🇨🇦 Michele Mosca & Nick Sullivan — In Conversation
Join Cloudflare's Head of Research, Nick Sullivan, for a special conversation with Michele Mosca, co-founder of the Institute for Quantum Computing at the University of Waterloo, researcher and founding member of the Perimeter Institute for Theoretical Physics, and professor in the department of Combinatorics & Optimization at the University of Waterloo.
Tune in all week for conversations with Canadian tech leaders as we celebrate the opening of Cloudflare's new office in Toronto, Canada! 🇨🇦
Learn more on the Cloudflare Blog:
Transcript (Beta)
Hello and welcome to another episode of Master of Computer Science. I'm very pleased today to welcome Professor Michele Mosca from the University of Waterloo.
Waterloo, Ontario in Canada is very close to Cloudflare's new Canadian office that we're launching today.
And so this is a very special conversation for me because Michele was one of my professors at Waterloo and taught me, I guess, my first class on quantum information theory.
And so this is a class he continues to teach today in one variety or another.
So welcome, Michele. Yeah, it's great to be here.
Yeah, great to reconnect. Nolan, congratulations on setting up shop in Canada.
It's a delight to see that. Yeah, absolutely. So let's get to know you a little bit better.
So how did you get your start in computer science? So a bit of a circuitous way.
So I'm actually my background is in mathematics. And I did CS in first year.
So we have to. I was good at it, but I didn't. I kind of really preferred mathematics.
So I kept working in crypto. And then I met Scott Vanstone, who was sort of championing public key cryptography and elliptic curve crypto.
And he was sort of mentoring me.
I did crypto work terms with him. And I really love that part because it married kind of deep mathematics.
But I really like the algorithmics and programming things and trying to basically break cryptography.
And I didn't realize it at the time, but I was really doing theoretical computer science, right, which is at the interface of math and CS.
So that kind of got me into it.
But, you know, I did my co-op terms involved programming. This is a long time ago.
I was working in beta versions of Windows 3.1 and even got to use some Cray supercomputers at the time.
So, you know, inevitably, I find myself doing theoretical computer science.
So I really like the algorithmics. OK, and so you went to you were an undergrad at the University of Waterloo.
Yeah.
Where did your path take you after graduating? So, you know, at Waterloo, I really like cryptography.
I like discrete mathematics and number theory. And Oxford had this master's program called Mathematics and the Foundations of Computing, Computer Science.
I thought, oh, this is great because it's got the mathematics, but the application area included cryptography and communications, you know.
So I went there and that's when I first, well, I'd heard of quantum cryptography and thought it was like the dumbest thing ever.
And then I was there. Actually, I did hear about quantum computing before I graduated and I thought it was a joke.
Then I was there and I'm doing my thing and working on discrete logs and finite fields and these sorts of things as part of my dissertation project after learning all this great mathematics.
And I'd started learning complexity theory more formally.
And my supervisor saying, you know, the physicists say they can break these discrete log algorithms.
I'm like, oh, whatever. That's just nonsense. And he kind of said, no, well, look at it.
And he went to this workshop. That's when I realized I was wrong.
So I met this very, very fledgling quantum computing community and they were half crazy, but not totally crazy.
And that's when I realized, you know, I didn't understand what quantum theory really meant.
I'd seen it.
I was like, I thought it was just esoteric, esoteric physical frame, you know, physical theory, explain a few things, you know, at the subatomic level, but who cares?
But then I realized, no, it's actually a, it's a paradigm. Like it's a new mathematical paradigm for all of physics.
And I also realized complexity theory has to be framed in a realistic physical framework.
So I'm like, oh, like this is really like in cryptography is all computationally secure cryptos based on things being computationally hard and some realistic physical framework, really.
So this is sort of redefining what was hard and what wasn't and what was possible and what wasn't because now you can actually do information, theoretically secure key agreement through a non-confidential channels and so on.
So it kind of redefined, you know, the foundations of everything I was, I cared about.
So then I started working on quantum algorithms.
I thought, you know, I thought I'd do a bit of quantum algorithms, a bit of quantum crypto, a bit of quantum error correction, but I never got to the crypto error correction.
I was just really busy with the quantum algorithms when I was at Oxford.
So, Chad, I don't know if I've answered your question, but, you know, I just, you know, reluctantly, but I'm definitely, thankfully sort of pivoted, not didn't sort of abandon, but just sort of evolved what I was doing in a highly non-trivial way, let's say.
So what was it about this intersection of mathematics, physics, and computer science that makes quantum computing such an interesting and unique field?
Yeah, because it's, it impacts, you can almost think of it like a stack, right?
You know, in computer science, you might have applications are built on, you know, on platforms that are built on infrastructure, you know, you have this sort of stack mentality, you, well, you can have a similar stack where you have math and computing, figuring out what's efficient, what is it and programming it and all that.
But all the, all the assumptions of what you can do, what you can compute is based on physics.
And the physics is based on the mathematical paradigm for physics, which is quantum theory.
So you have this disruption very low, like this is below infrastructure, right?
This is like the physics out of which, you know, stuff is made, including infrastructure.
So there's this massive ripple effect. Like initially I thought, oh, this is just a nuisance for people who build infrastructure, build technology, you know, build materials, like, no, no, it redefines what's doable, what's possible, what is it, what's efficient, what is it?
So you have this ripple effect, you know, implied by this new mathematical framework for physics, and the mathematics comes in, so it's at the bottom and the top.
So now we have to figure out what's possible, what isn't, what's safe, what isn't.
And, you know, I guess I just found it fascinating to be working at the intersection of these three very broad and deep areas.
But again, it was a very foundational change, and that's what really got me excited about it.
