Loading summary
Transcript113 lines
- [00:03]
A
Hello, this is Malcolm Gladwell and you're listening to Smart Talks with IBM. Every year, Tech Week brings thousands of people together to network and learn about what's emerging across the technology ecosystem. And at this year's conference in San Francisco, I had an amazing opportunity to sit down in front of a live audience with Jay Gambatta. Jay has been with IBM for years and was recently promoted to Director of Research. In this job, Jay has an important mission, helping the company build the future of computing. In the last episode of Smart Talks, I began to learn about quantum computing from IBM Chairman and CEO Arvind Krishna. But this conversation I had with Jay went even deeper and convinced me that the development of Quantum isn't just a fun, exciting new paradigm of computing. It may be one of the most important scientific achievements of my lifetime. Jay, morning.
- [01:02]
B
Morning.
- [01:03]
A
Welcome to Smart Talks with IBM.
- [01:05]
B
Thank you.
- [01:05]
A
Special live recording here for Tech Week. And congratulations. How long have you been head of research at IBM?
- [01:12]
B
Since October 1st. It's October 10th today.
- [01:15]
A
Nine days.
- [01:16]
B
Nine days.
- [01:17]
A
Can you just talk a little bit about the position? This is one of the most important positions in research in the world.
- [01:25]
B
IBM research has been around for 80 years and it's done some tremendous technology, a lot of inventions in fundamentals for semiconductors, algorithms, AI. Yeah. I think if we look back to where a lot of the innovation and the technology of the world comes from, I think you can find IBM's footprints on it and you can find IBM research. So, yeah, I'm very excited for the opportunity, but I'm also aware that there's big shoes to fill and I'm looking forward to how we take IBM research forward. Obviously, I'm going to be bringing a lot of the Quantum side, which we're going to talk about later. But beyond Quantum, there's important work that needs to happen in AI hybrid cloud, and I think we're going to also enter into this new period of mathematics where we get to use quantum machines and also AI machines. And there's some really good hard mathematical questions to answer.
- [02:20]
A
How many people do you have working for you?
- [02:22]
B
I've been researchers in the 3,000 researchers across many different labs around the world. Our main lab is in Yorktown, but then we have the lab actually out on the west coast in Armadan or SVL now. And then we have one in Zurich, Japan, and a few others around the world.
- [02:39]
A
Tell me a little bit before we get into quantum. I'm just curious about your path. So you're Australian.
- [02:45]
B
Yep.
- [02:45]
A
You were talking about earlier backstage, your accent has become muted. You should crank it up because it's.
- [02:52]
B
Yeah, I'm slowly losing my Australian accent. I've been in the US since 2004,
- [02:57]
A
so it's an asset, you know, to sound very Australian.
- [02:59]
B
Yeah, but how do you practise it? Maybe I gotta go back to Australia, hear more Australians say, g', day, how's it going? Things like that.
- [03:08]
A
And you, you didn't grow up thinking you were gonna be a scientist one day?
- [03:12]
B
No, I grew up in a pretty normal life. My dreams as a kid was building things, so I was either gonna be a carpenter or a me. But I had some great teachers that inspired me to go to university. And I didn't even know honestly what a scientist was. And then I found myself at university doing science, particular physics, and I ended up loving it.
- [03:34]
A
So you go from there to, where do you do your PhD?
