January 31, 2022
Like many Caltech undergraduates of his generation interested in physics, Dave Bacon's original plan was to pursue research in string theory or some other area of high energy physics. During his teenage years, Bacon was already interested in software programming, and it was at Berkeley during graduate school that he saw great possibility in quantum error correction - a monumental challenge that could draw on his interests in computers and physics. His thesis research posed a foundational question: can quantum computers correct their own errors?
Returning to Caltech as a postdoctoral scholar, Bacon joined the Institute for Quantum Information when topological quantum computing and quantum entanglement were major fields of inquiry. After research appointments at the Santa Fe Institute and the University of Washington, Bacon joined Google, first working purely in software engineering, and then becoming a key player in the company's embrace of quantum information research. A brief appointment at IonQ gave Bacon a fuller appreciation of quantum computing from the perspective of a tech startup. Looking to the future, Bacon is optimistic in what he calls a "middle way" that will blend the different approaches that Microsoft and Google are applying to achieve a scalable quantum computer.
DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Monday, January 31, 2022. I am delighted to be here with Dr. David Bacon. Dave, it's great to be with you. Thank you for joining me today.
DAVID BACON: Thanks for having me. I'm excited to talk about some fun stuff.
ZIERLER: To start, would you tell me your current title and institutional affiliation?
BACON: My current title is Senior Staff Software Engineer at Google, and I'm based here in Seattle. I work on the quantum computing group for Google.
ZIERLER: Let's start with research fields and see where there are intersections. Of course, your educational trajectory is in theoretical physics, there's software engineering, and there's computer science. Where do these different fields converge on the major research topics that you're after?
BACON: My research has always been in all of these. For a lot of us in quantum computing who started in the 90s, when it really started becoming en vogue, it was very natural because a lot of us were computer geeks who liked physics. It was like, "This field is just set up for us." [Laugh] I currently am a software engineer, so I do a lot of building tools and things like that for quantum computing, and I do a little bit of research on the side. I've been lucky, I've been able to do all the different areas, both in physics and computer science, even into this area of building tools for actually making research easier.
ZIERLER: I'll ask a series of questions. To the extent that you're able to talk about your work, wonderful, and where you're not able to, totally understand. Just as a snapshot in time, what are you currently working on, and how does that fit in with Google's overall project in quantum computing?
BACON: Google started pretty seriously when they hired John Martinis, which is, I guess, about five years ago now. They'd previously done some work prototyping D-Wave's quantum system. But they sort of went through this period of doing what they call the quantum supremacy, which we now call the beyond-classical experiment. I got dragged in to work on software that was necessary for quantum computers. Our current big thing at Google right now is pushing toward error correction. I'd say the main thrust of the team is working on showing that error correction works. We're sort of at a very fascinating time where error correction, for the first time, is starting to be implemented in the lab. Amusingly, for a lot of the experiments right now, if you're trying to do error correction, and your gates and things aren't good enough, what actually happens is error worsening. It makes things worse, it amplifies in the wrong direction.
We're doing a lot of error worsening experiments right now. But there are ones that are sort of pushing the boundary of actually performing error correction. The team is heavily focused on that. The other thing the team is heavily focused on is what we could use these quantum computers for, thinking about what people call the NISQ era, which is a term John Preskill came up with, and then for fault-tolerant quantum computers, when we have them, we really need to bring down the costs associated with all the algorithms. A large chunk of the team is focused on trying to understand what the applications are, when they'll arrive, trying to make it quicker by building better algorithms.
The Relevance of Software to Quantum Computing
ZIERLER: What are the ins and outs of software engineering when it's applied to trying to build a scalable quantum computer? In other words, do you need to have a quantum software program in order to make it applicable to what will be a quantum computer?
BACON: Essentially, there's the quantum part of the system, and we're sending in classical control sequences to these machines. We have to build the software that controls those classical signals. What's interesting about that is, it's a pretty complicated system you need to build, and you need to be able to scale it up. On top of that, you need to be able to orchestrate this error correction algorithm. The joke about quantum computers is, they'll mostly be doing error correction. Their current vision of doing active error correction is this idea that we're running this very complicated program at the lowest level just to make the machine work. You have to build and architect that so it'll actually work. It's almost like having a classical supercomputer sitting right beside a quantum computer to do a lot of computing to suppress those errors. There's sort of this quantum operating system you need to build that orchestrates the quantum computation you're doing and the error correction.
Further up the stack, you need to think about how you program these. Do you write it in a new language? Or do you use an existing language and augment it in order to do that? One of the things I've done is worked on a program called Cirq, which was basically a python framework for programming quantum computers. What's kind of interesting is, we're very much at the beginning of this story. These original tools we're building are very much version zeroes or version minus-ones of what's going on. It's kind of exciting because there's a lot of space to improve in that software. The first time you do it, you always mess up. We're at that era for quantum software. It's pretty cool.
ZIERLER: Operating in an industrial environment, what aspects of your work and the work of your group are simply fundamental research that's supported by Google, and what aspects of it are really responsive on Google's overall strategy, where it wants to see quantum computing going?
BACON: In the case of quantum computing, the way to say it is, the smart money knows it's still a long push to get to very practical applications. At Google, our CEO actually has a background in materials science, so he sort of fundamentally understands that this is an important tool to build. They very much understand that it's a long march. Some of the stuff that people do is closer to what you would call fundamental in the sense that it's exploratory. Trying to understand errors in our systems is very much trying to analyze the behavior of the physics of your system. If an academic were doing this, you'd be like, "Oh, yeah, that seems about right." The actual process of building it has a lot of parts like that, but a lot of it is pushing engineering to scale this thing up and all the management and coordination you need to do that. It's hard to know what the actual mix is at Google, but there's definitely a mix between them. The thing I actually think is fascinating is that you see the team pivot back and forth. They sort of go into science mode to try to figure out what's going on.
Once they understand it, they go into engineering mode to try solutions. They go back and forth. I'm always really impressed with the people who are doing this because I'm not great at it myself. But watching that process back and forth is really fascinating. Even in the best times in academia, that's always occurred. This notion of being not dependent on the hardness of actually carrying out your experiment and engineering it has always been there. It's a pretty rapid cycle at Google. I'd say that's sort of the lay of the land. On the theory side, it's still all very research-y in a more abstract sense. People are thinking about algorithms. Some of it is practical in the sense that they're trying to actually see how an algorithm would behave. But a lot of that work is still very much, "What can we do with this quantum system? What are the real algorithmic questions here?"
ZIERLER: Being at Google, what insights have you gotten in terms of quantum computing as really a merging of both theory, experimentation, and engineering or materials science?
BACON: This is actually the best part about quantum computing. It really is about bringing fields together. I have a story from my IQI days about that, which is that I was originally a physicist, then when I showed up at IQI at Caltech, I was more interested in some computer science questions, algorithms, things like that. I got put in a room with Sean Hallgren, who was a computer science PhD from Berkeley. I was a physicist from Berkeley. We knew each other a little bit in passing, but it literally took us, like, a year to be able to talk to each other. But it was one of the most productive things that ever happened to me because being able to speak the language was key. I think that's actually the most interesting thing to watch at Google, watching people try to break down the barriers between the different disciplines and talk to each other. And there's a huge benefit when that happens.
It is amazing though because most scientists are these really brilliant people who just don't speak the same language. They've been taught different ways to think about things, and it's hard to convey to the other person without having gone through the experience of training in whatever background you had, that understanding. But as it happens, you start to see a lot of benefits. It's probably the best part of my job, that I get to speak to physicists, engineers, even spend time in the product world and finance. It's everything at once. [Laugh]
ZIERLER: As you know, major companies, peers of Google, are involved in quantum computing. Honeywell, Amazon, IBM, the list goes on and on. Do you feel to some degree that all of these companies are in a race? And what might be Google's unique approach, if its aim is to get their first?
BACON: In some sense, it is definitely a race. The field is still fairly collegial in the sense that for many years, quantum computing was an outsider's field. You'd get very dismissive comments when you'd talk to physics departments or try to get a job anywhere. For that reason, the community was very tight-knit, I would say into the 2010s, when the commercial interest started to come around. Some of that still remains. I'd say it's a race, but people are very cognizant that we all really want to build the quantum computer. If somebody else starts doing a really good job, fundamentally, most of the people who have been in the field are like, "That's pretty awesome." [Laugh] That's good. I would say Google has a strong culture of doing hard things. That's what I always found fascinating about this place, that you hear a lot about Google products that were canceled and things, but there's a lot Google does that you never hear about. People are trying hard things, which means, of course, sometimes you fail.
And this is one of these sweet spots. This is a really hard thing that appears to require a lot of engineering right now. Our best ideas do require a lot of engineering. The weakness Google has is, we're not traditionally a hardware company. We've moved in that direction, but hardware is hard, as they say. I think there's definitely a culture of trying to do hard things. We do have the advantage, also, I will admit, that Google's software engineers are very good. Getting software engineers to work on the quantum team is super easy because everybody wants to work on it because it's cool and exciting, and we have really high-caliber engineers. One team member took the physics GRE before he interviewed with our team to transfer. I think he did better than I did on his score. [Laugh]
ZIERLER: A few questions as they relate to process. As you well know, for example, Google's approach to quantum computing is very different than, say, Microsoft's, for example. The question there is, one, do you have any insight as to why Google chose the course that it did, and two, is your sense that the approaches are different, but the aims of the outcome are the same? Or is it possible to conceptualize two different processes leading to two different types of quantum computers?
