit too late to participate in erecting the Standard Model. I started graduate school in 1975, and so, there was still controversy about what the right electroweak model was at that stage. But the J/psi had been discovered the year before, all the great stuff had been done. Not that there wasn't still a lot of interesting particle physics to do.

But our big hope for beyond the Standard Model physics, we were going to be the generation that would unravel the origin of electroweak symmetry breaking and all the new physics associated with it, and the SSC was going to be the source of the rich phenomenology that we thought we were going to mine as theorists to probe more deeply into nature's secrets. And when the SSC was canceled for complicated political reasons, even though they had already been digging the tunnel in Texas and had sunk a couple of billion dollars in it, one realized it was going to be quite a while before we were going to have the experimental input that we needed to really understand what was going on with physics beyond the Standard Model, which it was generally believed would be discovered when we got up to those energies. And so, I was sort of in a mood to think about different things.

And in fact, while I was sort of waiting, as many of us were, for the SSC to come along, I had already been doing things which were not very phenomenological, like thinking about black holes and how they process information. So, I had sort of become acquainted with principles of quantum information—which were not so widely known by physicists except for a small community—just because I thought that might be useful for understanding what's going on with black holes. And when Shor discovered this factoring algorithm, about a month later Artur Ekert, who was a pioneer of quantum cryptography, visited Caltech and gave a talk. And he mentioned this recent breakthrough that Shor had discovered that you could efficiently factor with quantum computers. And it's possible I've embellished the memory in hindsight, but I was quite amazed by this.

Feynman had been interested in quantum computing, and I knew that. And I wasn't very impressed by the whole subject. I didn't quite see the point of it. But I realized with the discovery of Shor's algorithm that it really meant there was a big difference between what problems we'd be able to solve and which ones we'd never be able to solve with computing technology because it's a quantum world instead of a classical one. Things became possible thanks to quantum mechanics that just wouldn't be possible in a different type of physical world. And I still think that's one of the most amazing things we've ever learned about quantum physics, the difference between quantum and classical. Sorry, I'm giving a rather longwinded answer to your question.

Zierler:

No, this is the intellectual origins of the Institute.

Preskill:

Correct. And so, I had a colleague at Caltech, Jeff Kimble, still at Caltech, and he was also quite excited about this surge of interest in quantum computing. He was a quantum optics guy. And he had made experimental advances in squeezing states of light and using them for metrology and stuff. But it's different now. Atomic Molecular Optic physics is widely recognized by most physicists as an intellectually rich and exciting field. And that's happened largely, I think, in the last 25 years, because of the connections with quantum computing and because of the connections with condensed matter, the possibility of simulating interesting states of quantum matter using these AMO systems. But back then, there was sort of a feeling in the quantum optics community that they didn't get respect because, from the perspective or somebody like me, what was the point, you know? As a particle physicist, I was trying to understand new laws of nature. But what can you do with quantum optics in the lab that you couldn't just figure out with pencil and paper what was going to happen?

And quantum computing, at least in principle, kind of changed that. Because I think it drove home that you should be able to do experiments where you're learning something from the experiment that you couldn't just simulate or calculate. And so, Jeff would have to say himself what got him excited, but I think he realized he had experimental tools that were very relevant to exploring quantum information. And I got excited from the theory perspective and wanted to learn more about what was experimentally possible. So, we formed what we called the Quantum Computing Club at the time, and we started having joint group meetings. And so, I learned some things. I've never been deeply knowledgeable, really, about how experiments work, but I learned a lot more than I had known about what you could do with quantum optics tools. And meanwhile, I was trying to understand whether you could protect quantum computers from noise, which led to the development of the idea of quantum error correction.

But actually, we wound up getting a DARPA grant. This was kind of interesting. The Department of Defense agencies had an immediate interest in quantum computing after Shor's algorithm because of the applications to cryptology. And they were the early supporters of the research in the field. Including the development of experimental tools. And DARPA, in particular, put out a call, and we submitted a proposal, and we got funded for a project which we called QUIC. I guess it was Quantum Information and Computation, but QUIC for short. And there were five PIs, and that helped Jeff do his first teleportation experiment in the lab, and I worked on quantum error correction ideas, and stuff like that. It was a five-year award, and after two years, they cut it. There was a new program manager. This was, I learned, not unprecedented for DARPA --- a new program manager comes in, and what you think is a multi-year commitment turns out not to be.

That's what Sean has. And that's what John Schwarz had, actually, when I first came to Caltech. That's another story, speaking of someone who was underappreciated for a while. Yeah. And so, we had Kitaev, and we had this amazing group of young people. And then, a lot of students came through and trained. I think probably in terms of impact on science, leading the IQI and establishing it is the most impactful thing I've done when you look at all the people who came through and how they've become scientific leaders. But anyway, to come around to answering your question, for ten years, we were just the IQI, and we went through several cycles of renewal at NSF. And Jeff spearheaded this. I wouldn't have thought to request a grant to start a theory institute if Jeff Kimble hadn't been pushing me, so I'm grateful to him for that.

And in 2010, we applied for the Physics Frontier Center program at NSF, very competitive thing. There are ten of them in the country in different areas of physics. And that turned out to be successful, and as a result, what had been the IQI expanded to a larger center that did encompass both theory and experiment, pretty much as we had envisioned back when we originally proposed it in 2000. We had something like that in mind, but NSF at that time said, “We only want the theory.” But in 2011 we became the Institute for Quantum Information and Matter. And now, that's been around for almost ten years, since 2011, and has been very successful.

Zierler:

John, that's a great overview of your current titles and affiliations. So before we take it all the way back and develop your personal history, I'd like to ask a very in-the-moment question. As you say, of course, we're all working from our home offices now. As a theoretical physicist, I wonder if in some ways, these past 11 months have been more productive for you because the social and physical isolation perhaps has given you a bit more headspace or bandwidth to work on some equations or problems that you might otherwise not have. On the other hand, I wonder if your style as a scientist really depends on in-person, interpersonal interaction, and in many ways, your research agenda has suffered as a result.

Preskill:

Well, as the question suggests, it's a complicated issue with tradeoffs. One big change for me is I was traveling a lot. And I get, of course, as we all do, lots of invitations, most of which I turn down. But for opportunities to lecture, attend conferences, and things like that, there were a certain number of them which I really thought I had to accept. So the last couple of years, I had been making lots of trips. And it was really a bit of a relief to put a stop to that for a while and not be chasing around so much.

On the other hand, the kinds of interactions you have when you visit other places to attend a conference or give a talk and so on, the kinds of informal interactions, those are not very well simulated in the Zoom era, although there are various attempts to do that. And so, you do miss that kind of thing where you go to dinner or lunch and just chat. And sometimes, that's a good way of probing questions and coming up with ideas. So, I think we've all suffered a bit from missing that kind of interaction.

In my group, it hasn't been too bad. We have our group meetings on Zoom, and I'm able to keep up with what students and post-docs are doing, and so on. But I think it's hard for the new students. They can attend meetings and stuff like that, but it's hard to become sort of integrated into the community in our online existence compared to when we're able to hang around and chat in our offices or at a coffee break. But in response to your question, yeah, I think I have had a bit more time for reflection than was the case, say, in the previous couple of years, and that has been helpful. And it's also given me a little more time, maybe, for reading and catching up on things.

One thing that I had been increasingly feeling was missing from my education or knowledge base was the students are more and more interested in machine learning, and I really just didn't know much about it. And I still know only a limited amount about it. But I did take some time to read textbooks and papers, and I also am collaborating with some students who know a lot more about machine learning than I do. And so, that's been a plus over the last year.