- [03:38]
B
So I did my undergrad in Australia. I did it actually in laser science. So I think I watched a some TV show and lasers seemed interesting, so I wanted to learn about lasers. And then I realized in trying to understand lasers, there was this quantum mechanics. And so I was like, all right, I want to actually understand this quantum mechanics. So I did my equivalent of what you in the U.S. call Masters, we call it honors in Australia. But we do a research project. I said I wanted to shoot lasers into atoms and measure cross sections. And I got really into quantum physics. So then I decided, all right, I don't understand this quantum physics. I want to do my PhD in interpretations of quantum mechanics. So I jumped in and said, all right, what is this quantum mechanics? Why is everyone arguing on these different interpretations? Then I finished my PhD in Australia doing that. Then I moved over. At the end of my PhD interpretations, it's more people arguing about the equations. Whilst I think it's really important, I decided if it's going to be like a collapse equation versus many worlds or a hidden variable model, or that just quantum mechanics decoheres because we don't see superstitions in the everyday world because it interacts with its environment. The only way to answer that question was to build a quantum computer. And so then I decided at the end of my PhD, I wanted to work out how to build a quantum computer. And then I left there and I went to Yale. And then at Yale, that's where I got into superconducting qubits, which just a few days ago, one of the professors there just won the Nobel Prize this year.
- [05:19]
A
Oh, wow. Yeah. I'm very interested in tracing because your career follows the arc of quantum computing in a certain way. Right at the time when you asked the question, what I really want to do is to figure out how to build a quantum computer. Where are we in quantum computing at that point?
- [05:36]
B
Yeah. So that would have been 1990. So there was Shor's algorithm came out, let's say 95. There was a lot of theory. And then the reason I went to Yale is because people had started to show that they could see quantum effects in electrical circuits. So these macroscopic objects, they were starting to behave quantum mechanical. There was a really significant breakthrough in 1999 where Yazu Nakamura in Japan showed that a qubit could exist in these electrical circuits. And then I found out the group at Yale were really trying to take these electrical circuits and couple of them together. And so it was like, if I can build something using electrical circuits and they're big, that's the best way that you can sort of test and understand whether quantum mechanics breaks down at a macroscopic scale or not. Can we actually make them behave as qubits? And I agree. When I came to Yale, the qubits were not very good. They were actually a couple of nanoseconds, they were unstable. An electron would jump onto the chip, and then they would change all their configurations. So you have to restart your experiment. And so for the first time at Yale, it's kind of what the challenge there was. How do we make a qubit? How do we make a stable qubit? And that took about five years, and that took us up to 2007. And I think the rest of the world looks and says quantum's like, just blowing up. But it's actually been like almost phases. Theory showing that we got the algorithms. How do we make a qubit? How do we couple of the qubits together? And now we're in the scaling phase.
- [07:16]
A
Describe for us, because many people in this room, me included, have only a kind of surface level understanding of what we mean when we use that phrase. What is the difference between classical computing and quantum computing? What does that word mean?
- [07:31]
B
Yeah, so you can go down the physics way and talk about superposition and entanglement, which we can go in later. But I actually feel that's a bit of a distraction. So when you think of classical computers, what they were is they were machines that were very good at adding numbers together, like simple addition. And they really showed that they could add these numbers together really, really fast. And now with GPUs and other AI accelerators, we can add those numbers together in parallel and so the whole classical computing can come down to just arithmetic, just adding numbers together. It turns out that there's a math that is new, that the quantum mechanics shown to be true. It's more like a group theory type structure. And the way quantum works is it has a different math as a primitive. And if we can exploit that new math and build a machine that does it, it allows us to answer different questions. And so think of it as a branching from classical compute that is very good at adding just numbers together to something that allows us to work with an algebra and that is much, much harder to represent with addition. And that algebra happens to be the same algebra that defines the fundamental equations of nature. Schrodinger's equation. So this is why you say it computes the same way nature does. But there are many other interesting problems. So the way I explain it to people is think of it as bringing a new primitive to computer science and allowing us to work how to go with it. And I like the analogy. Well, actually maybe go back. So if you went back in time, so we're 100 years at Quantum, and you went back in time and you asked, what is the foundation? Is it chemistry or physics? What would have, probably the scientists of 100 years ago would have said is they would have said, you know, chemistry is about the small, physics is about planets and things like this. And 100 years ago, when Heisenberg, Bohr, Einstein, all the greats, Schrodinger himself invented quantum mechanics. It was this concept that nature is discrete, not continuous. It actually brought all the physical sciences together. And now quantum mechanics is like, it is the foundation of the science. And so now what quantum computing is by that analogy is computer science. The foundation of the math is coming together with the physical science to allow us to compute using math. That if you were to try to represent it with classical computers, it takes exponential time.