BACON: Let me take that second one because that's actually a fascinating one. I spent a year at a startup called IonQ that builds trapped-ion quantum computers. These are very different beasts than the ones Google builds, which are superconducting circuits. The trapped-ion quantum computers have some benefits in that each qubit can talk to every qubit in a trap, whereas on a superconducting circuit, you lay it out on some grid, and only nearest neighbors can talk. There are different behaviors. Superconducting circuits are much faster than trapped ions, which are slower. When I made that switch, one of the things I really was thinking about is that there are some places where trapped-ion quantum computers seem better and are closer in the sense of quantum sensing applications, where you're trying to use the computer as part of a sensing network.
Those types of things might appear quicker. In the long run, when you look at the speed of the trapped ions, you're kind of like, "Hm." If you're a factor of 1,000 slower than your competitor at the base rate, that can be a problem. [Laugh] But I really do think, actually, there are probably multiple endpoints here, and we'll start to see them emerge probably in the next five years. I do think things like quantum sensors are going to have a major impact, and some physical systems are better for that than others. The other question you asked was about Microsoft and Google, why Google went the direction they did. Terry Rudolph has a very funny quote that's something along the lines of, "Everybody in quantum computing works on what they do because that's what they did in grad school," which is actually really true. It's not entirely true, because there are some people have moved.
My personal thing is, hire people who move across disciplines because that's really hard, and they have a different perspective. But it really is true that the early days of quantum computing companies have been people who did it as their grad work. IonQ, it's Chris Monroe, Jungsang Kim, and all of the graduate students and post-docs who were building that. Google was superconducting circuits coming out of John Martinis. It's these academic cores that started getting good results, and the companies either emerged from them, or in Google's case, they were trying to understand at the time what the best possible path was, and John Martinis and superconducting circuits seemed to be the right fit. There's another tradition, which is sort of associated with Microsoft, which is topological quantum computers and this idea that we can build naturally robust quantum computers. Caltech has a huge part of that story, actually. But it's been more challenging than you would expect.
Actually, John Preskill likes to say, "I remember Dave Bacon used to always say that we would never build a quantum computer unless it was topological." [Laugh] I kind of stuck my foot in my mouth on that. But in some sense, I believe something like that's still true. I still think that the ideas of topological quantum computing are extremely important. Google's system they're trying to build is building the topological code. In some sense, we can't build it directly, we're sort of simulating it and doing the error correction part of it. It's kind of fascinating that those ideas about how to robustly store information in two-dimensional systems are key to a lot of our current efforts. They're not as far apart as people think they are in some ways.
Quantum Startups Big and Small
ZIERLER: Thinking about comparing and contrasting your experiences at Google versus a startup like IonQ, what does it say about the quantum computing field that you can have really tiny startups making contributions, and you can have behemoths like Google and Microsoft working on it as well? What does that tell is in historical perspective about where quantum computing is right now?
BACON: I think what it's pointing out a lot of is just that talent is the key thing right now, knowledge about how to do these things. In startups, it's always tended to be small groups of experts. If you look at the industrial efforts, that's true as well. If you look at Google's stuff, it came out of John Martinis's lab originally. If you look at Microsoft, a lot of their efforts in topological quantum computing came from people like Charlie Marcus and Leo Kouwenhoven. There are sort of these academic histories of that. Because of that, I think it just means that we're still in a stage where we're taking academic ideas and moving them across the divide. It's also true that Google's team was always fairly small. Honeywell may be the largest they combined with CQC. They have probably the largest number of people. IBM is also large. We are starting to see the emergence of these much larger teams trying to do things. It's also this emergence of trying to think about it as a product. I'd say Google is still far from that.
We still believe this is a hard problem. We want to be prepared for when the product comes around, but it's not an end unto itself right now. And that's very nice. It's sort of liberating. I worry about some of the companies just trying to be too much of a product too early. IonQ, which went public, one of the challenges they're going to face is now, their CEO has to go talk to investors every quarter, which can be a pretty big distraction. It can be a motivator, too. [Laugh] But the business can be a pretty big distraction from the real challenge, which is build and scale that trapped-ion system out. That's going to be interesting to watch because I think you can get distracted a lot by the product, and what's really going to happen is, technology's going to win, so that's what people should try to focus on.
ZIERLER: I'm curious how much of a challenge the process is where, if we don't yet have a good understanding of what quantum computers will be good for, how difficult does that make the process of conceptualizing how to get there? Or is that really not relevant because we know how to build it, we just have to be able to actually do it, and then we'll figure out what it's good for?
BACON: We do know some things. We know it breaks RSA, so it scares the poop out of all the three-letter agencies. Then, there's the original story that goes back to Feynman, which is quantum computers as a simulator of quantum physics. And in some sense, you do have to actually hash that out and see how it works because it's not obvious that–maybe there are classical algorithms. It's true that we have classical algorithms that, in some places, are very good for some of these things. It's kind of squishy exactly where that boundary is. The bigger worry is, is that big enough? That's an important tool that could lead to very interesting new science and technologies, but it's kind of weird in the sense that it can lead to companies making breakthroughs in fields, but how do you value something like that? It's very different. We don't have a broad set of algorithms. We have a lot of polynomial speed-ups, and these are very challenging, in our current way of building hardware, to have practical impacts. There's a lot of worry that we just don't have that larger thing. What's interesting is, in the last few years, we've seen people getting their hands on small quantum computers.
They've started thinking about these variational algorithms like, "Let's just push it and see if there's something there." And we still haven't seen as much of what I'd call abstract algorithm development. That's always been hard. That's one of the things I worked on while I was at Caltech. One of the reasons I worked on it was because nobody was working on it. It got a reputation as really hard. People like Peter Shor, Manny Knill, all these very smart people had worked on it and not made much progress in the late 90s after Shor's algorithm. I kind of worked on it because I was like, "If I can make a small amount of progress, I can give convincing talks. 'If even Idiot Dave can make progress, you should work on it, too.'" And we still don't have a ton of that. Partially because so much money's being dumped into the field, I'm starting to see again some of a resurgence in people thinking hard about new algorithms. Maybe we're right at the time where we're actually going to start seeing the major changes I've been waiting for.
ZIERLER: From your perspective, where you are at Google, what is so difficult about quantum error correction? What's the challenge?
BACON: A lot of quantum computing is, you'll see a press release, "Researcher's come out with this great qubit that has a very long coherence time," or something. And they focus on one particular part of building the quantum computer. Control of a single qubit gate, doing the two-qubit gate, or leakage, things like this. But really, you have to do all of those things at once. You have to make sure that all of the benefits you get in your individual experiments actually work when you're trying to build this larger system. A lot of times, that's extremely challenging. I'd say that's kind of what the field is dealing with right now, how to get the performance they know they can get at the smallest scale and get it across, say, 100 qubits at once and still maintain that so you're low enough to do error correction. Then, you have to coordinate it and do it, so there are some long-term challenges. The error correction algorithms you need to run are, right now, not fast enough. The superconducting circuit ones are not fast enough because those are pretty fast quantum computers. For trapped ions, they might be better because they're slower quantum computers. Those algorithms, in the next three to five years, need to be able to show they can operate at a high enough speed. But I'd say it's really this question of getting everything you could do on a small scale to work at a 100-qubit scale.
ZIERLER: Putting on your theoretical physicist hat, thinking about the potential impact of quantum computers for physics research itself, what are some of the exciting possibilities, having a well-developed sense of all of the ongoing theoretical challenges in physics? Where might quantum computers provide some insight?
BACON: One of the things that's fascinating, I left Google while they were doing a bunch of these experiments and then came back, but this team has had a series of what they call physics experiments, where they're looking at things like time crystals, out-of-order correlators, and these are all related to simulating many-body systems. If there's hope for the NISQ era right now, it's really that these, say, 100- to 1,000-qubit machines will be an incredible simulator of all these questions in condensed-matter physics that come up. And I do think we're right at the boundary of that. I can sort of see that happening. The things that we're starting to get there are, can we start to use it to gain insight into these physical systems? I'm always a little bit worried because quantum computing people always say things like, "Well, we'll use a quantum computer to simulate high-temperature superconductivity, then we'll get room-temperature superconductors."
There's this giant leap in logic that occurs between those two statements. [Laugh] Not to say that I don't believe that simulating it might help us understand better theoretically what's going on. I think it's still challenging to use a simulator to help boost yourself up this level of what you can and can't do. But I do think that a simulation of these condensed-matter systems is going to be fascinating. Again, here, there's a place where different architectures are very interesting. Neutral-atom quantum computers look to be way better simulators because they get a lot of qubits together. They don't have exactly the right control necessarily as easily, but they sort of naturally fit into this simulator system better. It is interesting, also, and maybe this is no longer true, but five or six years ago, it was true that Europe sort of pivoted very hard to thinking more about simulators than quantum computers.