Well, according to family folklore, in Lithuania, it had a similar sound. And I've sometimes pondered whether it's related to names like Peskin and Peshkin. But we think in Lithuania, they were saying it more or less the way I do as Preskill. And there are several alternative spellings that were adopted when people immigrated. So, there are some Preskills still around the Chicago area, but there are also other spellings, like Preaskil. So anyway, my grandfather's business that he started was a harness shop. He would make the rig that you would use to attach your horse to your buggy. But when automobiles came in, he realized that wasn't a good business model, so he opened a hardware store. And when my dad was a kid, he used to work in the hardware store. So even in later life, he considered himself to be an expert on tools.

And my mom has a rather different origin story. She actually converted to Judaism when she married my father, and then later in life, she actually got bat mitzvahed. Much later in life. But she did not grow up Jewish. She grew up in Cleveland. Her father was a lawyer, and his family had been, for many generations, farmers. And my mom's mother also came from a family that had been farming in Pennsylvania and Ohio for many generations. We think they go back to before the American Revolution in Pennsylvania. But when my mom was a kid, she would work in my grandfather's law office. He was a probate lawyer, did wills. And very successful in the sense that he was very highly regarded in his profession. He wrote a textbook on Ohio probate law that was widely used and had some high-level connections.

One of the famous family stories is my mom, as a teenage girl, was working the switchboard in the office, and she cut off the Vice President of the United States by pulling the plug while he was talking to my grandfather. She wound up going to law school, and it was pretty unusual for women to attend law school. She was the only one in her class at what was then called Western Reserve, now Case Western Reserve in Cleveland. And I think she might've gone into practice with my grandfather. But then, World War II came.

And going back to my dad for a minute, he wound up going to the University of Chicago. He was very good student. He graduated high school at 16, and he graduated law school at 20. At the University of Chicago, he was able to get a bachelor's degree and a law degree more or less concurrently. And that was 1932. It was the Depression. Nobody wanted to hang around in school. Everybody had to go out and earn a living if you could get a job. So, he was in a big hurry. Because he was 20, he couldn't take the bar exam because he was still a minor. He had to wait til he was 21. And he passed the bar exam and worked for a law firm for a while. But when the war came, he was 4F because of a medical condition. He was about 30 then, but he couldn't enlist.

So, the way he did public service was he became a federal employee. He moved to Baltimore and worked for what was then called the Federal Security Agency, which was setting the legal foundations for the Social Security system, which was still sort of being fleshed out. And that's where he met my mom. When the war came, she also thought she should work for the government, and they wound up in Baltimore. The reason they were in Baltimore is a lot of the federal agencies moved out of DC because that was being taken over for military purposes. And that's where they met. They were both lawyers in the same office, and they got married 1944 and moved back to Chicago.

Preskill:

That's right. But I realized, by the end of my freshman year, several things. One of them, I think, I'd known all along. I'm just not cut out intellectually to be a mathematician. I'm just not good enough at that kind of thing. Meanwhile, I was taking freshman physics, and in the spring term, we used this book by Purcell, Electricity and Magnetism, which is a great book, and it's still used in some places. And that really impressed me because I was learning in my math class calculus on manifolds, and about differential forms, and things like that. It was all very abstract, and very beautiful and fun. But no hint of what it was good for.

Although, of course, he was a mathematical physicist and proved theorems, when he was young he did more practical things. He worked out details like how ionizing radiation deposits energy in materials and things like that. So, he had some appreciation for that type of knowledge building. Actually, another college teacher who had a big impact on me was John Wheeler. My sophomore year, he taught a class that I took for the whole year, covered everything in physics. We called it Honors Physics. And we did classical mechanics, and E&M, and stat mech, and quantum physics, and waves all in one year. And it was a very idiosyncratic course, to put it mildly.

Of course, to us undergraduates, there was something kind of god-like about Wheeler. So, this was 1972. The thought that he had worked with Niels Bohr seemed unimaginable --- that anyone could be that old. He was 61 at the time. Here, I'm 68, so it doesn't seem so old now, but at the time, it sure did. And he always came to class in a suit and tie, and that also made him seem like a denizen from another generation. And, of course, he had this marvelous ability to use the blackboard to draw intricate illustrations on the spot. But the thing that was most memorable is–here's what he did on the first day of class, or at least how I remember it. We're going to do classical mechanics. We're going to use Goldstein. We're going to learn Lagrangian Mechanics. And we're going to learn Hamiltonian Mechanics from this book. And I'd already dipped into the book a little, I was excited.

And so, I figured he was going to tell us about the calculus of variations, and the Euler-Lagrange Equations, and stuff. I kind of had a hint what that was about. But he comes in, and he goes up to the board, and he draws A on the board and B. And then, he draws a line going from A to B. And he said, “An electron is going to travel from A to B. How does it know how to go? What path should it take? Well, of course, it takes all the paths. It adds them all together with an E to the iS …” “What?” He was trying to explain that what we were learning was the classical limit of quantum theory. Although Goldstein wasn't saying it that way, he thought it was important for us to know right from the start that that was the context, and that you could understand why this calculus of variations stuff was relevant by thinking about how the phase when it's stationary would add up constructively.

Every Princeton student does it. So, it's a big deal. You spend a lot of your senior year doing it. Actually, there were junior papers as well that I think everyone had to do. In physics, we had to do one the first term and second term. And actually, looking back, maybe this was sort of formative as well. So, you're a junior, you don't have any idea what to do for a research project. You're supposed to knock on doors and talk to faculty, see if they have suggestions, say you're interested in working on something. “What do you propose?” And so, I don't know why, I guess maybe I was assigned to him, I went to see Marc Davis who's a cosmologist, he's been at Berkeley for many years now, but he was at Princeton then. And so, he asked me what I was interested in. And what I said was, I was interested in the interpretation of quantum theory. And he said, “Well, you know what --- you might be interested in is the EPR Paradox,” which I had never heard of.

And so, he explained a little about what it was. He didn't really know. But that piqued my curiosity, and it turned out that there was a new instructor who had just arrived at Princeton that year named Stuart Freedman, and he had just done an experiment with John Clauser to test the Bell Inequality. And so, I went to him and asked him to fill me in a little bit about that. And he said something that stuck, which I thought was really weird. He had done the experiment with Clauser, which seemed to confirm violation of the Bell Inequality. But there was a competing experiment that had found a different result, that the Bell Inequality was satisfied, so the idea of local realism seemed to be confirmed by that competing experiment, which was done by a Russian group. And I said, “Well, how do you account for the discrepancy?” And he didn't give a scientific answer, he gave a political one, which was, “Well, it has to do with dialectical materialism. So, there's a bias in favor of local realism.” I thought, “Boy, could that really be it?” Anyway, it just kind of shocked me that he said that.

No, not so much. Who would I have been aware of? Of course, I knew Sid Drell because I had read his textbook.

Zierler:

Bjorken, for example?

Preskill:

Yeah. I don't know. I wasn't too excited about Stanford or SLAC. And at Caltech, the feeling was the glory days were behind. That's what the students were saying in the mid-70s, that Gell-Mann and Feynman were in their declining years. I had one friend who graduated a year ahead of me, good friend, Orlando Alvarez, who had been a Princeton undergrad. And he went to Harvard. I talked to him a lot, and I figured, “Well, if it's good enough for Orlando, that's probably where I should go.” And I was aware of Steve Weinberg. He had been hired at Harvard relatively recently, I think in '73 he had moved from MIT. He was sort of supposed to fill Schwinger's shoes. And I was aware of the Weinberg-Salam Model, so I knew he was supposed to be a big deal.

Well, everybody wanted to work with Sidney because he could explain things so clearly, and he was receptive to a certain degree to supervising students. But in a way which was only half-joking, I rather vividly remember him saying, “I have graduate students like a dog has fleas.” And, of course, he meant it as a joke, and I actually thought it was funny. But he really had a lot of students, and his personal habits were different than they were in later years. This was before he was married and before he'd been diagnosed with adult-onset diabetes. And he was a very heavy smoker and kept unconventional hours. He'd stay up all night and then go to sleep at dawn, and he'd come in in the afternoon. He always insisted that his lectures be scheduled for the afternoon because he would be sleeping in the morning. And when he would arrive in mid-afternoon, sometimes late afternoon, students would be queued up outside his office because they wanted a moment with Sidney. And I just thought, “Who needs it? I've got to stand in line to get a few minutes with my advisor?” So that was part of it.