- [10:06]
A
Yeah. Anyways, a classical computer and explain this in a way that someone as well informed as I am can understand it. A classical computer works primarily on problems that can be easily represented in numerical form in numbers.
- [10:19]
B
Yes.
- [10:20]
A
Quantum allows you to step outside to a class of problems that don't necessarily have a simple numerical representation.
- [10:27]
B
Yeah. And so imagine I got some medicine or, or some set of operation, they'll call it A, and I then follow it by a different operation, B. If A followed by B gave a different answer than B first followed by A. So in mathematics we call that commuting. But like you can think of a correlation there, one gives you a different outcome to the other that means there's an algebra behind it, that representing that algebra traditionally on classical computers is really, really hard. Whereas that algebra, if we can get creative, we can come up with ways of representing that math. So we step, as you say, we step outside of the simple math to a new math to allow us to calculate interesting problems.
- [11:13]
A
So quantum, in a sense, complements. It doesn't replace traditional classical computing.
- [11:18]
B
I think this is one of the. This is. You're exactly on. People think quantum is going to be replacing classical. If your problem is good at adding numbers together, you should just keep using classical computers. I think the future is going to be heterogeneous accelerators, and it will definitely have quantum as one. But in some sense, the next generation of superstars are going to be those applied mathematicians that know, how do I write a problem using the simple math of classical computers or the more complicated math for quantum computers, and how do I actually iterate between them? And things like this. This is where I think the next generation of students are going to come up with much more novel ideas. I can give you examples of what we want to do on quantum, but you're giving them a fundamental, foundational new thing. And so I'm optimistic that we'll do much better jobs than my generation will.
- [12:11]
A
Yeah, we're going to get to some examples in a moment, but I wanted you to, in the most kind of down, you said as a kid you thought you might want to be a mechanic because you like to build things. Describe to me what it takes to be build a quantum computer. Like, what are you doing that's different from building a classical computer?
- [12:29]
B
Yeah. So maybe I'll give you an analogy and then I'll go in. So the way classical computers, we've got them to get to smaller and smaller sizes, like 5, 7nm, 5nm and things is actually inventing material to kill quantum effects. So you actually put dielectrics and other things in there to kill the quantum tunneling effects. And you want them to behave more classically in the quantum world. You want to get rid of all the classical effects. So you want to get rid of the ability of the qubits to interact with the environment. And in the sort of technical world we call it this quantum conflict, the more ways you want to control a quantum computer, you open it up to interacting with everything else, like interacting with its environment. So the biggest challenge has always been how do we give more control but don't bring in other sources of noise. So I want to be able to do Gates on the qubit, but I don't want it to decohere. I want to couple the qubits, but I don't want them to couple to other things. So the hardest challenge is the energy inside the qubits is 9 gigahertz. And if you times that by h bar 10 to the negative 34, with 9, you're at 10 to the negative 23 or something in energy. That's a tiny amount of energy. So you're trying to have a tiny, tiny amount of energy to control, and you don't want that to interact with anything. So you have to cool them down, you have to isolate them, and you have to make the quantum effects dominate over the classical effects.
- [14:10]
A
So practically, if I'm trying to do that right now, how big are these machines?
- [14:15]
B
So the qubits themselves are not that big. So the qubits themselves are like a few microns, but, yeah, most of the size. So you can see some of our. I got the pleasure of showing you around to one of the machines in Yorktown. You saw that they're like 20 foot by 20 foot in size. Most of that is all that equipment to isolate the qubit ship, which is only a few millimeters, when you put it together, from the rest of the environment, we will, as we get better at that, miniaturize all the isolation. But that's cooling it down to a few millikelvin, so about a thousand times colder than outer space. It's isolating the noise on any electrical signal so that no noise from the outside world gets into the system. And so that's a lot of isolators, filters, and things like that that we've had to invent to allow us to make the quantum properties of this chip go.