Google and others sort of went this other direction. I suspect that both are going to pay off, but there was sort of a division in some ways, where it was sort of thought that only industry had enough capability to do that scaling question, so others were trying to work in other places where they didn't have to deal with that. But my dream is always a room-temperature superconductor. That's just one I think is fascinating. And it's a curious one because people don't really work directly on it. You sound like a kook if you say, "I want to work on room-temperature superconductors." People say, "That doesn't exist." And you're like, "Well, yes, that's exactly right." It kind of makes me mad about academia, like, "That's what you should be thinking about." It's explicitly one of these, "Why not?" things. Instead, they justify it by, "We'll do simulation, then we'll understand it, then do it." It's like, "Well, maybe you don't even need to understand it to at least have some vague understanding." But it's a fascinating area. [Laugh]
ZIERLER: Let's go back now to undergraduate days and get a sense of where your interests were circa 1993 when you got to Caltech. First, was it physics all the way? Was that what you came to Caltech for?
BACON: Yeah, I caught the physics bug pretty early. I was also a hacker. I spent a lot of time running Cellular Automatas, neural networks even back then. But a lot of it was for simulating things. I simulated physics. That was one of the first things I did when I got a computer. I didn't know trig at the time, so I did what's essentially a small-angle approximation. I got these really weird orbits, and I knew they were ellipses. Eventually, I had to teach myself trig in order to figure that out. [Laugh] I was always on the boundary in some ways, but when I showed up at Caltech, it was very clear almost everyone took Physics 1. There were a few people who placed out, and my roommate was one of those, super genius smart physics guy. But what was pretty clear was, a lot of people came in, they start taking physics, and for some reason, physics is harder. If you can do it, you should do it. I remember the computer people had way more experience with all–I grew up in a very small rural area.
I would go over the hill an hour away to go to the bookstore to sit and read the manual to learn how to program the thing. [Laugh] That was sort of my access to knowledge, whereas these people were on the internet already and had vast knowledge of these things. I was like, "I don't think I can do that." I literally took no straight computer science courses while I was at Caltech. I took a ton of computational neural science, which was really fascinating. Especially, at that time at Caltech, Hopfield was there, so I got Hopfield networks from Hopfield, which was pretty cool. But I pretty quickly realized I really loved physics and was good enough at it to do at Caltech's level, which doesn't always happen. [Laugh] A lot of people come in loving it, but physics has a lot of challenges. I went that direction.
ZIERLER: A generational question. 40 or 50 years ago, physics was, by far, the dominant undergraduate major at Caltech. To fast forward to today, computer science is. When you were an undergraduate, where were you in that broader transition.
BACON: It was starting to transition more to electrical engineering and computer science. You could see that starting to occur. We hadn't been through the dot-com bubble yet, so we weren't at that point. Physics, math, hardcore sciences were very much what a lot of people were fascinated by. We were, I'd say, right at the beginning of that. In some sense, it's hard not to be nostalgic about it. Back then, being a computer nerd meant you were a nerd, nobody really cared what you were doing. It would still have a very good feeling of a community of nerds who were trying to do interesting things. That definitely transitioned when the dot-com bubble came around.
ZIERLER: What kind of interaction did you have with John Preskill as an undergraduate?
BACON: I really only had one interaction that I remember. I did a SURF project in quantum computing, '96, '97, the summer of that. I worked on whether quantum computers could solve NP-completely problems efficiently. We don't think they can, but at that time, it was, "Be optimistic and try it." I remember after my SURF talk, he came up and said, "That was a hard problem you worked on." [Laugh] Which is kind of an understatement. If I had made progress on that, I would've been Peter Shor part two. He had students I knew who worked with him. I was a Tombrello person, so I worked on Physics 11. I got into research very early at Caltech, which, for me, was fascinating. That was what kept me at Caltech really strongly, learning there was this thing that was very different than coursework that required you to creatively think about things. I got sucked into that. The other thing about that time I have to bring up is, I lived in Ruddock House, and in a lot of the dorms, people would write their names behind the mirror. My neighbor, who I was good friends with, we pulled off a mirror and looked behind it and saw the name Peter Shor. At the time, Shor had just come up with his algorithm. I knew his name, I was super excited about it because I always heard and knew about quantum computing as it came out, but there was this very famous physics computation conference, and I printed out all the papers and read all of them. They were all physics and computation, and they were really fascinating. I remember being super fascinated by that. I knew Peter Shor's algorithm. I was like, "It's got to be the same guy." When I first met him, I went up to him and said, "Have you ever written your name behind a mirror?" He was like, "Uh…" Like, "What is this random dude asking me about?" [Laugh] I described, "It's in Ruddock House," and he was like, "Oh, that must have been me." [Laugh] It was, indeed, right there behind it. That was one of my early Caltech connections I always find fascinating.
ZIERLER: As you mentioned, it was clear that the term quantum computing was something you heard and thought about as an undergraduate. Was your sense that quantum information was a field of study? Was that an area that you could think about? Or was it still too early in the game for that?
BACON: You could think about it. I think one of the things that was occurring when I was an undergraduate that I didn't know a ton about was error correction. That research really was ongoing, and John's group was in the middle of that. His SURF students worked on that. I remember they were looking at qudit codes. But it was only when I went to Berkeley for my PhD that I realized error correction was extremely important and really demonstrated that quantum is not just analogue computing. There was sort of an early generation of graduate students, Michael Nielsen, Chris Fuches, people like that, and I associate with them. At the time, IBM was sort of the center of quantum information and quantum computing. As an undergraduate, I don't remember there really being an established thing of quantum information. I definitely remember thinking about it as just quantum computation. And there's this whole tradition going back to foundations and Bell inequalities, which I'd always been also deeply fascinated by and knew a ton about. I sort of linked it up in my mind with that. This was coming out of this long tradition saying, "Quantum looks weird. [Laugh] Maybe it's not just weird but useful." It's a pretty natural thing to ask. In retrospect, it seems silly, but it took us so long to ask that question. But definitely, when you hear it, it's like, "Of course, that's true. That's what Bell's theorem says. It says it doesn't behave like classical information processing."
That was my Caltech. And my SURF was on quantum computing. It was with Nicholas Cerf and Chris Adami, two post-docs who were under Steve Koonin. I showed up at Caltech, I'm a new kid straight from the sticks, I'm walking around campus, and I see this kind of short guy talking to this very tall guy. And the short guy looks just like the guy from Honey, I Shrunk the Kids, Rick Moranis. I'm like, "I wonder if that's Rick Moranis." I felt like I should ask for an autograph because I really did think it was him. Later on, I realized that was Steve Koonin talking to Jeff Kimble, very famous Caltech professors. I was like, "I'm glad I didn't interrupt their conversation." [Laugh] They were probably talking about something secret because they're both JASONs. Probably something national security-related. [Laugh]
ZIERLER: As an undergrad, did you have a sense, in working with post-docs and seeing what the graduate students were doing, that this was a research community that had a momentum that would build into what would become the IQI?
BACON: I certainly knew that John was at the middle of a lot of this. John's historical interests had to do with the black hole information paradox. As he said, he didn't understand information, so he needed to really try to understand it. That part of the story, I sort of knew. I didn't know Michael Nielsen, and I don't think I knew people from his group at that time. I think it was only when I went to Berkeley, which was when he gave his first class in quantum information. And his notes were incredibly influential. It sort of was the physicists' version of how to learn it because he's very much a physicist. He had sort of these quantum field theory notes, which are also incredible. That was, for me, a big transformation, getting ahold of those notes and learning them. But I don't remember a momentum there, I just remember Caltech seemed like a great place to be doing this because it was a field that sat on multiple disciplines, which Caltech traditionally has been strong in. A defining characteristic is that the physics faculty would show up and talk to the biologists, and some crazy thing would come out of that. For me, it felt very natural, but I don't remember the momentum yet at that time.
ZIERLER: When you started to think about graduate school programs, was it quantum information?
BACON: No. Like all undergraduates at that time, you were groomed to do string theory or head in that direction, high-energy physics of some form. If you looked at my trajectory, that might've been what my admissions essay said. But I got very interested in astrophysics, so I applied thinking I wanted to go somewhere that was very strong in astronomy. Part of it was more cosmology, high energies out there in space we can learn some fundamental physics by looking at. That sort of thing. I was lucky that at Caltech, I took enough of the courses and graduate courses that I didn't have to take the first year of Berkeley, so I just took all astrophysics. About a year after that, there was a quantum computing researcher who was doing some things related to error correction. I met with him, and he sort of explained to me error correction, and for the first time, I sort of understood why you would need it. That's how I got sucked back into quantum computing. It had always been there, but when I applied, it wasn't to go to quantum computing, I was going to do astrophysics. Then, it just so happened that the field started to exist. [Laugh] I looked around and remember thinking, "All these astronomers are super smart. They're working on problems they've been working on for decades. Here's this new field, and there aren't very many people in it. They're all smart, but it's new. I should work in the thing that's new because that's what research is." I definitely remember having that discussion in my head and being like, "This field looks like a place where there's a lot of low-hanging fruit to work on right now."