Well, as I mentioned, what I thought was the most exciting thing in my first few years of graduate school that was happening was these topological ideas coming into particle physics, particularly in the theory of magnetic monopoles and instantons. But topological ideas were also becoming increasingly useful in condensed matter, where in different materials, there can be topological defects associated with spontaneous breaking of symmetries. Not usually gauge symmetries in the case of the magnetic monopoles. Well, actually, in the case of a superconductor, a vortex in a superconductor is an example of a topological defect in a gauge theory. That is sort of a prototypical example of such a topological defect. Bert knew everything about superconductors. He knew all about vortices. But also, there were point-like defects that occurred in three-dimensional materials like liquid crystals. And Bert knew about that, too.

Furthermore, what I was interested in is what would happen if there was a phase transition in the early universe. If it's SU5, for example—which is the gauge group—it had been understood in the previous few years that at high temperature, even if the gauge symmetry is spontaneously broken—if the Higgs phenomenon occurs at low temperature, at high temperature that symmetry would be restored. So, you would expect very early in the universe that the SU5 symmetry was still intact, but as the universe cooled, the symmetry breaking would occur. There might be a sequence of phase transitions, but there should at least at some point be a transition to the phase in which SU3 cross SU2 cross U1, the symmetry of the Standard Model, is the unbroken remaining gauge symmetry. And one could show that the breakdown of SU5 to the Standard Model would give rise to stable magnetic monopoles. The question was, how abundantly would they be created?

And there were a couple of ways of looking at that, one of which was really the idea that Kibble had discussed, although I didn't know his paper at the time, which is that there's an order parameter, which is fluctuating around, you're in the symmetric phase, but then it freezes out. Like, for example, if you're cooling a material, and it goes from paramagnetic phase to a ferromagnetic phase, the magnetization locks in, and all the spins line up. And that is the same kind of phenomenon where the symmetry is restored at high temperature and then becomes spontaneously broken at a critical point.

And I thought the same thing would happen in a unified gauge theory. And that that would give rise to the possibility of magnetic monopoles, for one thing, just because of relativistic causality. When the magnetization turns on, the magnetization at one point in space has no way of knowing to line up with the magnetization of another point in space because there hasn't been time for a light signal to travel between this domain and that domain. And so, as the spins start to line up, there will be knots that get locked in. Those are the topological defects. And that's how magnetic monopoles can form.

I had that idea, but Bert told me a different idea, which was that even if the phase transition were a smooth phase transition, if it were second order, that because the order parameters would be fluctuating, you would expect to get a lot of defects, even not taking into account the effects of relativistic causality, and that's what he showed me how to calculate. Which, because I wound up writing a very short letter-length article, got squeezed down to, like, a paragraph or something without many details. That was another thing. Steve did give me one piece of advice. I asked him where I should publish the result that he didn't find very interesting, and he made a suggestion which really surprised me. He said Nature. The particle physics students didn't read Nature. That was where there were papers about astronomy and stuff like that, or biology. But Steve at least had the notion that there was something of broad interest about what I was doing because it related to cosmology, to particle physics, and even to condensed matter ideas.

Zierler:

It was of broad interest, but not particularly interesting to him.

Preskill:

Not interesting enough, but broad. [laugh] But Bert said, “No, Physical Review Letters would be better.” So that's where I submitted it, and it was rejected because first of all, it wasn't novel enough. It'd actually be interesting for me to dig up that referee report. I think I still have it. But also, that it didn't seem to be right because I had overestimated the abundance of the monopoles for some reason which the reviewer didn't explain. But the editors, to their credit, said, “Well, we'll give you a chance to respond and resubmit.” But it was a very bad day because I was already in my fourth year of graduate school. I had no papers. And this was my first one. And it got rejected. And I was pretty depressed. So, I remember my wife thought this would cheer me up—we went shopping, and we bought a color TV set. Up until then, our TV set was this little black and white TV set. And it did cheer me up a little to have a color TV. But anyway, I resubmitted the single-author paper, and it did get accepted.

I knew a lot about magnetic monopoles. Yeah. But anyway, I talked to Alan a lot. I don't remember, was their paper out yet? Not sure. But they had some of the same ideas, and I guess you've already talked to Alan. But I'd been pretty careful in analyzing if a significant number of monopoles were produced, how many of them would survive. And I had an argument, which I thought was pretty convincing, that unless something nonstandard happened in the cosmology, it just couldn't work, that the production of the monopoles was unavoidable. It would be copious, they wouldn't annihilate fast enough, and the universe would be closed by them many times over, and that couldn't be our cosmology, so there had to be some way out.

And during the following fall, I probably should've thought about that more. Because I figured there had to be something about the phase transition that was unconventional. But by that time, I had gotten interested in a different topic, which is what I did end up writing my thesis on, which was technicolor, as we called it at the time. The breaking of electroweak symmetry by strong interaction. So, of course, Alan famously continued thinking about it and had the insight that inflation could blow the monopoles away, but he also, to his credit, realized that that could explain the flatness and isotropy of the universe. And, of course, that idea was very explosive when it came out. That paper had a lot of impact right away. I remember him coming to Harvard and giving a talk about it, which was received with a lot of excitement.

Zierler:

Did you see the transition to technicolor as switching gears? Was it related?

Preskill:

It wasn't that closely related, but I thought the ideas were quite exciting. I was particularly inspired by a paper by Lenny Susskind, which is actually a little ironic because Steve Weinberg had written a related paper. He didn't call it technicolor, but he did call it dynamical breaking of electroweak symmetry, which is what it is. And his paper was, in a way, sort of typical Weinberg style. He calculated everything, and he correctly discussed all the issues. It was a little dry.

And Lenny Susskind is also one of my heroes because of his creativity as a scientist, but also he's a very charismatic communicator in writing and in person. And this was an inspiring paper. And what he realized, which Weinberg had not, was something quite simple, which was, in the Weinberg-Salam Model, the so-called rho parameter, which basically says that the ratio of the W to Z mass is determined by the Weinberg mixing angle theta-W. Steve, by the way, always claimed the W in theta-W stood for weak, and he wouldn't call it the Weinberg angle. He called it the weak mixing angle, in a burst of modesty.

But at any rate, Murray Gell-Mann always liked to say in his snide way, “Oh, we call this angle theta-W because W stands for the last letter in the word Glashow.” [laugh] Anyway, Murray and Steve were not fans of one another. So what Lenny said in his paper was that you could understand how the Weinberg angle was related to the W of Z masses just from a symmetry consideration, and that in the dynamical symmetry breaking scenario, that symmetry would naturally be present, and that the dynamics that you needed was dynamics that we already understood fairly well from QCD—the breaking of chiral symmetry in QCD, which is responsible for the pion being much lighter than other hadrons. That could occur with this new strong interaction with a similar structure to QCD, but which becomes strong at a higher scale, at the weak scale like a TeV, or a few hundred GeV rather than a few hundred MeV, as in QCD. That could account for how the electroweak symmetries get broken. And Lenny called this new strong interaction “technicolor.”

And what I found so appealing about this was that because it was dynamical, it should be highly constrained. One thing I found very curious and was very interested in, for some years, starting when I was in graduate school is, where do the quark and lepton masses come from? In the Standard Model, they're just free parameters. You write down Yukawa couplings, they can be anything, and those determine the mixing angles like the Cabibbo angle and the Kobayashi-Maskawa matrix, it's all free parameters. Same thing for all the masses. And what fun is that? You'd like to be able to explain where those masses come from. And I thought in a dynamical scenario, we'd be able to do that much better.