- [15:10]
A
Yeah, it's like the Princess and the pea. Mounds and mounds and mounds of mattresses trying to isolate the impact of this little thing in the.
- [15:19]
B
Maybe that's the best way to describe it. Yeah. And you got to keep it really, really prestige.
- [15:24]
A
But that when you showed me. So in the lobby of the Watson Research center in Yorktown, which, by the way, is just the coolest building.
- [15:33]
B
It's like a modernist, it's awesome masterpiece.
- [15:37]
A
Anyway, in the lobby, there's. There are these. Is it two machines?
- [15:42]
B
It's inside a container that has three machines.
- [15:45]
A
Three machines, yeah. So what can you. Can you tell me what would one of those machines cost to build right now?
- [15:51]
B
So typically, we put them together in a way where we upgrade them because we want to. As I was talking about before, One of the things we want to do is always get algorithms done on our machines. And I've got a roadmap of building bigger and bigger machines. So usually one of those quantum processes two day is out of date in six months. So we want to build this future of computing that leverages quantum computing where every six months we've outdated a quantum processor. Eventually, hopefully we get to a point where it's like stable and it can be many years operating. But we want to get as large a quantum computer in the hands of people to explore the math as possible to come up with those new algorithms. So we've had a philosophy of having them open, working with universities and things like that. So to answer your question of cost, yes, there's cost in building the system, but we are operating in the much more in a service model where people pay to use the machine because we have to continuously calibrate it and operate it. And so depending on various different things, professors, we have a credits program where they get free access. Some universities and enterprises, they can buy premium access and get more access. So think of not like a cost of it because it's almost like a continuum. I want to make sure that the best quantum processes that I can build get in the hands of students and professors and interested enterprises that want to explore these machines as fast as possible. And typically every six months we upgrade it.
- [17:29]
A
Yeah, you don't start over, you upgrade.
- [17:32]
B
We upgrade various different pieces, the processor, the electronics. Some upgrades are just simply replace the processor. But as an example, I think many people have probably seen photos of quantum computers and you see this like scary thing with all these wires hanging down, as I've heard it referred to as the chandelier. And it's got all these wires with loops and things like that. They're called coax cables. When we first put the quantum computer on the cloud in 2016, you could probably only fit about 50 qubits inside one cryostat. We've had to upgrade all those cables so that we can fit around 1,000. I want to get to 3,000. And that's about miniaturizing it. So to answer your question and upgrade, it depends. It can be either just the processor or it can be the complete insides. And we're actually in our third generation of our electronics to control the systems, to make them faster, less noise. Internally we got exciting results of going to something like cold cryo cmos. So you can bring down the cost in terms of energy of running these quantum computers almost to negligible. And you could Imagine future quantum computers are not going to require much energy to run. So unlike classical compute that requires lots of energy. The biggest machines that we envision is only in the few megawatts, but we have to upgrade to future controls that use less energy. So it depends is my long answer, short answer to how it upgrades and it depends on what it is.
- [19:04]
A
The only observation that I felt I could was capable of making when you showed me the quantum machine is it's gorgeous work of art.
- [19:12]
B
I've always believed that. And I think there's an IBM saying good design is good business, but we've always taken pride in making sure what we build. I don't know, I feel if you're going to build something that is new, that can change, you should take the time to make sure it looks and feels good.
- [19:33]
A
Will you donate it to moma when you're through with that particular.
- [19:37]
B
Actually, I think we just put a old version of one of our insights with United Airlines and the aps, which is the American Physical Society and the University of Chicago. There's a replica right now if you fly into one of the terminals in Chicago, you can walk and see one.