It wasn't some, you know, oh, here's a new theory or one new algorithm.
It wasn't ephemeral at all. It was really a foundational shift in computing.
Yeah, it's fascinating that the very lowest level of how you actually instantiate a computer would reflect itself all the way up the stack to the point where it enables new algorithms in cryptography.
And so what new capabilities does using the underlying methods of quantum physics and the capabilities of quantum physics enable you to do in the field of mathematics?
Yeah, in mathematics.
So to first order, we don't know. We're still just discovering it.
Like, the idea of quantum meets information theory started in the 80s, maybe even the 70s, you could say, but it wasn't until Shor figured out it would break RSA cryptography that, you know, hundreds, you know, initially dozens and eventually hundreds and now thousands of people are thinking really hard about that.
In fact, it's taking off even more in the last couple of years because the hardware is now becoming available and palpable.
So what does it change? Well, we're still discovering it, but obviously it changes some computational number theory problems.
Initially, we thought discrete logs and factoring was super polynomial.
In a quantum world, actually, they're polynomially hard. There's a number of other number theory questions whose complexity changed substantially.
We can do, you know, brute force exhaustive search, not super polynomial more efficiently, but we can do it quadratically more efficiently, which is not as big a speed up, but it's much more broadly applicable.
But in the last few years, people have figured out all sorts of current heuristic, you know, classical heuristics, like backtracking algorithms, which are amenable to quantum speed ups we hadn't known before.
Like rapidly, you know, Markov chain, so random walks, which you can model as Markov chains.
There's a general purpose framework for quantizing them and getting the sort of quadratic speed up.
So it's very broadly applicable in terms of the tools that enables, you know, the applications.
Well, we're still discovering those.
But like I alluded to, we're here in the last 18 months or so, just a greatly intensified effort by industry and others to figure out.
So while we're, you know, while the researchers are figuring out fundamentally what kinds of problems and algorithms and mathematical problems, like can it speed up machine learning?
Well, maybe, right? It can speed up the solution of sparse linear systems under the right conditions, you know.
Okay. So we're developing this tool chest, but, you know, without necessarily knowing what it's good for, but now we're getting more people coming with the hard problems.
And we're trying to map, connect these tools with the real world problems.
So I guess how close are we to seeing real quantum computers intersect with these challenges, these problems that they seem well-suited to solve?
And how, I realize this is very close now, but when you got back, got into this field, did this, how far away did it seem?
Oh yeah. I mean, oh, I remember my master's defense, right? I just, you know, reluctantly wrote a chapter of my master's thesis on quantum computing and then ended up switching to quantum algorithms for my PhD or my doctorate.
And my thesis examiners, one of them asked me, when are we going to have a quantum computer?
I didn't know, but so I didn't want to answer.
So I changed the subject and they're like, okay, now answer the question.
Okay. And they said, you're going to pass just, just, but we want to know what you think.
I said, look, I don't think we're going to have a large quantum computer in the next 20 years.
I think in 20 years, we're going to have about 20 qubits.
And then the path will be clearer because by the time you have 20 qubits, it's a lot to get there.
You'll have learned a lot.
You'll understand the errors better, you know, you know, much clearer picture.
And a lot of people thought I was being really pessimistic. That was more or less spot on, but that was 96.
But from 96 to now, what's the chance of breaking RSA in 20 years?
Zero. That's what I said in 96. But a few years later, I was like, you know, probably, you know, 1%, maybe 10%.
And now I put it at over 80%, right?
To the point where don't even ask 20 years. Now the question is, what's the chance of it happening in 5, 10 or 15 years?
Because we really need to do that risk management now.
Because it will break public key crypto in the Internet and HTTPS, IPsec, VPNs, everything depends on, not everything, but most things depend on this.
So I've been doing my own estimates for many years. I've been public about them the last few years.
So last year and a half ago in Seattle, you were there too, speaking, I said, one in five chance within 10 years.
And the probability that we break RSA 2048.
So very precise. Now, so but what does that mean for quantum machine learning or for the design of new materials or whatever?
Well, that's a rough, so you got a 10% chance of about 4,000 logical qubits or 20 % chance.
So how many qubits do you need for what you care about? Because if it's only 400, well, it'll be a few years before that.
If it's 40,000, it'll be a few years after that, right?
So we don't know is the answer. But it's not like we have no clue.
You get these cynics who make these, you know, unfounded statements about, well, they used to always say, oh, quantum computing, it's always 20 years into the future.
Not true. Like that, that's never, that's not been the consistent statement.
And now they're even changing it to saying it's always 10 years into the future.
Not true. Like anyone who's seriously analyzing it, you'll see that the chance of it happening in 20 years has continually gone up.
The chance of it happening in 10 years has continually gone up.
And now even in five years, like IBM says it's going to achieve a fault-tolerant logical qubit in 2023.
Our own analysis, so we surveyed 44 thought leaders around the world last summer.
It's part of our quantum threat timeline report. And 30 out of 44 said it's about 50% likely or higher that we'll have a fault-tolerant logical qubit within three years.
So three years from last summer. And actually a quarter of them said it was about 50% likely or more likely.
Some of them said over 70, you know, over 70% likely within one year.
So again, I don't think it's going to happen in a year, but it could let, you know, serious people thought it might.
So once that logical qubit is achieved and IBM says like one to 10 or something like that, how long before you scale the 4,000?
We don't know. And people will start saying, oh, let's assume a Moore's law scaling.
Well, why? Like what basis do you have?
You know, we're at, we're not at T equals zero, even you already have a scaling law.
So I, cause you know, you don't need to miniaturize initially. You don't just make a bigger chip.