I was very interested in those ideas for a couple of years, and it turned out that these dynamical scenarios are so constrained that it was very hard to come up with a viable phenomenology. Because you didn't have the same kind of freedom you do when you have Higgs fields, where you can choose Yukawa couplings to be whatever you want. Explaining the masses of the quarks, and leptons, and all that became very challenging. Actually, I'll tell you something funny. When I first came to Caltech, that was in 1983, I thought the important problem in particle physics was to explain those quark and lepton masses. I thought, “If we could do that, if we could understand what that hint was telling us, that would be a good path to understanding what's beyond the Standard Model.”

And so, to remind myself that was important, I made a chart which showed all the masses, the spectrum of quarks and leptons, and I posted it in my office on the bulletin board so I'd see it every day to remind me, “This is the important thing to think about.” And then, a couple of years went by, and one day, I was talking to Mark Wise in my office, who occupied the office next door, and we looked at that chart, which had been on the bulletin board, roasting in the sun every afternoon, and the masses had all faded away. They'd been bleached by the sun. And we took that to be some kind of metaphor for how this problem somehow was too elusive to admit an easy solution. And by that time, I wasn't thinking about it anymore.

Zierler:

I want to ask, at this point, when you're really starting to solidify your identity in theoretical physics, going from magnetic monopoles to technicolor, did you feel at the time that you dipped a toe into cosmology, and then went sort of back to your home intellectual environment of particle theory?

Preskill:

Well, yeah. I don't know if I looked at it that way. But because I got excited about technicolor, I sort of dropped the cosmology ball for a while and focused on technicolor. My interest in cosmology got reactivated partly because of another experimental false alarm. In 1982, Blas Cabrera thought he saw a magnetic monopole. It was on Valentine's Day, 1982. I was still at Harvard. By then, I was on the faculty. And that seemed incredible and really exciting. And hard to explain. He had this little loop of superconducting wire and saw the flux jump, which he interpreted as evidence that a magnetic monopole had passed through the loop. And so, one needed to understand why magnetic monopoles would be plentiful enough for Blas Cabrera to detect one, and at the same time, not do other things, which the astrophysicists told us would be bad. Parker, in particular, had gotten a bound on the abundance of monopoles from observing that if there's a magnetic monopole plasma in the galaxy, it'll short out the galactic magnetic field on some time scale short compared to the galactic rotation time, which cranks up the dynamo.

And so, was there something wrong with that argument? Guess that wasn't really cosmology. But at any rate, we did realize that if the monopoles were very heavy, the story was changed because Parker had assumed they got relativistic velocities, which for the types of monopoles predicted by grand unified theories, needn't be the case. They'd more likely have typical virial velocities in the galaxy like 10 to the minus 3 c. Of course, it turned out Blas Cabrera never saw another magnetic monopole. But it was exciting for a while. And actually, that helped to elevate my star a little bit maybe because now everybody was excited about magnetic monopoles and where they came from. And I was asked to give talks about that and things like that.

Zierler:

To clarify, when you say that Cabrera never saw another one, is that to suggest it's possible that what he saw was a magnetic monopole?

Preskill:

Well, it seems extremely unlikely, right? Because he would've had to be incredibly lucky to see that one and be consistent with other bounds we have on the flux. So no, I don't think he ever explained, or at least never publicly explained, what went wrong or what the right interpretation was of the event he saw. But no, it wasn't a magnetic monopole, sad to say.

Yeah, so then, the next foray into cosmology which had some impact concerned axions and predicting that they could potentially be the dark matter. And probably Alan Guth told you about this workshop, the Nuffield Workshop in 1982 in Cambridge. I was there. It was organized by Stephen Hawking and Gary Gibbons. It was a pretty exciting event. And the big topic there was whether inflation could explain the origin of galaxies by seeding the density perturbations from which galaxies grew. And there was a lot of disagreement at the beginning of that three-week workshop about what inflation predicted.

And I'm sure you discussed this with Alan, but after the idea of inflation, which seemed very exciting, trouble was brewing because how inflation ended was unclear. And Alan and others had done computations of how as bubbles of true vacuum appeared in the false vacuum in a phase transition, whether those bubbles would succeed in filling up the universe and giving rise to a reheated universe that would then be described by Big Bang cosmology, and he couldn't get this to work. But then, around I guess it must've been the end of 1981, the idea by Andrei Linde, and Albrecht and Steinhardt, that instead of having to go through a barrier, the universe could sort of roll off the table to end inflation. The energy density would be high because you'd be on a plateau of a potential function, but rolling along, and then you'd start to oscillate in the potential after you roll off this flat part. And that would give rise to reheating.

So, what everybody was interested in was what kind of perturbations of density would be produced in that transition from the inflationary phase to the more standard radiation-dominated phase. And so, Alan, and Starobinsky, and Turner, Steinhardt, and Bardeen, and Hawking, they were all trying to calculate those things. So that was sort of the focal point of excitement. But I went there to talk about magnetic monopoles and to think about what axions might have to do with cosmology. And Frank Wilczek was there, too, who had an interest in axions, as the founder of them—

Actually, just to backpedal for a second, this is sort of a funny story. Or maybe, I don't know, sort of a typical experience of a graduate student. In the fall of 1977, I crank up my courage, and I go to see Steve Weinberg. I'd like suggestions for a research problem to work on. And so, he responded immediately with the thing that he was thinking about that day. What was it? Well, he had just read this paper by Peccei and Quinn that would explain potentially the solution to the strong CP problem, why CP is a very good symmetry of the strong interactions. And their idea has something to do with the Higgs sector, and how you can introduce another Higgs field, and that can help. “So what might be interesting to work out is, what's the phenomenology of this type of model with more than one Higgs field?”

Weinberg, Coleman, Georgi, and Estia Eichten, who was junior faculty at Harvard at that time. I do have a very disturbing memory of my exam, actually. You're not going to believe it when I tell you this, probably. Well, here's the thing. I didn't understand what a PhD defense was. Somehow, I didn't realize I was expected to give a presentation. How could I have not known this? All the other students seemed to know it. So, I thought, “Well, they've all read my thesis, and they'll come in, and they'll ask me questions about it.” I had nothing prepared. My thesis was related to technicolor. Didn't have anything to do with cosmology and magnetic monopoles. But actually, it was something Steve was very interested in.

I thought it was still the case, and I thought a transformation was needed. And in fact, they made multiple offers. They tried to hire Michael Peskin and Lenny Susskind. And I think Larry Yaffe, who had been a post-doc at Caltech, in fact, and then wound up going to the University of Washington. So, they realized they needed to do some rebuilding. And if Mark weren't going, I would've been more hesitant about it. Now, the other person who was active in the particle theory group was John Schwarz. At that time, he was not on the professorial faculty. And, really, at the time, string theory was not so respected in most of the particle physics community. And John was plugging away in his work with Michael Green and making rather brilliant progress in string theory, and identifying type 1 and type 2A and type 2B, and so on.

But it was very underappreciated. And that changed overnight in 1984 when they discovered the anomaly cancelation. And suddenly, Schwarz was a hero. He became a Caltech professor, I think, not long after. But yeah, so there was John Schwarz doing his string theory, which was sort of seen as outside the mainstream, Feynman and Gell-Mann, not so active anymore, David Politzer, still active, but maybe not on as steeply upward a trajectory as he had seemed to be a few years earlier. Steve Frautschi, Fred Zachariasen, also not very active. So, it was a group in need of some fresh blood.

Zierler:

Were your interactions with Feynman intellectually, in some ways, the starting point of your budding interest in quantum information?