Zeroing In On Quantum Error Correction
ZIERLER: Who was the researcher who opened you up to the importance of quantum error correction?
BACON: That was Daniel Lidar, who I used to call my post-doc because we worked really closely together. We'd work on ideas, produce papers. I had a great time. He was a post-doc of Birgitta Whaley, who still works in quantum computing as a chemist. I moved to the chemistry building but still was a physics student. Daniel was the one who really explained error correction to me. The one bad thing about Daniel is he changed his name. Right before I met him, his name was Hamburger. He was Israeli. He changed it when he moved to the US. We would've had papers that were Bacon-Hamburger. It would've been epic. In fact, he told me that if I came back to quantum computing, we could write a paper, and he'd change his name back. But ever since he said that, it seems to have disappeared a little bit. [Laugh]
ZIERLER: What can we read into the fact that Whaley is a chemist interested in quantum computers?
BACON: She was a physical chemist, so a lot of her work is in simulating these systems, tight binding calculations. I think she definitely understood the power of if we could simulate things. It would be obvious, in some sense. But she told me once she was always trying to get into a physics department, always on that boundary of the physics world. She was a great advisor in the sense that she would let people go off in directions. Daniel was the one who had previously worked on inverse scattering problems. When he started working on quantum computing and got hired over, he really did start that direction. She was one of those good advisors who tended to say, "That's great, let's go that direction." I always felt very lucky. Everybody needs a different type of advisor, but I think I needed one who would let me go off and explore. [Laugh]
ZIERLER: Was there anyone at Berkeley in physics or computer science who was thinking about quantum information?
BACON: Not really in physics. There were people a little bit on the periphery, some experimentalists, Dan Stamper-Kurn. In computer science, there was Umesh Vazirani, who wrote one of the earliest rigorous definitions of what a quantum computer is and showed one of the first super polynomial speed-ups, one of the first speed-ups showing that quantum computers really do offer advantage. He was over in the computer science department. There wasn't a tight connection between those two groups, but there were seminars and talks, people I met through those. As I was leaving, Scott Aaronson was starting as a graduate student there, so I overlapped a little bit with him. That was, of course, a fascinating time. [Laugh]
ZIERLER: Were there any administrative challenges being a physics grad student with a chemist as your advisor?
BACON: A little bit. The same things that always happen. The funniest thing was after they did sort of the one main presentation of your work. I think I already had my offer from IQI to become a post-doc when I did it. But the committee chair, Eugene Commins, said to me afterwards, "This is all good work, but we want to make sure you continue to do physics and not this computer thing." That was very much the old-school, traditional physicists, "This newfangled computer thing." I was going to do what I was going to do, it didn't really have any influence. I kind of think about it as a validation in the sense that I was doing something he didn't think was useful, whereas I'm like, "Computers are the future, dude." There was a little bit of a push in that direction. But besides that, it was actually pretty seamless. Berkeley's Chemistry Department is excellent, and Berkeley is pretty good about not having terrible barriers between places. I still think one of the key factors in a good institution is having low barriers between departments. Lawrence Berkeley Lab was up on the hill, so it was kind of assumed that you could do things with other affiliations than you were doing. I don't remember anything crazy like that, besides that one comment, which I found humorous. [Laugh]
ZIERLER: What were the big research questions that defined your thesis research?
BACON: My thesis research was on what people call decoherence-free subspaces and decoherence-free subsystems. I would say the main result in there is this discovery that you can quantum compute using just an exchange interaction. This idea that usually, you have single-qubit gates and two-qubit gates, and you have a universal set of operations you can do. But it turns out that if you just have this one exchange interaction, you can do quantum computation by working in an encoded basis. That was not previously known, and that still lives on today. HRL's effort in quantum computing uses exchange-based qubits. You take three spins and build one encoded qubit out of it. When you do that, in that encoded space, you use this exchange interaction to do your quantum computation. This is, of course, very much like error correction in the sense that you're encoded into a space and working on that encoded space. You take multiple qubits and work an encoded space spread across those.
The things we were working on were called decoherence-free subspaces, which meant that sometimes, the interactions of your environment with your system during error correction, the errors have so much symmetry that they're sort of preserved spaces that are not touched by that symmetry. What we looked at was, if every qubit interacts identically with coupling to some environment, then there are these subspaces that are preserved, they're decoherence-free. In practice, this has been actually seen. In that same thesis defense, I literally had all the theory up to a point, and then there was one slide, which was the experiment. One of the physicists on the committee said, "This is all good in theory, but what about in practice?" And literally, my next slide was this first demonstration of a decoherence-free subspace. [Laugh] I kind of lucked out on that. But the thesis was mostly about symmetry in quantum computers, how you can exploit it for these decoherence-free subspaces, for building different error-encoded qubits.
Also, the end of my thesis, kind of the most interesting one, was about this idea of self-correction. Can you build quantum computers that can correct their own errors? This relates to the stuff that has a strong tie to Caltech, which is topological quantum computing. I was sort of trying to build versions of that which were even more robust than the stuff that Kitaev did. I was trying to build very, very robust quantum systems, self-correcting ones. I wasn't able to do it in that thesis, but I at least started laying out a vision of that direction.
ZIERLER: 20 years-plus, what aspects in thinking about decoherence in quantum computers are resolved, and what are as fresh as when you first started to confront these problems?
BACON: There are still a lot of fresh things. For example, when I got to Caltech, I wrote a paper that ended up having an error-correcting code that people later named after me, which is embarrassing. [Laugh] But that code was discovered many years ago, 2001-ish. Only recently, my friend, Ken Brown, and Roger Calderbank discovered that if you look at the four-by-four version of that, it actually can have decoherence-free subspaces in it. This is literally my thesis. I should've seen this, but nobody saw it. Error correction has a lot of stuff like that I think will continue to happen. But it's the nature of research that you always think you've solved the problems, and you haven't. I do think fields get fossilized. Error correction did get a little bit when it got obsessed with threshold calculations. The threshold is this number about your error in your quantum computer. Basically, if your errors are below one of these thresholds, the error correction works and makes things better. If you're above the threshold and run the algorithm, it makes things worse.
The threshold is very important because you want it to be very high. You want it to be able to tolerate as many errors as possible. If you could get 30% error for every cycle, if you could tolerate that and do quantum computation, that'd be awesome. But the field did get sort of obsessed with that number as the only thing. And I don't think that's true. Like any complex problem, there are a lot of reasons that error-correcting codes and schemas will work in the real world. That's not the only important number. I think that field has matured, but it's entering the new phase where they're actually having to do the error correction in the real world, and that's going to change everything because it's always theory until it's the real world. All of a sudden, you realize there are things we didn't think of, things we didn't understand, different alternative approaches. Other times, they'll seem simple. It'll be like, "Oh, just do this." But those are the best. You're like, "It's just simple, and nobody thought of it. Now, we have it, and things are better."
ZIERLER: Even before you started to think about post-doc opportunities, did the creation of the IQI register with you? Was that a big deal in the field?
BACON: I do remember the IQI being founded. John also would always invite tons of people to go talk. I do think at the time, IBM was the center of theoretical. Charlie Bennett and David DiVincenzo, that group was considered the place to be for anything in quantum information. And there really was a very quantum information-y feel to all that, even though some of it was quantum computation. There was a ton of stuff coming out of teleportation, measures of entanglement, things like that very much centered over there. But I do remember as a graduate student, I definitely visited Caltech multiple times. What I most remember was visiting Michael Nielsen, who was in the process, I think, of writing his textbook. And I think Michael explained to me the Schmidt decomposition. I was like, "That's beautiful." I don't think I'd ever seen that before. He explained it to me. He's one of the more fascinating people in quantum computing. That had a big impact. Maybe I only visited once. I'm not sure.
There's another part of the story, which I think is why I knew more about Caltech's effort than anything else, which was there was a thing called the SQuInT network, Southwest Quantum Information and Technology, started by Ivan Deutsch, Carl Caves, probably Ike Chuang. They started early on having conferences that were a mix across all experiment and theory, and they'd intermingle all these talks, so it was very cross-disciplinary, and Caltech was definitely part of that. David Wineland in Colorado at Boulder was there, so every year, they'd give a talk about the greatest experiment they'd done. It was a really key meeting point for graduate students at the time, and they had summer schools and things like that. I think at least one of the SQuInTs was at Caltech, but I don't remember if I was a post-doc or a graduate student.
ZIERLER: It's a different station in life when you came back to Caltech. What had changed, both for you and where quantum information was at Caltech?
BACON: It was awesome to come back because you're not a student anymore, so you can actually enjoy a lot more. [Laugh] There were two things. One, it was weird to not be an undergraduate at Caltech because Caltech undergraduates are so tight-knit, it's a very strong community. I was no longer part of that community. I probably could've passed as one and had fun with the undergraduates, but it's sort of a different stage in your career. But you also were under the stress of constantly doing homework. I loved my Caltech experience because my goal was to go to a place that would make me understand my limits, and I found them. [Laugh] It turns out you can't take four physics classes at once at Caltech. It's not possible. Your brain explodes, and you don't have enough time in the day to do the problems. But coming back as a post-doc was great because you're now part of this amazing community.