- [19:55]
A
Oh really?
- [19:56]
B
Yeah.
- [19:56]
A
Well, the most advanced thing at o' Hare, I'm sure, probably.
- [19:59]
B
But yeah, hopefully I think yeah, we're open to that. But I, yeah, I, I appreciate that you love the design. Beautiful.
- [20:07]
A
So I. Last week I interviewed for another episode of Smart Talks, your CEO Arvind Krishna. And when we got to the quantum question, I mean he's always elegant and brilliant and. But quantum, he's like lit up. Am I right in thinking that IBM is much more invested in quantum than anybody else? Is that a fair statement?
- [20:30]
B
Oh yeah, most definitely.
- [20:32]
A
Why did IBM choose to kind of make this such a priority?
- [20:36]
B
So when I talk to the history of the physics side, there's this interesting thing in the history of computing. So we build computers like classical computers today, using bits and CMOS and they consume energy. Do you know that there is a way in classical where you can actually compute without using energy? It's called reversal computing. Turns out to be a terrible idea. It's not practical to build, but IBM investigated that with Ralph Laura and Charlie Bennett early on and they proved the concept that reversible computing, the first use of quantum information theory, one of the first actually was from IBM. When I did my PhD. I remember actually picking up this paper on quantum teleportation and seeing IBM written there. And at that time I rem thinking, don't they make PCs? What the hell are they doing this foundational paper on quantum teleportation. Why are they doing it? So to answer your question, actually IBM was the first in quantum information science because it's the fundamental of computation. Can we actually come up with compute that we can go beyond the classical. So way before anyone was talking about it they were doing fundamental theory. And then as we've built it we've always. When I first came there the experimental team was small. In 2011 we've had a small team that were focusing on single qubits coupling them. I think in 2012 was the first time we showed really good two qubit gates and no one was talking about quantum computing then. And then I remember in about 2016 I said to actually Arvind was the director of research then. Can we actually put our quantum computer on the cloud? Well it was probably 2015 and it was always supporting that. So as we've done more and more we've been able to do it. It's had this program going now I agree is very visible because we're in the scaling phase and so we're invested to keep scaling it and to get why is at IBM research what we always do is answer what is the future of computing? Whether it's coming up with new algorithms, coming up with better AI, coming up with quantum or coming up with just how do different accelerators go together. It's our DNA to answer the question of what is the future.
- [23:05]
A
Isn't it a perfect problem for IBM because you kind of need to have a legacy of building stuff, building actual physical machines.
- [23:14]
B
Yeah. It's why I came to IBM. I wanted the experience, the culture of building hard things that others have not done before.
- [23:26]
A
Where do you imagine we are in the timeline of this technology? There will come a point when it will mature. My cell phone is a mature technology at this point. How far are we from that point with quantum.
- [23:39]
B
So I think there's various aspects of it. So we sat in 2017 we set our goal that in 2023 we would be able to build a machine that was beyond classical computers to simulate it. And we achieved that in 2023. So to run run a big we call it a quantum circuit. The details of it don't matter but to run a quantum workload that if you were to simulate that workload how a quantum computer operates on a classical computer, you couldn't do it. So we set that as our first and now I've made it publicly that by 2029 we'll build the first fault tolerant quantum computer that is one that can completely handle the noise to the level to allow you to run a very, very large problem.
- [24:27]
A
It's an example of a large problem.
- [24:28]
B
Yeah, a large quantum problem. So for around a couple hundred qubits and 100 million operations, you're talking still interesting science problems like simulating a molecule or calculating a small optimization problem or calculating say some part of a matrix update in some type of differential. So it'll still be scientific, but it'll be at the point where it's beyond, well beyond any classical approximate method. And then I think that's 2029. That's 2029.
- [25:04]
A
So we're four years away from something that can start to handle interesting, serious problems.