Preskill:

I'd have to say no. I did have interactions with him having to do with QCD, which I think we both enjoyed a lot. We were both interested in quark confinement, and chiral symmetry breaking, non-perturbative QCD. We talked about those things a lot. And I actually knew a lot of things about that. Feynman would often give the advice to students, “Work things out for yourself. Don't necessarily pay attention to what other people are saying.” Not such good advice, in my opinion, for most students. That was not my style. I tried to read everything. Maybe I overdid it. But in particular, I knew most of what was going on at that time, people trying to understand quark confinement.

And among my heroes were Sasha Polyakov and Gerard 't Hooft, and I knew the things that they had done. Which was actually related to what Feynman was trying to do, but he hadn't studied their work, and because I had, he appreciated that I had something useful to tell him. I did not disguise the fact that I was mostly reporting what I had learned from these other papers. But he didn't seem to pick up on that. He didn't really seem to care whether it came from 't Hooft and Polyakov. He was hearing it from me. And so, he developed what I interpreted as admiration for me, and he would often want to discuss those things. But while I knew he was interested in quantum computing, I was not so interested. At one point in the mid-80s, I was made aware, I think it came from Charlie Bennett, that David Deutsch had written an insightful paper about quantum computing, which I read. And I got the incorrect impression that quantum computing was really not much different from randomized computing, from rolling dice in a computation, and was just sort of that dressed up in a fancy style, and that it wasn't really more powerful in a meaningful way than ordinary computing. And, of course, that was wrong.

This is something that's actually puzzling to me because I did a little dipping into Feynman's papers for his 100th birthday, because I was asked to give a talk at an APS meeting about Feynman's scientific legacy. His paper based on his 1981 talk at a conference on the physics of computing called Simulating Physics with Computers, was really, very insightful, and he has the idea and expresses it fairly clearly and correctly that when it comes to simulating complex, many-particle quantum phenomena, ordinary computers just won't be able to do it efficiently, but with a quantum computer, you could. Now, this is a truly great idea, and 40 years later almost, it's still a very current and relevant idea. And then, he wrote one more paper, which was also rather insightful, about quantum computing in 1985.

But when I said I dipped into his papers, what I mean is, he taught a class jointly several times with Carver Mead and John Hopfield about the limitations and potentialities of computing machines, in which he talked about quantum computing. And in those lectures, in the notes and in the recollections of the students and even TAs, he did not, in class, ever say anything about quantum computing being more powerful than ordinary classical digital computing. That great idea from 1981 didn't make it into the class. The class did include content from the 1985 paper, and I think the first time he taught it was before that paper appeared. He seemed really interested in, was there a thermodynamic cost to computing? Charlie Bennett had written a famous paper about that, saying in contrast to what Rolf Landauer had said earlier, that you can compute for free, in principle, that processing quantum information has no unavoidable thermodynamic cost. What does have a cost is erasing information because that's a dissipative process. So if you just want to compute, and you don't insist on erasing, in principle, you don't need a battery. I emphasize in principle.

And so, that was a great insight. And Feynman was interested in whether that would still be true quantum mechanically because Bennett's analysis had been classical, and he concluded yes, it was still true quantum mechanically. And that was what he focused on in the class, not that quantum computers would be able to compute things that we couldn't possibly compute efficiently with classical machines. Now, he was well-aware of this, and it's far more exciting, it seems to me, and it would seem to almost anyone, that that statement is true than that you can quantum mechanically compute without paying a power bill. But that wasn't what he chose to emphasize. Anyway, in answer to your question, we never really talked much about quantum computing. I knew he was interested in it. We talked about other things. I wasn't that interested. And, of course, I regret that now because eventually, I became very interested.

Zierler:

What were you most interested at the time you joined the faculty at Caltech? What were you working on at that point?

Preskill:

Well, I was still interested in the question of the quark and lepton masses at first, and models in which quarks and leptons were composite is one of the things I was working on. I'll tell you a story about that, too. When I first came to Caltech, I had been at Caltech to give seminars a couple of times, and Feynman had been there. In fact, he had given me a hard time. My first time I had gone to Caltech to give a seminar, I think it was in 1981 when I was making that visit to Santa Barbara, I came down for a couple of days and gave a seminar. Feynman and Gell-Mann were there, and Feynman was quite hostile. And in fact, I was talking about the idea that quarks and leptons are composite. And he made some quite disparaging remarks. But when he would do so, Gell-Mann would defend me. And then, I said something that Murray didn't like. And then, he made some rather cutting remarks, and then Feynman would defend me. I had often heard that when Feynman and Gell-Mann gang up on you at the seminar, it can be a very unpleasant experience. But things didn't go that badly for me. I always had one of them in my corner.

I don't know if Dick [Feynman] remembered that. Probably not. But anyway, when I first arrived, I thought, “Well, I'll have to go up and introduce myself to Feynman.” I went to Helen Tuck, who was the group's secretary, and I said, “When's Feynman going to be around?” “Oh, he's at his vacation home. Probably be back in a couple weeks.” I said, “Well, let me know when he's in town. I'd like to chat with him.” And so, one day, I'm in my office, and then I hear this kind of drumming. Feynman's just walking down the hall, drumming on the wall with a kind of pent-up nervous energy. So, I know that's got to be Feynman because who else would be drumming on the wall? And, indeed, it is.

So I popped out of my office, and Helen saw we were both there, and she introduced me to Feynman. And she said, “Professor Feynman, this is Professor Preskill, our new faculty member.” And Feynman looks at me like he has no idea who I am. And he said, “Oh, yeah? What group?” And I said, “Particle theory.” And he said, “Oh, well, particle theory, that could mean anything. Some people work on quantum electrodynamics. What do you work on?” And I said, “Well, I've been interested in the connections between particle physics and cosmology.” And then, I said, “And I've also been working, without a great deal of success, on the idea that quarks and leptons can be composite particles.” And he kind of looked at me, and he said, “Well, I must say, your lack of success has been shared by many others.” And then, he pivoted and went into his office. That was my introduction to Feynman as a Caltech faculty member.

So anyway, that's, I guess, partially the answer to the question that you asked-- that's what I told Feynman in 1983 I was interested in. I actually got interested in other things having to do with topological defects. I got interested in cosmic strings and whether they could be responsible for the formation of galaxies. And that kind of led me into other properties of defects, including anyons, particles with more general statistics than bosons and fermions.

And then, I guess a couple years after I got to Caltech, something struck, which some of us called “Cole-mania.” This was an idea that came from Sidney Coleman that we could explain the cosmological constant being zero using the idea of wormholes in spacetime. Topologically nontrivial spacetime histories. And I thought this idea was very exciting. I worked on it for a couple of years. It sort of petered out, but in a way, it's come back quite recently, but in a different and more successful form. The idea of wormholes and Euclidian quantum gravity has had surprising success in the last couple of years in helping us to understand how information leaks out of an evaporating black hole, encoded in its Hawking radiation. And it's rather amazing because it turns out that you can compute the information content of the Hawking radiation coming out of the black hole without using string theory or any other microscopic theory of gravity, but just by doing semi-classical gravity and the path integral formalism in imaginary time.

What was your understanding of Stephen's analysis of the developments that caused him to concede where Kip had not?

Preskill:

Well, actually, it was based on Euclidian quantum gravity, which he had long argued was the right approach for answering deep questions about quantum gravity. And it was pretty close to the idea which I mentioned a moment ago, which has arisen just in the last year or two, of using Euclidian quantum gravity to understand how information can get out of a black hole. But it was based on a somewhat different observation, which, I think, doesn't really hold up to scrutiny. He argued for technical reasons—which I think are not convincing—that when you do the Euclidian path integral, there are different topologies that you can include in the histories of the space time, and that there's a reason to include only the simplest topology. And the interpretation of that simplest topology is the black hole doesn't form at all. And the other topologies, which were reputed to describe the effect of a black hole that forms and evaporates, he argued, gave zero contribution to the path integral when you computed it.