I spent a lot of time while I was there doing what I'd call television-watching, which was instead of watching television, I would just choose a random seminar from a field I didn't know and go to it. If you'd go to the synthetic chemistry ones, it'd be like, "I have no clue what any of these words are." [Laugh] But I remember going and learning about simulating the dynamo in the earth. It turns out you do this by taking molten liquid sodium, putting it into a big sphere, putting a propeller in there, and spinning it around to see if it generates a magnetic field, which was crazy. But it was awesome to be back for the good parts of Caltech, lots of crazy, interesting science going on. When I started, we didn't have a ton of post-docs. I think Kitaev had moved to Caltech. I don't remember the exact date he was there. He was my neighbor. Sean Hallgren and I shared an office.
The thing that was incredible when you showed up at the IQI is, John literally started inviting everybody in the field to come visit. There was just this train of people. A lot of post-docs end up spending a ton of time on the road, traveling, giving talks. In some sense, that's good for your career, getting out there into the field. But I kind of got spoiled and didn't have to travel very much because it all just came right through the door. In retrospect, having seen a lot of things, there aren't a lot of places that achieve that. KITP brings in a lot of programs. It's like on that level, where you just get these huge groups, but it was like all of quantum computing was flowing through that place, and I was in the main place we'd have the group meetings. It felt like everybody would come in and give my talk right outside my office. It was perfect. [Laugh]
ZIERLER: What were the big ideas that were animating IQI in the early years?
BACON: Definitely topological quantum computing, that was always a big one. Error correction in general. When I look back, there was sort of the Daniel Gottesman era. He wrote a very important, influential thesis on error correction and that history of really deeply understanding error correction, what it can and can't do. And then, these questions about building naturally resilient systems, which was originally Kitaev's main contribution, "These anyon systems can be used to quantum compute, and it seems to be protected." There was a lot of interest in that. I would say in general, people were allowed to go in their own directions. As a post-doc, I don't remember having any conversation with John about what I should work on at all, actually. Except to the extent that you'd talk to him about what you were working on, and he'd say, "Maybe think about this," that type of thing. It was not prescriptive at all. We were all given a lot of freedom to do things.
A good example of this was Guifré Vidal, who was a post-doc as well. Guifré had done a bunch of important early work. One day, he gave his group meeting, which was kind of hilarious. He started off and said, "I have been working on quantum entanglement. Then, I thought, 'Why am I working on quantum entanglement?' I realized I'm working on quantum entanglement because my advisor, who's very smart, worked on quantum entanglement." [Laugh] He was like, "That does not seem like a good reason." I'm like, "Yeah, that's not a good reason." His story was, "Then, I thought, 'Why would you study quantum entanglement? What use does it have?'"
Then, he proceeded to describe this idea about the relationship between how much entanglement system has, how hard it is to simulate. This eventually turns into sort of tensor networks. He basically rediscovered methods that had been known before but recast them in a way that was totally novel. There were things like that, which just occurred, and of course, people got very interested. It's like, "Oh, here's this new tool, tensor networks, that we can use in all these different amazing ways." That animated a lot of us because we were like, "That's pretty crazy and interesting." [Laugh] A lot of stuff I remember from that time is stuff that was just coming out of the air from wherever research stuff comes from.
ZIERLER: In terms of the exuberance at the time, maybe even some naivete, were people thinking that this jumble of theories was going to lead to an actual quantum computer? Or was that way too far afield, thinking about the engineering and the condensed matter side of things?
BACON: I think, in many ways, at Caltech, it was very pure research just trying to understand these things. I call error correction the shield in quantum computing. In that era in particular, you'd go talk to some group in a physics lab, and they would have a conversation with you, and they'd be skeptical about quantum computing. "It's not possible." You'd sort of probe and ask why they thought that, and sometimes they had reasonably good answers, and most of the time, they didn't have. But then, you'd also approach them about whether they knew error correction or not. If they didn't know error correction, we'd started to ignore them. It really is a transformation of your worldview, this idea that you can take these noisy components and put them together to build a robust machine. That's an amazing insight.
A lot of it wasn't really focused on how long or hard it's going to be, but just convincing people it was possible at all. For many years, people didn't know the story of error correction, so there was this huge gap in understanding why a bunch of us are optimistic on a 20- or 50-year time scale. We weren't really in the depths of, "Can we do it?" Jeff Kimble had been doing cavity QED-type stuff. There were always experimental people coming through and talking. A lot of that got mixed up with the SQuInT network, which had a very good view of that, so most of us would go to the SQuInT as well, and that would help us see the experiment. But we weren't obsessed with building or thinking about that. We were just like, "There are a bunch of really hard, interesting problems here. Let's work on them."
ZIERLER: What was the research culture like? Was it post-docs working with post-docs? Where were the interactions with the professors and the graduate students? How did all of that work together?
BACON: A lot of post-docs working with post-docs, post-docs working with visitors, graduate students working with post-docs. I worked with Ben Toner on some very foundational questions. I would say the research culture was shaped by the infamous group meetings we'd have that would last a very long time. John would come in, and he had his nice, comfy chair. My office was right behind him. I would sit behind John. I had the best seat because I didn't really have to pay attention. He couldn't really look back. He'd really have to turn to see me. John's a huge baseball fan, and at the time, Barry Bonds was clobbering homers and things like this. My job was often to go in and report on the scores from the baseball games that were on. I think just recently, I saw that somebody asked him what career he would take instead of physics, and he said a stats guy for baseball. If he could live his dream job, it would be that.
But that group meeting had this amazing culture where everybody would go around and talk about what they were working on. Sometimes, this was pretty bad because people like myself or Andrew Landahl, who's now at Sandia in New Mexico, who would talk a long time. Everybody afterward would feel obliged to say a lot, and these meetings would just go on for a long time. Part of it was keeping John up to date, part of it was sharing the knowledge. It'd be interesting to hear from the graduate students. As a post-doc, I never felt pressure, I just felt like if I had something interesting to share, I would share it, then I would get critiqued, or somebody would get very interested in it afterwards. Of course, we also had people like Kitaev there. Kitaev wouldn't come every time, but he'd come, and every once in a while, Kitaev would rip off one of these, "Well, I've been thinking about quantum gravity," and then he would describe some amazing new thing that people would be like, "Where did that come from?" [Laugh]
ZIERLER: Where do you see any intellectual links between your work in decoherence in graduate school and what you were doing at the IQI?
BACON: The decoherence stuff, one of the things that did happen is, I was working on what are called subsystem codes, and these are different types of codes. It turned out that in my thesis, I had found a code that was very interesting. I didn't really know how to describe it as an error-correcting code. While I was at Caltech, people sort of realized that these subsystem codes could be used as error-correcting codes, but they didn't have good examples of them. I sort of had this example in my thesis that I knew explicitly was one of these codes. The decoherence-free stuff was all about symmetries, but the error-correcting stuff was very similar in nature, so there was a very sort of natural overlap there.
And then, it was true that I was just trying to think about how to build physical systems that were error-correcting, and that's like topological quantum computing. What I was trying to do was build versions of this which were stronger than that. And John had written a paper while I was in grad school about the four-dimensional Toric code. If you had four space time dimensions, it turns out quantum computing would be way easier in the sense that there's a physical system that exists in four spatial dimensions that preserves quantum information extremely robustly at a temperature much higher than the energy gap of your system. The way to think about it, in some ways, is it's like when you store information in a classical system in magnetic domains, you have a bunch of spins pointing up or down. If you flip a spin, the neighbors sort of feel that it's oriented wrong, and it costs them energy to do that. If you do this in one dimension, it's not very good because if you flip a spin, it only has two neighbors. And they're sort of like, "That's OK, it only costs a little bit of energy."
You can flip the next one, and it only costs a little bit to do it. It doesn't have an amount of energy that scales like a perimeter. In one dimension, you can't really store classical information very well. You can, but it's really challenging. And the same thing's true in quantum, but it turns out that in quantum, you just have to double everything. Instead of one versus two dimensions, it's two where it's hard but possible, like the Toric code, and four, where it's easy. We were really trying to find systems in lower dimensional spaces that did this. Could we get it to work in 3D? And we still don't know the answer to this. There have actually been kind of fascinating codes that people have discovered that somewhat work in three dimensions, but it's still sort of a big open question in the field.
ZIERLER: I'll ask a question about the job market at that time that will tie together several of your affiliations after that, the Santa Fe Institute, University of Washington. Were you thinking about tenure-track faculty positions, but quantum information was just so far out there that it wasn't really yet a field institutes were hiring in?