- [25:10]
B
I do believe the scientists will find interesting heuristic problems before that. And so over the next four years, you're going to continue to see more and more, let's call them heuristic, not provable quantum problems that run on quantum computers that come out. We're seeing more and more come from many of our partners and ourselves. Heuristic problems have value, but they have to be tested. They have to stand up over time. You have to run them many, many times. You have to try different ones. And many times heuristic can lead to formal problems. So you're going to see because we're beyond now the point that you can simulate these quantum computers with any classical computer. They're kind of like a scientific tool. So they're exploring the heuristic.
- [25:51]
A
What do you have to get done between now and 2029 to get there?
- [25:54]
B
So we had to reinvent how we wanted to do error correction. So we have to demonstrate modules and if we can demonstrate these error corrected module. And our goal is actually it's called Kookaburra. I name all our chips after birds. So it's called Kookaburra. It's named after an Australian bird. I think I still say Kookaburra the way Australians do. We need to then show that we can make a single module and then we want to connect two of those modules together. And I call that one cockatoo, which is another Australian bird. And then if we can do that, so that's 26 and 27. And then we want to scale them, scale those modules and that we call Starlink. And we want to scale that in 2029. So get a module, join two modules together and scale. And so each module is going to be around a thousand qubits. Yeah.
- [26:44]
A
The challenge to getting there is it finding the right material or how would you describe what needs to be done,
- [26:51]
B
that's the beauty of it is if we would have been here two years ago, I couldn't tell you how it would be done. So we had a huge breakthrough. We came up with a new code, a new quantum error crushing code. And that code, the biggest part of that code that is the most important is its modular in nature. So previous codes, without getting too technical, they were very monolithic and you had to build a very big device. And I wouldn't have known we would have to invent tools, like new CMOS tools to do that. So we came up with this new code. We started on 2019, we published in 2024. We kind of had most of the things worked out in 2023. That's why confident to release the thing. So the biggest breakthrough we had is coming up with a code that's modular in nature. And think of that as like a blueprint. And so now we have the blueprint, and now we're doing engineering tasks to implement every part of that blueprint.
- [27:52]
A
And so the minute you had that breakthrough, then you began to have confidence that some of these goals could be met.
- [27:58]
B
And then you can't. And then anyone that's done engineering will know what I'm talking about when I say this is cycles of learning. It takes so long from test idea to build to test. In hardware, the cycles of learning are much, much slower than software. Like you can be really, really fast in the software. So then we've planned out our iterations over the next few years and so we have to successfully demonstrate them. I may slip, because sometimes you may estimate your time wrong. But we now have exactly what we want to do for the next four years.
- [28:35]
A
I want to go back to that breakthrough for a moment. What does the word breakthrough mean in that context? Like, it's not that you get a call in the morning from somebody who says, I did it. Do you see it coming or is it a surprise when they get there?
- [28:47]
B
So the way this one worked is Sergey Bravi, who's an algorithm person at IBM, one of the smartest in Quantum information.
- [28:55]
A
Don't mention his name. Silicon Valley. They'll come for him.
- [28:59]
B
Everyone in Quantum already knows his name. I don't think there's an idea that has not originated from him in Quantum. I mean, I. So we were looking at other codes and we're going, all right, we gotta get serious about these codes. And others were starting to propose to bring these, we call them LDBC codes, from the classical space into the quantum. And I asked him we need to get ahead of this and understand what they're doing it. He's like the most modest. Puffy is like, jay, let me learn about them and I'll generate a report for us and we'll read through it. And then I said, great. Then I don't know, six months later he comes back with 100 page report on. Everyone had done an LDPC code. So I'm like, awesome. So I started then to read from them and then we said, all right, how do we, under the assumptions of the hardware we can build, can we get an LDPC code knowing what we can build? And that's a great question. And so we put a small team together to investigate and it honestly took two to three years. And we iterated and we used the constraints. So we had the sort of theory and then we had the constraints of what we could build and we iterated for a few years and then at the end of that, we came out with a solution that, yes, it is possible to meet all the constraints of the hardware and build a code that will work.