I'm not saying it's easy, but you know, it's going to be sort of choppy.
They'll probably just, they can print a bigger chip. It may or may not work.
They might need to improve the quality or not. They might want to miniaturize maybe enough qubits fit on a chip.
Probably not. They probably need to build bigger fridges or network the chips or miniaturize the chips.
But so it's, we're not going to get into some Moore's law status for a while, but we don't need to miniaturize it.
It'll just be, we need to scale. So I don't care if it's a big thing.
If you network several fridges or if you build a huge bat, you know there's lots of ways to potentially scale.
We don't know is the answer, but I certainly from a cybersecurity perspective, you don't bet against people figuring out how to build a big fridge, right?
Or how to, you know, again, I don't want to underestimate the challenges.
They're huge, but we've gotten this far and you got tens and tens of billions of dollars allocated to solving these challenges.
So again, I wouldn't bet against that. So I have my own estimates and you don't have to just trust my views.
I've re-interviewed 44 like true, like global thought leaders and got their opinions.
And my views are kind of middle of the pack.
You get some people a little more optimistic, some people a little more pessimistic.
So I'm talking about that's fault-tolerant logical qubits I've been talking about.
That's where you can really take an arbitrary quantum algorithm and build a circuit to implement it.
In the short term, you hear a lot about the so -called NISQ or noisy intermediate scale quantum.
So if you have, if you can build your components, let's say a one in a thousand error rate, very loosely speaking, you can get a circuit, you can get a quantum computation with about a thousand operations in it before it kind of falls apart, right?
Before errors just take over.
So, you know, I call it like a quantum soap bubble. Like, you know, it'll expand, but at some point it'll pop.
And the question is, if you're truly milking, if you're getting the full quantum power of those one over epsilon operations, that's exponential power.
Like you're getting something that would take exponential in one over epsilon effort classically.
So if that's exponential, like two to the 100, you got something there, right?
Like you good luck trying to do that classically or two to the 200, right?
So it might burst at some point, but this whole NISQ era, the people trying to do something useful with NISQ, and they've already done some things which, you know, outperform classical, what we know how to do classically.
It's not something useful probably, but that's normal.
Of course, the first proof of concept demonstration of an advantage is not likely to be something commercially useful, but it validates this outrageous hypothesis that there's this exponential power lurking within the quantum world, right?
Like, you know, when I eventually understood the paradigm, I'm like, yep, it's there.
We just got to build devices to do it. Other people are like, well, I'll believe it when I see it.
So now we're starting to see it, right? Now, is NISQ going to be useful for anything commercial?
Some people and many companies are hoping yes.
Others, you know, I think one application of NISQ is to build a fault tolerant logical qubit.
Get a few hundred of these things, maybe a few thousand, sophisticated error correcting code, and you get one robust qubit.
And there we know how to program these things.
But if you can actually solve your problem with 100 or 200 NISQ, you can do your chemical simulation or simulate your material, you know, test whether some new material is superconducting at a high temperature without fault tolerance.
That's great. Like, we're excited about that.
People are trying. They may or may not succeed. Like, you know, people say there's a lot of hype.
Maybe. I think the hype is hype. The quantum hype is often hype or exaggerated.
So of course, there's a lot of vaporware and exaggeration. So, you know, filter out the smoke.
It might actually work, this NISQ stuff. It might not.
We don't know. I think for many organizations, it's too big. Like, the thing with quantum, like, you know, it's like lightning.
You don't know where it's going to strike.
But when it does strike, it's going to hurt. Like, it's going to be impactful, right?
And so, you know, large enterprises can't just wait and see how quantum might completely disrupt their business, right?
And so they have to investigate and see if NISQ might help.
And if it doesn't, you know what, that effort isn't lost.
Because what are we doing when we try to, when we work with people to see if, you know, quantum computing will change their business?
Well, we find, we understand their problems.
We see how quantum can impact it.
And then we see whether we can compile those algorithms to run on available hardware.
And if they do, well, we've unexpectedly hit the jackpot. If they don't, well, we wait for this fault -tolerant hardware to come.
So I guess the other point I want to make is, so we've got the fault-tolerant qubit.
Again, I think we're going to start in the short number of years, we're going to get the small prototypes.
It's not until you have 70 or so that you can really outperform classical.
And you need about 4,000 to break RSA. And before that, right now, we already have NISQ.
And, you know, even before that, there's been quantum annealers around for a long time.
And so maybe your problem is even, you know, can run on an annealer, which is basically an array of noisy, I don't mean that, not the noise is too high to be useful for fault tolerance, but still low enough that you're doing something useful, right?
And so you're kind of annealing to not the ground state, but something close to the ground state, and you're potentially solving some hard problem.
And there's a lot of people trying hard to do that as well.
So I see sort of three generations of quantum or quantum-like computing platforms.
There's annealers, pioneered in Canada, a D-Wave, the NISQ platforms, which are being developed around the world, including the United States and China and Europe.
And then eventually the fault tolerance platforms. When, you know, again, the milestone or the, you know, the goalpost or what do you call it, the reference point is that one can use as our estimates for when we can break RSA, because then you've got all these experts who know what they're talking about, putting a lot of thought into when that might happen and giving it a risk profile.
You can leverage that and sort of extrapolate and estimate when these other things could become commercially useful.
There's a lot to unpack there, but that's a really great overview of how quantum computing is evolving.
But one thing that you mentioned was that recently something could be proven to be done as a computation on a quantum computer that is faster than anything that we could do in classical computing.
And so that's, I think the term that's been thrown about for that is called quantum supremacy, right?