BACON: Yeah, it was brutal. We used to always joke that Daniel Gottesman, who ended up going to the Perimeter Institute, was the canary because he had written a thesis that was very transformational, incredible work as a graduate student. And he was a physicist and he was trying to get jobs in physics departments. All the post-docs who were there were trying very hard to get good jobs and having a lot of struggle. Some of it had strong connections back to Europe, where there was a better job market, I think, but it was still not particularly great. The Perimeter Institute in Waterloo had started by the time I was looking, so it was like, "You can get a job up in the cold, frigid North," and some of my friends did that. Debbie Leung, who's still there, complains about the temperature every May when it's super cold in Waterloo. She's like, "It's May, why is it snowing?" [Laugh]
But it was pretty brutal. And part of it was this underlying skepticism about quantum computing. Because in some ways, quantum computing does tie back to foundations questions. You're sort of taught, "Don't worry about quantum mechanics, it works." And it was really hard to convince people, "Yes, it works, and there are these implications." There was a very deep sort of skepticism when you'd go talk to physics departments, which is where I was looking for jobs mostly, about quantum computing as viable. They thought it would never happen. People would say things like, "Quantum computing's just about two-level systems." We'd get that comment from string theorists and people like that. There was a big string theory boom at the time, so that was still very popular and would get you a lot of jobs. They were very dismissive of quantum computing. Of course, that shifted over the years, thanks in part to John and Patrick Hayden, who was one of the post-docs there, who really went off in this direction. Now, quantum computing is one of the hottest places in the more theoretical-minded part of the world for high-energy physics. That's just been awesome to watch because it's kind of like, "We told you so. Just a couple decades ago, you guys were so dismissive." But the job market was really brutal.
ZIERLER: What was happening at the Santa Fe Institute at that point? Was anybody doing quantum information there?
BACON: Cris Moore was there. It's kind of fascinating, the story about error correction actually goes back a long ways. Claude Shannon is involved in it, but also von Neumann. In fact, while von Neumann was visiting Caltech, he would drive out to Los Alamos from Princeton, which is nuts, but apparently, he also went out to California. He was at Caltech, and he wrote a set of lecture notes. I think, actually, they were maybe done after he passed away and were published. But these notes are basically about how to do fault-tolerant classical computing, basically the classical story of this. This question about how to deal with noise and build resilient systems has always had a tradition at the Santa Fe Institute, which was a place that was looking at complexity, but also early cellular automata work and thinking about robustly storing information in physical systems. I think that appealed when I applied to them. That was sort of why it was more interesting. I was working at the time on hidden subgroup problems, algorithms, so I had a real natural internal advocate for me. He wasn't my official mentor. My official mentor was actually Murray Gell-Mann, which is kind of fascinating. I loved being at the Santa Fe Institute because I got to listen to Murray tell stories, which was kind of like listening to Feynman tell stories. He was there, and he had a story about everything.
ZIERLER: Did you hear stories about his frustrations with Caltech at that point?
BACON: Oh, definitely. I told him this once in a while, I was a Tombrello kid at Caltech. Tombrello gave me a book about complexity and the Santa Fe Institute. My uncle lived in Santa Fe, so I knew about all this work in artificial life, so I was very fascinated by that. But I knew that he had left Caltech and was not particularly happy. His parking space was always empty at Caltech. I would park my beater car in his spot because it was right beside Tombrello's office, and I never got called on it. I told him that once, and he was not particularly happy with me. I don't think he ever used it or anything, but I was like, "I've been using your parking spot because I know you're not around." [Laugh]
ZIERLER: I know Gell-Mann was famously dismissive of areas of physics that he did not consider "real physics". Did you ever engage with him on quantum information and what he thought about that?
BACON: I did. He had sort of his own interpretation of quantum mechanics, which he would strongly advocate for. With him, the goal was always to agree with him about that. But he wasn't skeptical of quantum computing. I never got any feeling he was skeptical about it. I think if I had told him that Feynman invented the field, maybe it would've set him off. [Laugh] But because his work was very much at the intersection of thinking about information and how it flows in physical systems, I think for him, it was kind of a natural thing, and I suspect a younger Murray would've been in the field and pushing for it. He definitely was of that mind. The other great story from the Santa Fe Institute was Cormac McCarthy, who was the author-in-residence, was my neighbor, and the great thing about Cormac is he was a good friend of Murray, so he knows a ton of physics.
His novels are these very brutal, dark stories, but he's a very sweet man. I remember a conversation once where we were talking with somebody about Bell inequalities, and the guy didn't understand Bell inequalities, fundamentally didn't understand there's a difference between classical correlation and these quantum correlations. He missed the main point. The guy was going on, and Cormac was there listening, patiently letting the person go on and on, and at the end, he sort of said, "Actually, what it means is this." Afterwards, Cormac came up to me and was like, "I can't believe you let that guy sit there and say all that. He did not understand Bell inequalities." It was like, "Oh my God, Cormac understands Bell inequalities." That's definitely a Santa Fe Institute thing. [Laugh] This amazing novelist has a deeper understanding than probably 99.99% of the world about this foundational quantum phenomenon.
ZIERLER: The series of positions at the University of Washington, was your sense that there were people there who were pushing for quantum information to get bigger, and you were first in, but there was no real larger emphasis at the institutional level?
BACON: Yeah, right before I went to the Santa Fe Institute, my father passed away. It became increasingly clear that I needed to get closer to my mom. Pretty soon after, I started looking for positions on the West Coast. I'd collaborated and worked with Ike Chuang, and Ike had worked with Mark Oskin. Mark's a computer architect who had done some very early, way-before-it-should've-been-done work on how to actually architect these things, what a computer architecture looks like for a quantum computer. Mark was very interested in quantum computing and was able to scrounge up some money to get a scientist position. I applied for grants and got them, so I became research faculty. The really fascinating thing at that time was, there was essentially no junior faculty in the US.
Sean Hallgren, I think, was the only other person in theory doing algorithms. But because nobody was getting jobs, a lot of people were going to Canada or to Europe. Because of that, it turned out that getting grants was actually a pretty easy because there were a bunch of tenured people, then a couple junior people. You clearly were getting the benefit because you were junior, so it actually wasn't hard to raise money for the years I was at UW. It was just challenging to convince them that I should be tenure-track and quantum computing was something they should invest in. It just seemed so far out in the future, especially in computer science, where literally Google is exploding, there are all these other fascinating things going on that are so short-term. "Why would we think about quantum, which is clearly decades away?" I think they regret that a little bit now. [Laugh]
ZIERLER: Now, at UW, I assume in the computer science program, they were wondering why you were not in physics, and in physics, they were thinking, "This is computer science."
BACON: That's right. And it's a valid critique, actually. There's a lecture by Feynman that I've never been able to track down, I think he gave it at the University of Chicago, where he talked about the danger of cross-disciplinary work. He was like, "As a biophysicist, you give a talk to the biologists, and they're like, 'Oh, that person knows so much physics,' then you go over to the physicists, and the reverse happens. They're like, 'He knows so much biology.'" You can really get into this problem that there's nothing actually deep or interesting about it, but it appears to each side like they're doing great in the other field. I think there's actually some truth to that, and I think it's a valid worry. I think there was just a lot of skepticism when I would talk to the physicists about quantum computing. Then, of course, when you really are honest and think about it, you have questions like, "Am I a star? Am I the top person?" That's clearly what they're looking for. And frankly, a computer science star at UW is better than in the physics department. It was a tougher challenge to climb, in that sense. But I actually have no regrets about those times, it was a great faculty to work with.
Seattle as a Quantum Nexus
ZIERLER: Being in Seattle, were you tracking what was going on at places like Amazon or Microsoft? Were they starting to get involved in quantum computing at this point?
BACON: Microsoft had a long tradition starting with Michael Freedman, then Krysta Svore, who was a graduate student around the same time as me, switched over, then there was sort of a group at Microsoft. Microsoft was definitely going on. Amazon wasn't doing anything. One time I looked on LinkedIn, there was one other person, Oliver Downs, in all of the Seattle area who had quantum computing on their profile. He had done some early patent work for D-Wave. That was basically it. There was nothing beyond that. Definitely, at the time, I ran a quantum beer night. Throughout history, starting at Berkeley, we started a quantum beer night, and that was essentially Birgitta Whaley, some of the graduate students, Ken Brown, who's now a professor at Duke, were sort of the core of it. When I went to Caltech, we turned it into quantum margarita night. We'd go to Amigo's, not eat the food, and just have margaritas. We had it back in Seattle, and it started up again when I returned to quantum computing here. It's always been my one social outlet, bring beer night wherever I go. [Laugh]
ZIERLER: Would you visit the IQI during the UW years? Were you staying on top of what was going on there?
BACON: I'm sure I did. I do remember going to a conference at Caltech. It may have been right before I left for Google. It may have been John's birthday party because I remember talking to people who were like, "You're crazy, why are you leaving?" I remember talking to David Poulin, who has passed away but was at Caltech as a post-doc as well. I remember David said to me, "You really don't understand that you actually do good work." I was like, "Well, thank you, David. That's the nicest thing you've ever said." But he was like, "You have Imposter Syndrome." I was like, "Yeah, I know, I do. But I'm going to go try something else." He was like, "You're crazy." [Laugh]
ZIERLER: What were your interests during the UW years? What were you working on, say, 2005 to 2010?