It was a pretty technical argument. He never actually published it in detail, and I'm not sure he, in the end, really believed it was right. But he never reneged. He never changed his position again. From 2004 on, he said he believed that information could escape from black holes. In a way, it was slightly disappointing. What were we going to talk about from then on? Because we had had arguments about black hole information going back a number of years by then. And now that we were on the same side, it wouldn't be quite as spicy from then on.

Zierler:

We briefly touched on it, but I wonder if you could explain a little more the creation of the Institute for Quantum Information as a purely theoretical enterprise. Matter does not enter the scene until 2011. What were some of the limitations in the field, both theoretically or experimentally, that would have prevented matter from being included right from the beginning?

Preskill:

Zierler:

And I gave a similar talk, actually, at the Solvay Conference in 2011. That, really, the idea, just from a physics perspective, that we can't efficiently simulate in general quantum systems using classical systems is a very deep statement about physics. I think if you want to say, how are quantum and classical different from one another? This is one of the best answers you can give. Quantum systems, according to the Schrödinger Equation—I guess everybody knows this at some level, it's been discussed for decades—is really hard to simulate with classical computers. And over time, the computer scientists have come up with some useful perspective on the difference in complexity, that there's a gap between the problems you can solve with a quantum computer and a classical one. But, really, it means that quantum systems can do things that classical systems couldn't possibly do, except by taking unimaginably long times to do them.

I guess maybe I should make the remark that the separation doesn't have to do with what the computer scientists call computability. You can get a classical system to simulate a quantum system to any accuracy by evolving for however long you need. The difference has to do with efficiency. That as far as we know, and there are arguments of varying sophistication supporting this, that in order, in general, for a classical system to imitate what a quantum system is doing, there's an exponential slowdown. The classical system needs to take a time, which is exponentially long in the time that the quantum system evolves. So that's a very deep thing about physics. And the point I was making is, we're getting to the point where we can experimentally try to validate this idea.

And so, it's an important research objective, just from a fundamental point of view, to the extent that we can validate this idea to try to use the technology that we currently have or will soon have to perform some task, which might not be interesting for any other reason than for trying to investigate this fundamental question, which we can do with this quantum device, which we can argue is just too hard to do with any classical device, including the most powerful supercomputers we have now. And so, the Solvay Conference was in 2011, the next year, and I also felt this was kind of a milestone. Because, of course, you, as a historian, can appreciate the Solvay Conferences have a certain cachet going back to the beginning, and in the modern era, they've been held every three years or so. And with evolving themes, which had for some time now, you might know for how long, David Gross has been chairing the committee that comes up with the themes for the meetings. And they had been rotating among particle physics, and condensed matter physics, and astrophysics, which are three mature and large communities of scientists.

And so, in 2011, they decided they wanted quantum information to be represented in the meeting. And the theme was called Theory of the Quantum World, but there was a session about quantum information science, and there was a session, both on the theory side and the experimental side. And so, I was asked to be one of the speakers on the theory side. So, you can imagine, it's not a very big meeting because that's how they do these things. There's, like, 50 people in the room. But they're luminaries of theoretical physics, and some experimentalists, too. And there are things which are kind of second nature to the quantum information community, which just are not part of the way most theoretical physics, at least in 2011, think about quantum mechanics. And a very simple one is this distinction—which, from the quantum information community side, is very fundamental—between things that are hard to do and easy to do, what the computer scientists call computational complexity.

And so, I made a remark in that talk that Hilbert space is a myth, in the following sense. Hilbert space is vast in size if you think of it in terms of qubits, as every time you add an extra qubit, you double the size of Hilbert space. So with just 300 qubits, the dimension is 2 to the 300. And you could never, for example, describe a general quantum state in a 2 to the 300-dimensional Hilbert space using ordinary classical information. But more to the point, you can let the quantum system evolve for a long, long time, and it will never, if it evolves according to what we call a local Hamiltonian—which is the way we describe physics, where only a few degrees of freedom interact at a time—it can never visit anywhere except a tiny, tiny fraction of that huge Hilbert space. That's the sense in which the very large Hilbert space is a fiction. It's not something that'll ever be physically accessible, even to the quantum scientists and engineers far, far into the future. Of course, it's very useful to describe this gigantic space for the purpose of thinking about the system. But in a sense, it's not operationally meaningful because we can never go there. And like I said, this is a commonplace observation in the quantum information community. But physicists at the Solvay Conference coming from other fields thought this was a very cool observation.

[laugh] If you look at the trajectory if my career. But I have become increasingly engaged in discussions with experimentalist and experimental groups about, “What kinds of interesting things can we do?” Actually, I mentioned that when we started in 2011, one of the focuses was quantum optomechanics and the quantum mechanics of macroscopic systems, cooling oscillators to their ground states and stuff like that, where we had some real experimental successes. My colleague, Oskar Painter, had a world-leading group doing that type of science. And partly, I think largely, because of discussions that took place under the auspices of IQIM at our annual retreat, Oskar decided to change the direction of his program to developing superconducting devices for quantum computing and quantum simulations, and he brought to that some expertise which is complementary to others who are working in that area, in particular because of his experience with fabrication, and also the acoustic devices.

And so, one of the things that we're still discussing and working on is incorporating sound as well as light into systems with superconducting qubits, which has some potential advantages. I just point to that as an example of how, when you have this kind of hub of activity, and discussions take place, and you're thinking about the future together, it develops potential new directions. And I love seeing that sort of thing happen.

Zierler:

Of course, there's a whole big quantum world out there, and I don't mean in a physics sense, I mean, in an administrative sense. There's so many institutes across the world. In academia, in the national laboratories, and in industry, who have been, over the past decade or more, some of your key institutional collaborators or partners?

Preskill:

Well, let's unpack that a little bit. It's true what you said, all of it. Many institutes springing up in the last couple of years and a ramping up of activity in the national labs and in industry. And those are all good things. And Caltech, as I said, has some vision for the future of the field, and that is probably best reflected in the faculty hiring that I spoke of. But we're also interested in continuing to invest with further faculty appointments, and, of course, facilities that are needed. It's good to see the national labs getting involved. This has been partially facilitated by the National Quantum Initiative, which was signed into law in late 2018, and which called for increased funding for DOE, NIST, and NSF—and in the case of DOE in particular, prescribed the creation of national centers for quantum information science, of which there are now five getting underway, each with its own identity.

And then, there's industry, and there are some companies which were in quantum information early on. IBM is maybe the most conspicuous example because they had people working on quantum information when the field was really starting to explode in the 90s already, and they had a very distinguished theory group and some experimental activity at the time. And that has vastly expanded into making them a major center of quantum computing with superconducting devices, and they now provide cloud access to users around the world, including many students. So that's certainly been a useful service.

Is there a unified theory, or at least shared understanding, of quantum computing so that when it happens, everybody will agree that it happened? Is there a shared set of parameters that everybody goes by and accepts? Or even, are those basic questions still being hashed out?

Preskill:

There is a fair amount of discussion of how to quantitatively benchmark different platforms so you can compare them. Like, how can you say that platform A is better than platform B? And even that's a complex question because A might be more useful than B for one type of application, and B more useful than A for another. But the criterion that I would prefer to use is running on a quantum computer or quantum simulator, some application, which people want to run because they can answer questions that have some practical impact that they can't otherwise answer. So, we're looking forward to when that will happen. Hasn't happened yet, and we don't really know when it's going to happen. And I would say, harking back to the point I made earlier about the three big questions, I said one is, can we do something interesting with the near-term devices? And another was, how are we going to scale them up? And the third was, what are we going to do with them?

And one of the questions, which is an important one for the current stage of industry, is, are we going to need to use quantum error correction to do something that's practically useful? And this is a really key question because what we have now—in the case of ion traps and the superconducting circuits—is devices, which have a number of qubits of order 50 or so, and you can do the fundamental quantum operations that you need to do to run an application. The key ingredient in those quantum circuits is the two-qubit gates, the gates that act on a pair of qubits in some entangling fashion. And a measure of quality of the qubits is—it's a little subtle to quantify this precisely, but let's put that aside—how often is there an error in one of those entangling two-qubit gates? And the current state of the art is that those error rates are a little better than 1%.