But as of today, it's not something that's particularly useful and, you know, somewhere down the line and your estimate is less than 10 years, there will be some particularly useful things that will come out of quantum computing.
And I think, you know, proven there's some very plausible assumptions and, you know, it's faster than anything we know how to do classically, or you can do it proven in some sort of black box model, which are just to be a pedantic mathematician about it.
But yeah, I mean, to me, it demonstrates this quantum power.
And yeah, quantum supremacy is the term that is typically used, it's become a bit of a controversial term.
So often, we'll say quantum advantage or something.
But controversial for two reasons. One is a little more academic.
And it's like, well, it's not, it's not better than classical at everything.
So you shouldn't say it's superior, because it kind of connotes that it's just, you know, classical computers are now unnecessary, which is not what it means.
It's not what it intends. So to me, that criticism is a, you know, anyway, it's a valid point.
I'm not so persuaded by it. Others don't like it, because it just connotes racial supremacy.
Like, you know, there's a bit of debate as to whether, you know, that's a valid or you know, how, but I think a large part of the community tries to avoid calling it quantum supremacy, or call it so called quantum supremacy, or put it in quotes or something, for that reason.
But it's a bit unfortunate that I mean, when the term was coined, it was before.
I mean, no, obviously, it's always been a problem.
But it's sort of really flared up. Recently, we've become a lot more sensitive to it.
So but the term kind of stuck at the time. Okay, so after grad school, you went back to Waterloo, and became a professor and helped found the IQC.
So what is the IQC? And why was this something that needed to exist?
Yeah, so while so when I left Canada in 95, it was not a good time. I guess was, you know, people old enough for me, this is economy was, was, you know, economic crisis, or certainly, you know, serious economic issues.
There was a lot they were, you know, early retirements or budget cuts.
And like, a lot of my mentors took early retirement, went got jobs elsewhere, because, you know, we had to, we had to reduce our operating costs at the university.
So I thought, Oh, I'd love to come back to Canada, but I don't think there's any jobs in Canada, for people like me.
But in the late 90s, things started to pick up, right. And they started to invest in new programs and research chairs.
So while I was working on, I accidentally got into this quantum stuff at Oxford, still working in crypto, the center for applied cryptographic research was being founded.
And Bill Tutte, the fellow who kind of drove the cryptanalysis of the phish codes at Bletchley Park, he was the honorary director, who is amazing, because nobody knew he'd worked at Bletchley Park, but it was revealed in the late 90s that he had, and people like Scott Van Sonne, Ron Mullen, and Alfred Menezes, and Doug Stinson founded this new cryptography center.
And they were, they were, you know, part of their master plan was to get somebody working in quantum computing, as Scott Van Sonne used to put it, so they can keep an eye on me.
Right. So I think they recognized, if you really want to have a 21st century applied crypto center, you need to have some quantum stuff, right?
You have quantum cryptography, and you have quantum cryptanalysis.
So they recruited me to come back. It wasn't an easy choice, because, you know, I was having a lot of, I was really enjoying my time at Oxford and had the opportunity to stay longer if I wanted to.
But I wanted, I thought, you know, this is exciting, what they're trying to build here.
And Waterloo is sort of the ideal place for this sort of initiative.
It has a long history of, you know, why not, and doing new things.
We kind of had to, for many reasons, like starting new university, 100 kilometers or 60 miles from Toronto, you had to be innovative somehow.
So I came, but I didn't want to just be me, the lone, I really was trying to recreate the environment I had at Oxford.
And the cryptography center was very supportive and gave me resources to hire postdocs and initial students.
And I started sort of a grassroots effort to hire people in physics, physics and computer science.
And the university, you know, it was a slow grassroots, you know, building trust and confidence and awareness.
And it was great.
And we were making progress. Other departments were willing to allocate some reason if the right person, you know, came along and sort of thing.
And they were willing to cross appoint me and let me teach in whatever department I wanted to.
So I started this sort of grassroots effort with this quantum fledgling quantum computing group, you know, and CNO was willing to if you know, if the right candidate came along, and we did hire Ashwin Nayak, for example.
Then when I also was contacted by the secret project called the perimeter Institute.
So the inventor of the BlackBerry, Mike Lazaridis was setting up this secret at the time, theoretical physics Institute in Waterloo.
And the founding executive director, he was traveling around the world, asking famous physicists, you know, if you had a lot of money, what would you do to make a positive impact in foundational physics?
And one of them said, Well, you should talk to this guy in Waterloo.
So I met this person.
So I joined that rogue effort. So I was already, I already had my own rogue effort to start a quantum computing group at the university.
And then I heard about this other rogue effort to start this theoretical physics Institute.
And I said, Great, you know, I'll help with that too, because a foundational quantum information group there, and super timely, because there's amazing talent, no long term positions in the theory side of quantum computing, we can get stellar people there.
In the meantime, we're universities interested in growing a more applications and implementation oriented effort at the university.
Mike got super excited by this opportunity and said, Look, I'll help you build, you know, a whole Institute dedicated to well, he was kind of funny.
He's like, Well, why don't you hire more people?
I don't have resort, you know, you know, I can't just go and hire, we have to convince people to allocate their existing compliment.
And that's a long, you know, process of trust building. And, you know, academia, basically, so I'll help you.
So Mike offered to give us to match the resources we dedicated to building this quantum computing effort, you would, you know, give $2 for every $1 and so on, or you give one for every two.
So it started to snowball.
So you know, we had the grassroots effort, Mike was sort of driving it from above.
And, and we founded the IQC. And, you know, the mission to be, you know, the world's leading quantum computing center.
And over time, we continue to accumulate resources to really hire great people.
And at the time, you know, people thought we were crazy.