BACON: Still thinking about error correction. I really got deeply involved in this hidden subgroup problem, which is Shor's algorithm works because you can find the period of a function efficiently, and that means that your function has a symmetry, it's periodic, and there's an extension of that. For many years, we were trying to understand when quantum computers could actually solve that. One of the things I was working on was actually finding places we had groups which were non-abelian. Not as simple in some ways that we could solve it, and trying to understand why that was. I worked a lot on that. Then, again, a lot of it was on physical ideas for error correction, coming up with weird techniques for trying to build quantum computers and discovering weird codes along the way. There's a code we found from thinking about–this code has a very weird origin because it originally comes from a paper, a book called Mathematicians in Love by Rudy Rucker. He always writes sort of pseudo-psychedelic science-y stuff. I read the book, and I was like, "This is going to be a great idea." I called up Ken Brown, who was at Georgia Tech, and was like, "Here's an idea from this book. You should read the book." He read it and was like, "You really did get that idea from the book," and we wrote a paper on these weird ways of building codes in classical systems at high temperature. And that led into this other type of code that has a similar property. I was thinking a lot about weird codes that might have interesting properties. That was sort of the main direction, I would say.
ZIERLER: To contrast to what you said earlier about your time as a post-doc at Caltech, when it was all fundamental theory stuff, looking five, ten years down the line, did you get a sense that things were progressing so that a scalable quantum computer became more of a realistic proposition?
BACON: I would say, when you were in the field, it was hard to see the progress. The early days in the 2000s, when going to SQuInT, you would see these talks by Dave Wineland's group in trapped ions, and each one, they'd be building larger computations that seemed to be growing and getting larger. But at the time I was leaving, it did not feel like progress was being made at a rapid pace. I think that's partially because when you're in a field, it's incredibly hard to see progress. I do have this benefit that I left and came back. You just get a better appreciation of the time scale, which we have a hard time comprehending for some reason, and progress happens on five- to ten-year time scales, not by the year, and I could really see it. But I don't remember thinking, "We're on a great pace." There was a very famous road map that was put together by a bunch of smart people, and you go back and read it, it's hilarious. It's basically, "We will have an error-corrected qubit in two years," or something like that. Super optimistic, but not realistic at all. Whenever I hear people complain about quantum hype today, I'm like, "There's always been this optimism because we're all optimists." But very smart people have said really silly things about how fast this will occur. It's probably going to occur today. Very smart people are going to say silly things, and that's OK because that's just being an optimist. Now, there's hype, which is really bad and can be overdone when you're selling things. But on the other hand, this optimism will always be there.
ZIERLER: How did the opportunity at Google come about in 2011?
BACON: Part of the story is, I started working on a collaboration with the people who eventually founded IonQ, Jungsang Kim and Chris Monroe. Part of that was to do architecture design for their modular systems. I was like, "Let's think about the computer architecture-type problems there as well as a hardware development effort to actually try to build these things." As part of that, I went to a conference in Australia. I was going to write some code to start building tools to simulate this. It's a 14-hour flight, so I had, like, three batteries because this was back in the era where there wasn't enough juice in your laptop. I just wrote Python code. On the way back, I did the same thing. And it kind of occurred to me that I really liked that, I really enjoyed writing code. I built a site called SciRate, which is an overlay on the arXiv where people can cite papers.
When I left academia, I shut it down, but people actually rewrote it twice, and it still exists. It's almost exclusively used by quantum computing people, so it was actually a great thing for the quantum computer because I can now just go and look at the popular papers. You go there, and you can see what everybody thinks are the best papers over the last month. As a not-full-time researcher, it's a great tool. I had written that, I wrote some iPhone apps. I wrote an iPhone app that was an overlay for the arXiv for downloading papers and saving them, then I wrote one which was a random number generator. This was the very beginning of iPhone stuff, so my big thing was, you could take your iPhone and shake it, and it had one of those spin-y animations, and it would generate a random number. It turned out, the people who loved that were teachers because they would put their students' names in there and shake it. When they were calling on students, the students loved that. It was so novel at the time.
I'd done all these little projects, and in the CS department at UW, I looked around and thought, "Where are people going on their sabbaticals?" They were all going to Google at the time. I was like, "Oh, that's a sign. It's a good company, there must be fascinating things to work on." I just woke up and studied. I took an algorithm design manual, I read all the algorithms. I didn't do enough coding, so I basically bombed the coding parts of the interviews at Google, and I applied. Somehow, I got an offer. I was like, "OK, we'll try it." [Laugh]
ZIERLER: Was this explicitly a quantum information opportunity?
BACON: No. Nope, totally just a software engineer. Google would always hire generalists then try to place them. When they tried to place me, they basically just had me tour all the former professors who were at the Google Seattle office. They were doing more research-y type work, I'd say, and I actually chose none of them. I wanted to go somewhere I could learn to be a real programmer. [Laugh]
ZIERLER: Did you think you were leaving the field to some degree?
BACON: While I was at Google, I still worked on research problems on my own. I still worked on graph isomorphism and thinking about that. To the day I die, I'll be thinking about random problems like this. That's at my core being. It just became a hobby, not a full-time thing. But it was a transition to a totally different career, which was hard because you're no longer a professor, and all that expertise doesn't matter anymore. What matters is, can you build a system that's going to sell the product?
ZIERLER: The extension to Google getting involved in quantum information stuff, this is total serendipity, you do not see this long-term strategy?
BACON: No, not at all. And in fact, I helped build Google Domains, which is a domain registrar, then I started working on federated learning, distributed machine learning, privacy-preserving machine learning. Google, like I said, had some early efforts, and I sort of helped connect them with people because I'd watched them proliferate for a long time, even when I started, because I just had this other career. I was doing all these other things. Then, eventually, it was just kind of that they finally really did need software engineers to build parts of their system. I want to say it was seven-dimensional chess, but it was just more choosing a company that might somehow do quantum computing.
ZIERLER: Was the Google Seattle office sort of predetermined to be alive to future thinking in the field? Would that have been the place within the larger Google infrastructure that would've caught on to the quantum information revolution?
BACON: The Seattle office itself, I was working in research, and there were some people doing stuff there, but it wasn't quantum. The quantum stuff really did originate from Hartmut Neven, who had worked in Computer Vision and was one of the original people in Google Glass. He had sold a company to Google. He was down in the LA office. Really, at Google, a lot of efforts that are far-reaching come from directors or very senior people deciding, "This is something Google should be doing. We should be pushing on it." The key at Google is always, you don't want to get too big too fast. You don't want VPs to be keyed into what you're doing, really, until you've done it. Hartmut was at the right level where he could start a program and keep it small, build it up, and start to gain expertise in the quantum world. But it was always a real risk within the company, like, "Why would Google be doing this? Do we have any support from this?" Luckily, especially in the early days of Google, what mattered most was what Larry and Sergei thought. It turns out that Sergei had actually gone and talked to Ike Chuang at Stanford when he was looking for graduate student projects, and Ike, I think, gave him a project and didn't see him for some months, when they started working on PageRank.
And I think somebody asked him at one point, "Why did you do that?" Sergei said something like, "I looked at quantum computing, and it just seemed too hard to build right now," which is a good answer for the 1990s, when he was doing it. That was a great answer to not go into a field, "I don't think it's practical yet." Because of the errors. He didn't say it because it was on a bad model, it was because of error correction, how hard it is. Fundamentally, I think Sergei has always been a huge supporter of quantum computing. That helped, Hartmut had an ally to support him. Now, I think we're at the point where the team has done this major milestone beyond classical computation, so there's a lot of internal support because that's seen as good press, but also a new step in the quantum computing era.
Google Embraces the Quantum Revolution
ZIERLER: Do you have a specific memory of when and why Google jumped into quantum when it did?
BACON: No, the early stuff, they really did want to understand whether D-Wave's machine gave any advantage, and they developed a lot of expertise in optimization. They were really investing in that. When Hartmut started, he was new to quantum computing, and I sort of watched him learn the field and become a first-rate researcher in it. It's kind of crazy to be a director doing all this other stuff, then also mastering a pretty challenging field. I think he fundamentally believes this is a new field that we need to be on top of. The main story is, if it gives any advantage for optimization or artificial intelligence, then Google should be doing it. But the real big transition was the decision to hire John Martinis. That was definitely a transformative event because that was a transition. What's interesting is, the early days of that really were about building a D-Wave-like machine, then John Martinis clearly pivoted back to the more traditional approach.
But it was clear there was this experiment where they'd done classical error correction, shown things that were just below the threshold for error correction. In some sense, that was a pretty clear marker, where it was like, "The individual components look good enough now, and we just need to put it together." And that sounds like a Google thing to do. "Now, we engineer it." It made a lot of sense and built on a lot of the hardware in the field. This transmon qubit was a huge change in how people built superconducting circuits. There were these events that occurred that, without them, we would not be here.
ZIERLER: What did this transition mean for you? Were you holding up a big sign that said, "Hey, I'm a quantum guy, I'm already here. This is perfect"?
BACON: Like I said, I'd always kept in contact with Hartmut. What they really wanted to do was build a quantum-computing service so people could access the machine from the outside. That's sort of a very normal computer engineering-type thing to do. I just happened to be the right sort of person for that. In many ways, quantum computing is still my home. I have huge numbers of friends and people who I care about and care about me who have been through the experience of this challenge of being in an outsider field. I still was very deeply coupled to the field in the sense that I have a reputation, so Hartmut was very excited. It was and still is extremely hard to find people who know how to build and engineer software systems and also can do quantum computing and quantum physics. That's not a normal background for most people. I think Hartmut was super excited because I had both of those, and he was willing to have me stay in Seattle, which was my one condition. [Laugh] I was basically remote working on this. I really did just luck out. But I do think having multiple superpowers is a key thing for the new era. It's really hard to find people who can both build good software and be able to tell you about basic quantum computing, deeply understand the algorithms and the theory behind it.