So that means you do, say, 200 gates, and one of them's probably bad. And that's a real limitation on how large an application you can run. Because the errors, when they occur, they spread, and they can lead to the answer of the computation, when you finally measure all the qubits at the end, being wrong with sizable probability. Or a better way to say it is, there's some signal that you're trying to get out from this noisy system, and there's a lot of noise. And you need enough signal-to-noise so that by running the computation some acceptable number of times, you can extract some statistically significant signal. And like in the Google experiment, which they did with the 53-qubit device, they actually had to run a few million times a circuit, which had a probability of success which was about 1 in 500.

So, think about that. You run 1,000 times, and 998 times, you get junk. And 2 times out of 1,000, you get a useful result. And then, to extract that from the noise, you run it a few million times. And then, you can do it. And they were able to do that in a few minutes, to their credit, because they have pretty fast gates and a reasonable cycling time. And that's how they achieved what they claim to be a successful demonstration of quantum computational supremacy. But it's hard to go much farther with the devices we have now because they're so noisy.

So, I guess part of the big question is, can we—and how soon can we—have hardware which is much more reliable, so we can get those gate error rates down a lot? That will have a big impact. But another near-term question is, if we're going to live with error rates at the level of, let's say, 10 to the -3 or worse, how far can we go towards running a useful application? And obviously, you can't have more than a few thousand gates, or it's going to fail. So, what can you do with a few thousand gates? A few thousand is probably optimistic, at least right now. Well, that's where quantum error correction comes in. We understand in principle that it's a long-term solution to the problem. It adds a level of redundancy to the computation that protects against noise, just like error correction does in classical settings.

Of course, we use error correction in classical communication routinely, like in our cell phones. And quantum error correction was actually a huge advance, I think, in physics, of historic proportions in the mid-1990s, when we understood in principle how to do this, that we can encode quantum information so it can be protected from noise. But it has a big cost, a high overhead cost, so that we'll need many more physical qubits than so-called logical qubits. If I want to run an application using 100 really, really good qubits, then I will have to encode those qubits with sufficient redundancy to protect against the noise, and how much that cost is in extra qubits and extra gates depends on what application we're trying to run, and what the quality of the qubits are. But if I wanted to do something like factor a large number with the hardware we have now, it probably means thousands of physical qubits for every logical qubit, and the number of logical qubits you need in factoring is a few thousand. So, we're getting up into the 10-million physical qubit range. And just think about what a big leap that is from where we are now with 50 to 100 qubits. There's some vision of how to get there, but it's really hard. And it's not going to happen real soon, unless there's a big surprise.

Zierler:

One of the big storylines over this past pandemic year for experimental physicists has been-- they have increasingly turned to computer simulations because they haven't been able to get into the laboratory. One of the huge frustrations in particle physics is, absent the SSC, who knows what else is out there at those energies, right? Can quantum computing, quantum information get to a point where building something like an SSC or an ILC isn't, dare I say, necessary? That the information that we can get from quantum can be reliable enough that we can see things virtually and be confident in them without having to see it in the real world?

Preskill:

Well, we're trying to understand what the right theory is. For that, I think experiment, at some level, is indispensable. If there are different alternative theories, we're going to try to distinguish. But quantum computing can be very helpful for that because it will allow us to predict by simulation the consequences of different candidate theories well enough so that we can compare with experimental data to distinguish one theory from another. So at the LHC now, and in particle physics going back decades, the key challenge is—and this is true in experimental physics and other fields, very broadly—their backgrounds. And you have to understand them in quantitative detail. And you're looking for a tiny signal, and you have to be able to know how to subtract those backgrounds. And if you're doing a collider experiment at TeV scale, that background is mostly QCD.

Well, I'll tell you one thing I've noticed is this. I was, for some time, a freshman advisor at Caltech. I'm not doing that the last couple of years. But my advisees, I would have seven or eight a year, were supposedly assigned to me at random, not because they were interested in my research area or even in physics, necessarily. But I would usually, in my first meeting with them, just kind of be getting acquainted and try to probe a little bit, “What's exciting to you? What about science or engineering really intrigues you, thrills you?” And even though this was supposedly a random selection of Caltech freshman, maybe ten years ago, it would be things like string theory. And maybe a little bit later, LIGO. Kind of cutting-edge physics. And now, by far, the most common answer is machine learning. They see that machine learning is really changing the world in a practical way, but it's also changing science because increasingly, we're going to use those tools to interpret data and even to advance theory.

And, of course, that gives rise to questions about the future of the human scientist, which are intriguing. Are we all going to be supplanted, just like the Uber drivers and the truck drivers and the factory workers? Will the physicists also become obsolete when the robots take over? That's another question. But anyway, understandably, they see machine learning as sort of pivotal for the future of the world and for science. And so, I get pulled into that a little bit because I sense that excitement. And, of course, it's not just the freshman, but many of the young scientists, the graduate students, the post-docs, and young faculty are engaged in using machine learning in their research areas and excited about it.

You asked what courses I got a lot of pleasure out of. Well, my first teaching experience, I was still on the faculty at Harvard, and I taught one of the freshman physics classes. I taught that with Roy Schwitters, we co-taught it. It was my first teaching experience, except as a graduate student TA. It was freshman mechanics. And it's showmanship. Obviously, you're trying to convey the science, but you're trying to make it exciting and engaging. And it's really fun because you get to do things like ride a rocket car. And we would ham it up, put on a pair of goggles, and a scarf, and a little helmet, and come shooting through with a rocket car. So, stuff like that was fun. I guess it kind of appealed to my instincts as a performer. But I haven't done much of that. I did Physics 2 at Caltech a few times. But the demonstrations weren't quite so spectacular on the special effects side.

Another thing I really enjoy teaching is statistical mechanics. We have a class for the physics concentrators, but it's also taken by other people who intend to focus on scientific areas, including chemistry, and math, and stuff like that. We teach it to sophomores. And it's actually a three-term sequence. But the part that I think is the most fun is stat mech because you can, starting from very simple principles, basically counting, explain physics phenomena like–of course, you talk about the quantum gases, so you talk about degenerate Fermi gases and what that has to do with metals and white dwarfs and neutron stars, and you talk about the Bose-Einstein condensates, and then you can say something about the revolution that's occurred in atomic physics over the last 25 years because it's become routine to prepare Bose-Einstein condensates in the lab. And then, you get to talk about phase transitions, which is also great because you can give them a taste of the modern view of phase transitions in terms of critical exponents, and scaling, and things like that. So, I found that to be a lot of fun. I also like teaching classical mechanics because it's just so beautiful.

I think I mentioned last time, I learned classical mechanics as an undergraduate from Alan Guth. And he taught a brilliant course, and I've stolen a lot of it. [laugh] I have taught classical mechanics at Caltech. But the educational experiences which I think have been most transformative for me as a scientist have been in teaching graduate classes, particularly on special topics because it's an amazing learning experience for the instructor. If you have to teach fairly recent developments to graduate students—and at Caltech, our undergraduates are very ambitious, and so they take these classes in significant numbers, too. I taught a course on Non-Perturbative Quantum Field Theory in the early 80s, and it was about quark confinement, and chiral symmetry breaking, and lattice QCD, and stuff like that.