I mean, people thought we were crazy with perimeter.
And then people thought we're crazy setting up a quantum computing Institute, like, what is this stuff?
It's never going to work. And, of course, over the subsequent 10 years, that started to change, right?
The bet clearly was paying off.
So yeah, it was a lot. It was a lot of hard work, you can imagine, because again, back then, it was a rogue effort.
You know, people thought you were crazy, we didn't have all the political buy in.
And, you know, all the examples around the world we have now.
And I think, honestly, to a large extent, we kind of wrote the business plan for a multidisciplinary.
I mean, I don't want to say from scratch, obviously, I was modeling it after what I'd seen at Oxford, help my supervisor do at Oxford, but we kind of helped take it to the next level, in a sense, with more focus on not just the physics side, but more of the computer science and mathematics, and so on as well.
So it's a very balanced, broad Institute. Yeah, it's been a one crazy ride.
That isn't over, that's for sure. And you not only did the theoretical part, there's also experimental parts, you built quantum computers, right?
Yeah, that's another kind of fluky thing. You know, life is just a sequence of unexpected opportunities, and you can't jump at all of them.
But you kind of have to decide, well, okay, is this should I make like the quantum switching to quantum was I very, very reluctant initially, then one day, Jonathan Jones from the chemistry department at Oxford brings me this paper, well, we can do NMR quantum computing.
So we can do quantum computing in NMR, which is decades old technology.
And, you know, I'd seen it, but the first paragraph said something about solving NPR problems and poly time or something.
So I'm like, okay, you know, in the dustbin, but, you know, I wish I'd read it more.
But thanks to Jonathan, we looked at it, and you're putting that aside, because it was probably not the best way to explain it.
It actually made sense what was in that paper.
So I was communicating with Jonathan, and he was telling me what he could do.
I was telling him, like, this is what this would be an interesting thing to demonstrate.
Because actually, a few months earlier, I'd met this famous computer scientist in the States, he was talking about the old stuff I'd been doing.
And he asked me, what are you going to do next? Like, oh, I'm going to work in quantum computing.
And he's like, don't do it, you're going to throw your life away. It's never going to work.
There's so many interesting things for you. Like, and he was doing it from a place of concern.
But then I was like, oh, no, like, and I, but I asked him, like, so do you know, do you know a lot about the physics?
He's like, well, no, but I said, okay.
So, you know, I stuck with my guns there. But I did, because one of the things he told me is, it's just a bunch of mathematicians and computer scientists talking about matrices, and a bunch of physicists playing in the lab, but there's no connection between the two.
So I thought when I met Jonathan, I said, look, we're going to demonstrate the connection between the two.
So I said, tell me what you can do with your cytosine molecule. And actually, I needed one of my physics classmates, or, you know, colleagues are to translate because I didn't understand what he was saying half the time.
But I said, look, you know, I was saying, well, I need a Hadamard gate.
He's like, oh, I could do it with this y pulse, blah, blah.
And then I'm like, hold on, tell me more about this y pulse.
And he told me what a y pulse is. I'm like, all I need is a, you know, pi over two, you know, pi y pulse.
And let me I'll rewrite the software. I don't want you to, because you know, it would have introduced a lot more errors if he'd made his hardware conform to my algorithm.
So I understood what he could do natively, rewrote the algorithm, and we implemented the first quantum algorithm on a cytosine molecule.
And we did a series of things like he, we took the weaknesses of NMR and turned them into strengths, right.
So we did a lot of the early demonstrations, you know, I think we're way beyond that now.
And I suggest people stop doing those proof of those proof of concept demos, because there's a lot more interesting proofs of concept to do now.
But back then, I manually compiled algorithms.
So about a decade ago, I sort of one of the shifts in my research was to work it in, you know, quantum compile automated quantum compiling, because we can't be doing this by hand anymore.
But so I did it in the early days. And then I kind of come back to it in the last 10 years.
So was this actually like a medical nuclear magnetic?
Sorry, NMR is, it's nuclear magnetic resonance. So it's the same idea.
So usually with NMR, it's a test tube in a, you know, you got this giant vat sort of thing with liquid nitrogen, helium, and whatever.
And you put a test tube in this probe, and you lower it in.
And so the liquid helium and all that is to cool the magnet to be superconducting.
So you have the superconducting magnet, really strong magnetic field, you know, several Tesla.
And then so you got this zed or, you know, vertically oriented magnetic field, it creates an effective qubit in your spins in the sample, and you you know, you characterize all sorts of different chemicals and materials that way.
MRI is the same idea, except you got a person, instead of liquid in a test tube.
And you know, the cool thing is, you also often spin the probe to effectively homogenize the electric, the field, it's not gonna be perfectly homogenous.
And that kind of messes your data up a bit.
But if you spin it, it kind of averages out, you don't spin the humans, an MRI, though, that's an old joke.
But so you put the human in the magnet, but it's the same thing, you get this magnetic field.
And you look at, you know, you create this effectively zed orientation, you can start by, you know, every at every, you know, spin one half particle like hydrogen and certain, you know, carbons have a characteristic frequency.
And so you're basically looking at these, you're reading these, these, these oscillations, and inferring what's there and what what density is there.
So it's the same kind of technology. Actually, another cool thing I've worked on is so called algorithmic cooling, which you can use to because in NMR, and many of these experiments, you get your initial sample is the so called Boltzmann distribution.
It's not but which is almost completely random, but not there's a bit of bias toward the lower energy state, but not much.
But there's enough bias that you can do medical imaging and all these other things.
If you can increase the bias, you can get much better signals. And my colleagues at you know, in Berkeley, you know, Vazirani and Shulman figured out how you can use data compression.