ZIERLER: Moving closer to the present, tell me about your decision to join IonQ. Did you leave Google, or was it a leave of absence?
BACON: I officially left Google. It wasn't because I didn't like the Google team. I actually really liked the Google team. I'm back, so that kind of shows it. But I had known Jungsang Kim and Chris Monroe from my previous work with them. An advisor on that team is Ken Brown, who's one of my good friends. He is currently a professor at Duke. He's a founder of quantum beer night. They had tried to recruit me about a year before I went to Seattle. It was just hard to figure out how to make that work financially for startups because you're like, "I'm taking a big risk." When they approached me the second time, they were clearly about ready to go public and do this reverse merger, and I would get to participate in that. A lot of the motivation was to see that process, and frankly, to see another quantum computer. IonQ was building another quantum computer with another quantum operating system with all different challenges than the superconducting world. I knew I'd grow by seeing that new system. Google had invested in IonQ, so I'm still helping Alphabet make a lot of money. [Laugh] I really was there to participate in that and see that through. It was definitely a fascinating experience to see what investors thought of quantum computing, why they were investing now, having those conversations to try to understand where they are and what they were thinking about. That was the IonQ adventure. [Laugh]
ZIERLER: IonQ had gone public before you joined?
BACON: Basically, when they approached me, they were like, "We were thinking about doing this reverse merger, this SPAC, and we want you to join our team to be a part of that." Part of it is, Chris Monroe is Nobel Prize-worthy trapped-ion quantum computer person. He literally did most of the firsts or bests in quantum computing for many years. Jungsang is more of an engineer and comes from that background. They didn't really have a theory or people connected to error correction of software. I sort of was used as that for their pitching this, "Now we have somebody who does these things." A lot of the team comes from Chris Monroe and Jungsang's labs. They're very tightly coupled to that. They have a very interesting technology. I was sort of brought in as the software person plus helping them do theory, then doing the dog and pony show to talk about why trapped ions are not totally crazy. [Laugh]
ZIERLER: What did you learn about quantum computing in the sense that in the world of investing and hype, there's a lack of patience, or maybe there's an expectation that might not really align with where either the theories or the experiments are in quantum information?
BACON: The interesting thing is, when you do a SPAC, there's a company that's formed that has some amount of money that's trying to do a merger with you, but you also, as part of this, raise money from private investors who are coming in and investing at the same time you're doing this. We spent most of our effort on that, raising the PIPE, which ended up being, like, $300 million. There are a bunch of different types of investors you talk to. Some of them are the hedge funds, which are just being financially savvy. SPACs have weird warrants, so there are a lot of financial engineering-type things going on there. But we tended to try to stay away from that and look for long-term investors. One of the people invested is Breakthrough Energy, which is one of Bill Gates's funds. They're trying to solve climate change, so their time horizons are, "Solve climate change. 50 years." Very long. There's a class of investor that actually doesn't need to be hyped. This is the secret of quantum computing.
When I see a lot of the hype going on for these companies, it just is an indication of who they're talking to. I think the people who are good investors know that this is something you're doing that is ten years, not five years, 20 years. It's an early, early investment. The best people understand that. A lot of the hype is that there's a lot of money that doesn't understand it and just wants to get on the bandwagon of doing this. It leads to the cycle of overhyping, promising too much whereas the best investors understand. And I think about this as Google and Microsoft, they view quantum computing as an investment. Sundar is not doing this out of the generousness of his heart. He's doing it because strategically, he thinks that in ten years, Google will still be around and will need to be a world-class quantum computing company. The longer-term thinking of people out there exists, and that was the pleasant part of the experience. The non-pleasant part was the more, like, hype-y, "We need to get this product moving."
There's a very famous thing called cargo culting. Feynman has a great thing about this. Cargo culting is, like, after the Allies left islands after World War II in the Pacific, the people who lived there would come out and do motions to try to get airplanes to land because they had seen people make these motions, then all these things drop from the sky when planes came, and they associated the movements with the actual thing that's going on. In today's world, there's a lot of cargo culting, which is people going through the motions because something else did it. In quantum computing, there's a lot of making it look like a product when it's not a product yet. That part drove me slightly nuts because IonQ will have to focus now on revenue when really, what they need is what they got, $500 million in the bank, and in six or seven years, they can probably have some reasonably interesting quantum computer they can sell as a product. And that's the tradeoff. The problem is that they make revenue projections that are like, "That's going to be really hard to hit." I worry about that.
ZIERLER: The pull to come back to Google was pretty strong, though, after just a year?
BACON: It was. I really enjoyed the team I worked with. I think it cemented in my head this notion that we need to focus on the technology, not necessarily the product side yet. And I do have good experience with Google, however it thinks about its future going forward, I understand where the investors are thinking about this. I think it's a good perspective to have to see the temperature of what the world is like right now. Of course, interest rates are going up, so now it all changed. [Laugh]
ZIERLER: We talked about your current work right at the beginning of our talk, so for the last part of our discussion, two forward-looking questions to top things off. First, in terms of the things you'll be looking for, if you could extrapolate, where are the big initiatives where, for those people who might be impatient and don't have a timeline where they can conceptualize 10, 20, 30 years in the future, you might point to so you can say, "Here are the benchmarks. Here's where we want to be in X number of years. Judge us on that timeline"? Or is the premise of the question not even appropriate because things really are so fundamental that we don't necessarily have the ability to make those benchmarks because we're simply not there yet?
BACON: I think you could make the benchmarks, and people have tried to make them. I think the real challenge is that quantum computing is a technology with a ton of moving parts. Like any benchmark, it's hard to do one that's holistic enough for what we're looking for. When thinking about the future and progress to judge people by, it's, again, one of these things where I still think we're missing breakthroughs, and we will see them, and they will be very obvious when they happen in the sense that they will tick off all the boxes of what you need for a better qubit system. I suspect these will pop up as we go forward. It's hard because I think my answer really is, we'll know it when we see it. And I think we have enough experience now to know when things are not the breakthrough they seem. When somebody claims, "I've done this thing, and therefore quantum computing is easy," usually, that statement is almost certainly naive. If you said that to anybody who's tried to build a 10- to 100-qubit quantum computer, they would be like, "No, that's not true."
This integrative part of quantum computing is the big challenge. I think, also, there's just a major question in the field about which direction's the right one, not in terms of platforms, but Google is pursuing this brute force approach. "We're going to brute force error correction. We're going to have a large classical compute sitting there helping us do error correction constantly," whereas Microsoft is, "We're going to trying to build a qubit, which may have to do a small amount of error correction but as minimal as possible, because that's really hard, and instead, our qubits are fundamentally better because they use error correction in the construction of the qubit." To me, that's the big open question, which of those two is going to triumph. So far, Microsoft is losing. [Laugh] My good friends at Microsoft will not disagree with that in some sense. But their vision is different. It could be that on the cusp of totally upending the race. I think that's going to be the fascinating, exciting race, are we going to have better-encoded qubits, or are we going to brute force our way forward? I don't know the answer to that, and I suspect in ten years, we'll have a more definitive understanding of which of those two paths is the right one.
ZIERLER: Finally, for you, what are you most optimistic about going forward, and how might that influence or inform what you work on next?
BACON: I'm very optimistic about what I call the middle way, which is ways of building quantum computers that involve Microsoft and Google ideas. I think there's a middle path, which people are starting to explore. I think that's a fascinating era. I'm also optimistic about algorithms. I think the hype level started maybe three years ago around the beyond-classical experiment. What happens with hype bubbles is, they hit, and the early work isn't as good, in some sense. But the first graduate students who started joining because of that hype bubble are now mature graduate students starting to produce results. When I look at the quality of work coming from a lot of smart people doing a lot of interesting algorithms work, I think we're about to shift, and I'm super optimistic about that. I think we're going to have another couple years like post-Shor, where there's a ton of interesting, fascinating stuff.
The invention of error correction occurred during that era, huge amounts of quantum information were done. There was this blossoming of stuff. I suspect we're going to hit a similar thing because there's so much money, and that money is being used to support a vast amount of people. And a lot of it is going in the wrong direction. But when you have so much going on, it generates a lot of excellent work. I suspect we don't quite value how much brains are being dumped yet again into quantum computing, and that that's going to pay off, which makes me pretty optimistic. For my own work, it's mostly trying to find really smart people to work with because they're going to be doing fun and interesting things. "Where can I find the people who are going to be doing the stuff we're going to be talking about in another ten years?"
ZIERLER: No matter what, it's been a fun trip and will continue to be so.
BACON: Definitely. The fact that I left and came back has definitely been a wild ride. I never really thought I'd be coming back and talking about quantum computing. [Laugh]
ZIERLER: On that note, it's been a lot of fun speaking with you. I'm so glad we were able to do this for the IQI/IQIM. Thank you so much.
BACON: Thank you very much.