And the experience of teaching that really deepens your understanding. I have withdrawn that deposit from the bank many times in the ensuing decades. And I think last time, I mentioned how Kip encouraged me to teach a class on Hawking Radiation, which was also very beneficial to me. But teaching quantum computing was a huge deal because I did it for the first time in 1997. The field was pretty new. Shor's Algorithm had been discovered in ‘94, quantum error correction in ‘95, how to do fault-tolerant quantum computing in ‘96. It was all fresh and exciting. And I had the opportunity to synthesize it. And by that time, the technology existed to reach a larger audience by posting lecture notes on the Internet. In my earlier classes, I made handwritten notes which were Xeroxed and distributed, and I eventually scanned those and put them on my website. And people sometimes look at them.

But in the case of the quantum computing, I had the notes LaTeX-ed and posted in more or less real time as I was doing the lectures. And I still hear from people often that their introduction to quantum information came from reading those notes, which are still available on my website, and which people still read. So, it was an interesting vignette of how the Internet has changed scientific communication because, of course, a lot of us put a lot of work into our classes, and that's to our own benefit and to the benefit of the maybe a few tens of students attending a very advanced class. But when you can put your lecture notes on the Internet, and thousands of people read them, that really amplifies the impact. But most of all, I learned the subject much more deeply from teaching it.

Zierler:

On the graduate side of things, as an intellectual history question, one of the best ways to gauge a professor's research intentions is what kinds of graduate students they take on at any given stage in their career because one of the things you have to do as a successful graduate mentor is to be on top of the literature in, in your case, particle physics, in cosmology, in quantum information. So just over the years, where have your graduate students been? Have they roughly followed your interests over the years? Or do you still take on students that have interests that are more in line with where you were 20 or 30 years ago?

Preskill:

If we go back to the beginning, when I first came to Caltech, maybe we discussed this last time, there was a lot of pent-up demand among the graduate students in the particle theory group for advisors. So, I very quickly accumulated eight or nine students. That maybe set the pattern for my mentorship in a way because there's no way I have the bandwidth to be really deeply engaged on a day-to-day basis with the research of nine students. And so, my goal and style was to sort of create a community where people were talking, and I was giving regular feedback, and where the students were learning from one another and from post-docs in the group. And I could sort of set the agenda to some degree. But already back then, I was encouraging students to come up with their own problems, which was what I had done. And somehow, I thought it was a good model. Though it's not a great model for everybody.

But the most important thing you learn as a researcher is to figure out what to work on. And, of course, as a mentor, I'm supposed to provide guidance there. And to some degree, I do. And sometimes, I make rather concrete suggestions and collaborate with students as well. But in some cases, the students come up with their ideas, and I try to provide a little guidance pointing that idea in a direction which I think is most likely to be fruitful. Now, since I went through this transition in the 90s from a focus on particle physics and black holes to quantum computing, it was a very exciting time because I was excited about those ideas, and so were some of the students.

But it was transitional because I had some students who didn't go that way, who had started out working on other things. I think it was around 1997, I had five PhD students finish the same year, and they all worked on very different things. One was working on dualities in quantum field theory, and one was working on exotic statistics in condensed matter. And then, one was working on quantum error-correcting codes, and so on. So that was kind of fun to have all those balls in the air. But it made our group a little bit less cohesive.

And now, quantum information has become a broad enough field so that you can have different students engaged in different aspects of it. Like now, I have students who are interested in machine learning and its applications to quantum information and or in another one of the things we've discussed in this conversation, the prospects for simulating quantum field theory with quantum computers. Some are interested in quantum error correction and its applications to quantum gravity. And some are interested in anyons, and whether we can realize them, and physically realistic platforms, and what we can do with them.

So that's actually pretty broad. But there's enough overlap so that at least it's my hope and belief that we function as a community, and everybody can be interested in what everybody else is doing. I really value our group meetings where we all get to hear about the progress that others in the group are making. I think that benefits me to know what everybody's doing, but I think it benefits everybody to hear about it. And I've had a few cases where I've had students who were really exceptional, and I'm proud of making very good suggestions. Daniel Gottesman was one, when I was first getting interested in quantum error correction and steered him towards that. And he wrote a thesis which is still a standard reference in quantum error correction and fault-tolerant quantum computing. He got his PhD in 1997.

Another one was Jeongwan Haah, who's now a researcher at Microsoft. I gave him a very good problem, which was to try to understand whether you could make better quantum memory if it were configured in three dimensions instead of in two. And he discovered a new phase of matter, which the condensed matter physicists are still puzzling over. So both Gottesman and Haah wrote single-author papers, which had a lot of impact. That's something I'm proud of. Of course, I'm singling out those two as particularly impactful examples, but I've had about 60 PhD students, and many of them have had successful scientific careers and were great performers in graduate school, and I'm proud of all of them. And maybe we talked about this briefly last time, we've had a lot of post-docs go through the group, and that continues, who have had a huge impact on the field, and I'm really proud of that, too.

Zierler:

Last question, looking forward, so much of your scholarship over the course of your career works in the world of fundamental mystery, ongoing things that we still don't understand. If you were to combine your expertise in all that you've learned in cosmology, particle physics, quantum information, and brought them to mutual benefit, is there one problem in physics that stands out for you that you're either most optimistic about solving, or you just remain most curious about and are optimistic that all of these different research areas of expertise might come to bear on each other and provide something really fundamental for the remainder of your career?

Preskill:

Well, I can think of a few possible answers to that, but I'll give one, which is the one that most leaps to mind. And that is, I would like to see, while I'm able to enjoy it, significant progress in understanding quantum gravity coming from laboratory experiments with quantum simulators and quantum computers. And thanks to this idea of duality, that we can describe quantum gravity using an ordinary quantum system, this is not a ridiculous proposal. And it does combine all the things you talked about because it involves building and operating quantum platforms that are up to this task, it involves formulating the questions in a way which can be conveniently addressed by doing experiments. It connects with condensed matter physics because the way you would do it is to create a highly correlated, strongly interacting system of many particles, where the geometry of this quantum gravity theory would be a kind of emergent property. A lot of the most fascinating things in quantum condensed matter concern the emergence of an effective description, which is very different from, and not so easy to predict from, the underlying microscopic description. That would be a great way to combine together all in one cool package the things I've been interested in my whole career.

Zierler:

Which would tell you what, best case scenario?

Preskill:

Well, first of all, this idea of duality, that we can describe quantum gravity in terms of a theory that doesn't seem to have anything to do with quantum gravity at all, has been understood in a few rather narrow settings. Essentially, the looking under the lamp post phenomenon. There are a few cases where, for example, the theory has enough symmetry that we can kind of figure out what the duality looks like. But I don't think we have a very clear understanding of how far-reaching these dualities are.

And so, I don't know if it's the best case scenario, but a good case scenario would be that when you have many strongly interacting particles, they do things, which we can't very well simulate with classical computers because it's just too hard for all the reasons we talked about earlier, but where you can come up with an effective description which captures what's going on in a way that is intuitively appealing, and that description will involve gravity. Just an ordinary system of atoms or something interacting strongly with one another, evolving dynamically in some way, which we might not understand at first from first principles, but where we can describe what's going on by saying, “Ah, well this is behaving like an emergent black hole.” Or it's an emergent wormhole in space, or something like that. And in the long run, when we imagine using quantum computers to simulate many body physics, which goes beyond what we can simulate classically, we won't be satisfied, at least I won't be, unless we have some way of "understanding” it, of describing what's going on in some language which we think is useful for leading us toward future discoveries.

It would be rather disappointing if all we can say is, “All right, well, we saw how that worked. And I guess we won't learn anything more until we build an even more powerful quantum computer.” We want theory and experiment to advance together, as they always have in the past. And as we learn new things from experiments, we want to be able to integrate that into our understanding, so that we can propose future experiments, and predict what would happen in them, and either find those predictions confirmed, to our credit, or find the delight of being surprised, and therefore, being led to further new discoveries.

Zierler:

Well, on that note, it's been unbelievably fun to spend this time with you. Thank you for allowing me to convince you to do this. I really appreciate it.

Preskill:

Well, I have no regrets. I had a good time.