So ideas from computer science information theory, the sort of algorithmically cool.
And so instead of putting it and making it really, you don't want to freeze people, like in your, you know, in your, you don't want to freeze your samples, and so on sometimes, because it messes up the messes things up.
But you can algorithm can surgically kind of take the heat out, almost at the molecular scale, pump it out to some other degrees of freedom, and let it dissipate.
Because usually in data compression, you keep the zeros and ones, and you get all this extra zero that you initialize and reallocate to memory, right?
The zeros in medical imaging are the signals you want to see. And if they're all initialized to zero, it's much more visible.
So that's another kind of spinoff of my accidental work in NMR, quantum computing in the early days, because a later student years later, one of my students became interested in this stuff.
So we did some nice work in that as well. But yeah, so it is the same technology, but we work in NMR.
So not the giant medical equipment. So how do you decide what to research next?
What do you use to guide your intuition in this evolving space where, as you said, it's hard to know what's around the corner?
Yeah, it's, you know, everyone has their own approach.
Like, on the one hand, you don't want to be so stubborn, right?
And say, well, no, I'm working on this. And I'm going to work on this for the rest of my life, right?
Like, for some people, it works, and it makes sense.
But on the other hand, you don't want to flip flop and switch from, you know, all the time and really achieve nothing and not leave a mark.
So what's, you know, and I guess everyone has their own approach. For me, it's always been obviously, there's inertia.
And I think that's a good thing. Because, you know, switching, if they're equal things, it's better probably to keep staying the same thing, like equally impactful, equally interesting, probably stay with the area where you're established, know what you're doing, and you're more likely to make an impact than switching, you got all the startup costs, and so on and so on.
But you know, I've pivoted a few times. So you can't generally just keep doing new things, you have to abandon, there's an opportunity cost, right?
So I can always identify new and interesting things. And then I have to decide, is that more interesting than what I'm going to give up as a consequence.
So I've done it a few times, like switching from crypto to switch, I sort of evolved into quantum algorithms, which meant I did less of the crypto.
And then I did a bit, you know, eventually, one of my students is really interested in.
So there's, you have to evaluate the opportunity and the opportunity cost, how much am I going to invest.
So this algorithmic cooling thing, I really invested mentoring one student, he did the heavy lifting.
And I was able to keep moving ahead. Another pivot I did about 10 years ago was this switching this quantum software.
So I've been working for since 96, and the high level algorithmic development.
But at some point, you can't manually compile this, right?
You get 10s, hundreds of qubits, we need automated tools.
And you know, on paper, our tools were good enough. And that Oh, with polynomial overhead, right?
But some of the polynomials are like cubic. So I said, this isn't going to be good enough, if we're going to make the best use of this limited quantum hardware, we have to get really, really good at taking an algorithm and mapping it to run on the fewest quantum bits on that using the fewest amount of quantum resources possible.
And so I started working on that it was partly an opportunity in that the United States was IARPA had a program for this.
I'm like, great, we applied, we got all this money was able to build a team.
And we started tackling these problems.
And honestly, I had a strong intuition for this.
I said, Look, no one's ever really done this intensely or seriously. Because a lot of my colleagues, I think said, Well, what's the point that we know in theory, it's possible, you're just going to implement some code that correspond, you know, we already know, it's possible, you're just going to write some code to implement it, like, why are you wasting your time on that?
I said, because it's not going to be easy, we're going to discover problems and challenges, you know, so as we were trying to do it, you know, how do you do this optimally?
How do you do it in an automated way?
How do you get the best circuit? And how do you get it fast, right?
And we started with simple questions like, how do we synthesize one qubit?
Exactly. And we just using number theory kind of revolutionized that one question.
And then we got another question, which is how do you round off a one qubit unitary that you can't implement exactly?
How do you kind of optimally round it off to something you can implement exactly?
That was related to a number theory question.
So we started this whole algorithmic number theory approach to quantum compilation, which, you know, if you try to implement Shor's algorithm, and you want to implement a pi over 16 rotation, using the old methods, with like, 10 to the minus five precision, you need a logical circuit of depth about 100 ,000.
With our methods, it came down to 50, using number theory, right? And then there was this other problem, where we want to optimize the size of these error correcting code implementations.
And we're able to reduce the optimization question to a matroid optimization problem.
Just so happens because of my math background, my supervisor at Oxford taught me enough, taught me what a matroid was, because he wrote the classic book in it at the time.
So I recognized it as a matroid problem. My student looked it up, Jack Edmonds at Waterloo came up with a polytime algorithm for this matroid partitioning problem in the early 70s.
And we were off to the races.
This is a state of the art tool that is being used around the world for optimizing quantum circuits.
And so it was way more, I knew we'd find interesting things.
And there's many other examples I could give you with this one research program.
We've implemented in an open source tool suite called stack STAQ. So my gut feeling a decade ago was, there's a whole lot of unknowns here.
It's going to be mathematically interesting and very impactful.
That was kind of my gut feeling.
I think I was right. And so that's a big branch of my research. The other branch was the focus on quantum safe cryptography.
And these two branches do overlap because one of the things I do in quantum safe cryptography is help people estimate when we'll break RSA.
And to do that, we need to know how efficiently we can compile Shor's algorithm to run on a quantum computer.
So there's a bit of overlap.
So those are the two main thrusts of my research with a few other sort of complimentary side projects as well, which are keeping us at the cutting edge of algorithmic quantum algorithmic development.
So we do try to better ways to simulate molecules on quantum computer and so on.
That's part of my algorithmic development track.
But my main focus in algorithmic development is to break post -quantum crypto.
But figuring out how to simulate chemicals or design new materials, it's the same kind of part of the brain.
And it's also fun. And we spun off a company actually to commercialize that those innovations as well.
So you mentioned that you work in academia and you just mentioned that you've also spun off a commercial enterprise.
So you have a view of academia and a view of industry. How would you describe the differences with respect to how research is done?
Yeah.
So I spun off two companies that the quantum safe company called EvolutionQ. I did that first because I knew there was a long road.
Seeing my friends who did elliptic curve crypto brought it to be a global standard and so on.
It was a long road and we need to be ready.
And so we started EvolutionQ to help organizations evolve to a quantum safe state.
And then SoftwareQ was founded with Vlad Georgiou. And that's the happy quantum, right?
What can you do with a quantum computer? So those are the two spinoffs.
And actually, there's another one we're working on, but it's not public yet in a different area as well.
So what's the difference? The analogy I like to think of is a standard tool in computer science is the meat in the middle algorithm, right?
It's quadratically better than just brute force search from one end to the other.
So at one end, we have academics in universities.
And we're amazingly good at asking the fundamental questions, exploring the previously unexplored, like finding where these fundamental new knowledge and disruptive possibilities lie, where no one else has thought to look just because it's there, right?
So we're very good at exploring the previously unexplored and unknown and just discovering things.
What is it good for? Well, I don't know yet, right?
That's at the fundamental end. And then some of us work in, okay, what could these be applied to?
So we kind of start from the fundamental, right?
We're just digging and seeing, mining for fundamental knowledge. And then some of us are looking for applications.
And a lot of what we call applied is really applicable research.
It doesn't get applied until further up the stack.
At the other end, in industry, we're up there to solve problems, like create value.
And there's got to be some value proposition that is doing something useful for somebody.
And then, so you start with the problem, and then you start looking for solutions.
And what do you look for? We look for what tools do we have?
And we start going more and more closer to the research, like what have researchers discovered?
What capabilities have they developed? Now, matching these things up, like one application is, well, it's not necessarily a good application, but breaking codes, right?
You work backwards to say, well, how could I break this code?
And Peter Shore brilliantly made this connection between this esoteric quantum algorithm discovered by Dan Simon at the University of Montreal at the time, and the problem of cryptanalysis.
And he made the match, but it was very unexpected, right?
You have this exponential number of nodes coming from fundamental research, and there's exponential working backwards from all these problems, as many approaches you might try.
And suddenly you realize, hey, wait a minute, you know?
And if you meet in the middle, it's generally better than, like, realistically, it's not like you're going to start from academia and say, oh, here's Simon's algorithm.
Like, we would never have thought of, oh, look, I broke the code.
Like, you kind of have to start with the real problems and start articulating many different ways to solve them and just see where unexpectedly, you know, the potential tools lie.
So, in academia, we tend to discover things and develop fundamental knowledge and capability.
I'm not saying without an application.
Sometimes there's an application in mind, but we're really trying to develop fundamental things and potential applications.
Industry has to solve real problems.
And it can be very synergistic, right? Because the more fundamental and new the tool, the more potentially disruptive the impact is, right?
And so, it's a pretty, and I think societies that nurture both of these, you know, these efforts and nurture this connecting in the middle are much more prosperous and successful, you know, at solving important problems and creating value for people.
And this has all been done in Canada. So, tell me a little bit about Canada, Canada's evolution as an innovation hub in the world.
Yeah, you know, I don't know what the master plan was, but it's worked for decades, right?
I think on the research side, it really is, you know, punching way above its weight and second to none.
Obviously, there's other places that do well as well. But I think, you know, we discovered insulin decades ago.
You know, we did a lot of fundamental work in public key cryptography, which, you know, was brought to the world.
That's maybe more on the industry side within trust and, you know, global deployment of PKI and Nortel, and so on, on the industry side.
But, you know, we've played a central role in the academic side of public key cryptography, like with elliptic curve cryptography.
AI was an idea that was nurtured in Canada for decades, right?
People thought that's never going to work. And, you know, but we nurtured it, right?
You know, we had, we have a system, right? It's not a top down system, right?
We have some, we have an effective system for judging what are the good questions, who are the people doing good work, and we somehow make it work.
And we have a long track record of really, you know, we did some important work in blockchain.
One of our undergrads, you know, went off and did some important work there.
And quantum, again, is another area where we started, it was not a clear winner.
So we have a great track record as a nation. And Waterloo is certainly one of these hubs, Toronto, there's so many across Canada, of creating this fundamental knowledge and great talent.
Another really valuable thing, an industry, you know, we have already mentioned some examples, but in terms of lubricating this transition of the knowledge transfer, I think partly our co -op programs, Waterloo, sort of, I think, really pioneered it, but others are doing some version of it as well.
That is a really powerful way to transfer that knowledge.
It's a two way street, it helps us teach students better, they learn a lot while they're students, and then they kind of, I think, are ready to hit the ground running and make very impactful contributions, even while they're students, and even more, as you know, as you have done after they graduate.
So I don't know, you know, there's a lot of wise choices that were made decades ago that we're still, you know, benefiting from.
Okay, and so there's a lot of companies that are international companies that are flocking to Canada.
For one of the reasons is the talent that comes from places like Waterloo.
And so this is one of the big reasons Cloudflare is setting up our new Canadian office in southern Ontario.
So what, what things would, would you do to encourage other international companies to set up shop in Canada?
Or what would you tell companies who haven't really thought about it until now?
It really is, you know, there's several global caliber innovation hubs in Canada.
So I tend to give examples. So Waterloo, Toronto is certainly that corridor is one of the top in the world.
And there's other, there's also, you know, the Vancouver, Montreal, Ottawa area as well.
So they're all great.
They've all have great, you know, critical mass of talent and, and industry and so on.
There's a number of reasons, obviously, there's, you know, there's a high standard of living, it's a great place to live and all those, you know, things, but you're really at the source, you're right next to the source.
Like there's a global competition for the talent that Canada produces.
Obviously, we'd like to keep as much as possible in Canada, but it's a global market.
And a lot of Canadians go abroad, we of course, you know, a lot of for, you know, a lot of non Canadians move to Canada, like that's all, it's all great.
So there's a global market competition for great Canadian talent.
And obviously, if you set up shop here, you'll be able to engage with that talent more and more directly.
And you can also start engaging with them before they even graduate.
So there's the co-op program, which of course, you can do internationally as well.
But the, you know, there's a larger percentage of students that stay within Canada for co -op, though, we're very excited that proud that many of them go abroad for their co -op terms.
That's a great experience, of course, but you have greater access to more co-op students.
But you can also engage with a lot of the research centres at the universities.
And I guess one thing I think that I think is a really valuable thing strategically, is to engage, you know, sometimes less is more.
So have a small engagements with the real innovators in the areas of science that are, you know, or mathematics and computing that are closest to your business.
Because then you start creating that familiarity.
And, and that'll, I think, make more apparent the more the deeper, you know, relationships you might want to pursue later.
Right? Because by the time, it's sort of obvious that you might want to work with somebody who chances are someone else has already scooped up that opportunity.
So I think, to recap, we have a tremendous track record of producing world changing talent amongst, again, the most of the knowledge transfer, so called knowledge transfer is, you know, our students, our graduates, through our co-op program and hiring them.
And if you're in Canada, you're going to have front row access to that.
And I think the other, you know, is starting to engage with research groups, the many, many stellar research groups we have across Canada.
And again, there's probably ways to I personally would like to see a more coherent, like, I think every graduate student should have an industry mentor.
And, you know, and less is more, make it small entry to, you know, barrier to entry.
So don't complicate it with IP agreements.
This is just casual, twice a year, someone to tell about what you do just to learn your communication skills, and get a bit of insight into the real world job market.
That low bar is already now creating a relationship between industry and all the wonderful research groups you have at the universities.
And even if one in 100 of these 10s of 1000s of relationships suddenly say, hey, well, what, you know what, you should talk to so and so, right?
I think that this can open up many great, serendipitous, unexpected opportunities for these companies.
Well, like, look at my own journey. I think everyone's journey, it was all serendipitous and unexpected.
There's just, you know, many, many opportunities present themselves.
And then every once in a while, there's one that you're really excited about.
So I think, you know, there's a long track record of things that, like, look at the BlackBerry security, that was one, one young person they hired, you know, had been influenced by one of my mentors, actually.
And he got he realized that public key cryptography was a game changer in the 90s, right.
And that elliptic curve was an enabler, because it was much more efficient.
No one else had seen that, right. And it was fundamental to the success of the BlackBerry in the early days.
Right. And so that all this serendipity, you can't predict it.
But if you do enough of these little epsilons, every once in a while, you really hit hit a jackpot, so to speak.
Okay, well, this has been an amazing conversation.
Michele, thank you so much for chatting with me today.
And I'm sure everybody in the Cloudflare TV audience has really enjoyed this conversation.
So thank you again. And to folks online. See you again soon. All right.
Thanks so much. All the best. Everybody should have access to a credit history that they can use to improve their their situation.
Hi, guys.
I am Tiffany Fong. I'm head of growth marketing here at Kiva. Hi, I'm Anthony Voutas.
And I am a senior engineer on the Kiva protocol team. Great. Tiffany, what is Kiva?
And how does it work? And how does it help people who are unbanked? Micro lending was developed to give unbanked people across the world access to capital to help better their lives.
They have very limited or no access to traditional financial banking services.
And this is particularly the case in developing countries.
Kiva.org is a crowdfunding platform that allows people like you and me to lend as little as $25 to these entrepreneurs and small businesses around the world.
So anyone can lend money to people who are unbanked. How many people is that?
So there are 1.7 billion people considered unbanked by the financial system.
Anthony, what is Kiva protocol? And how does it work? Kiva protocol is a mechanism for providing credit history to people who are unbanked or under banked in the developing world.
What Kiva protocol does is it enables a consistent identifier within a financial system so that the credit bureau can develop and produce complete credit reports for the citizens of that country.
That sounds pretty cutting edge.
You're allowing individuals who never before had the ability to access credit to develop a credit history.
Yes. A lot of our security models in the West are reliant on this idea that everybody has their own personal device.
That doesn't work in developing countries. In these environments, even if you're at a bank, you might not have a reliable Internet connection.
The devices in the bank are typically shared by multiple people.
They're probably even used for personal use.
And also on top of that, the devices themselves are probably on the cheaper side.
So all of this put together means that we're working with the bare minimum of resources in terms of technology, in terms of a reliable Internet.
What is Kiva's solution to these challenges? We want to intervene at every possible network hop that we can to make sure that the performance and reliability of our application is as in control as it possibly can be.
Now, it's not going to be in total control because we have that last hop on the network.
But with Cloudflare, we're able to really optimize the network hops that are between our services and the local ISPs in the countries that we're serving.
What do you hope to achieve with Kiva?
Ultimately, I think our collective goal is to allow anyone in the world to have access to the capital they need to improve their lives and to achieve their dreams.
If people are in poverty and we give them a way to improve their communities, the lives of the people around them, to become more mobile and contribute to making their world a better place, I think that's definitely a good thing.
you