- [30:28]
A
I'm just curious about. So you have this task, this problem you want to solve, and when you set out on the task of trying to solve the problem, what's your certainty level that you'll get a solution?
- [30:41]
B
Well, that's the beauty of science. For things where you kind of have a few ideas, my philosophy is try a few for the ones that need to be in that, like, wow moment. It's honestly, you got to set the ambition really, really high, but know when to stop. It was a great team that went together to get that breakthrough and we knew that we needed to come up with a code that met the requirements of the experiment. And I think what was different before then is the theorists that were doing error correction codes didn't necessarily know the constraints of experiments. So it was like really more pen and paper. So this became one. All right. Given these sets of constraints, is it possible.
- [31:28]
A
One last question about this. Sorry. I love these kind of moments when things become clear. At the time the problem was solved, were you aware of the implications of the solution or did that take. You did. You knew exactly what you had.
- [31:42]
B
We set out exactly like, either we were going to have to work out how to cool down a very large piece of silicon, which would require a lot of engineering and building tools beyond what anyone has ever built in the silicon CMOS industry to implement the known codes, or we had to come up with a different one.
- [32:02]
A
Yeah.
- [32:03]
B
And once I knew that we had one that I didn't need to reinvent any tools to build. The implications are clear.
- [32:11]
A
How much time elapsed between the time you heard the problem was solved and the time you told Arvind and Krishna, the CEO the problem was solved?
- [32:20]
B
I'm sure the next time I spoke to him, I update, but I don't remember. The beauty of Arvind is he trusts that scientists will do it. And so he doesn't really check on us. We update him when it is, and he empowers us to do really hard problems.
- [32:34]
A
Yeah. So let's talk about uses. I mean, the really like cool, big, shiny machine. Think you'll get by 2029. But there's all kinds of really interesting problems you're already working on.
- [32:47]
B
Yes. This is like another interesting area is I can prove in pen and paper algorithms that we want to run that like, it's not that we don't know what to do with a quantum computer. There are hundreds of algorithms you can go to. I think it's called quantumzoo.com and you can see many, many algorithms. People are coming up with more and more of them that they prove by pen and paper. Imagine now we have a machine that you can't simulate. How do you actually discover algorithms in a scientific way? How do you look and discover algorithms using a quantum computer? We're in this exciting period right now. And so even though I can prove these ones that we can run in the future, there's a big white space between the machines we have and we're going to build and continue to do, and those ones that one, the provable ones. And I'm an optimistic person by nature, I think getting those machines in the hands of students to explore and look at heuristic algorithms. So looking at the equivalent of doing numerical algorithms on computers, which there's many histories of numerical algorithms being discovered on classical computers. Before we had formal proofs that we rely on today, people would even argue the way AI works was driven numerically, even though we have input into it. There are ones in optimization driven numerically. We are entering that phase. So the computer scientists now need to go play with these primitives. Our prediction is over the next couple of years, we're going to see valuable numerical equivalent algorithms emerge. And where the scientists are going is in four categories. One is simulating nature. So looking at either high energy, physics, chemistry, light problems. As an example, with our partners in Japan, they took one of our quantum computers and Fugaku, a very large classical supercomputer, and they ran a problem where quantum was just A subroutine of the problem that was running on all of Fugaku. And they were able to look at an interesting molecule. A molecule that if you would go by pen and paper, you would have said, it's going to take me a very long time to run that. They were able to run that quite accurately heuristically and already get results that are comparable with the best classical methods. So they are extremely excited because they want to push that further. And they're sort of showing that you can take a classical supercomputer with quantum as a subroutine and start to push the level.
- [35:25]
A
So they were. This was. They're trying to solve a medical problem
- [35:29]
B
is that this one is a. Like most people don't realize, like iron, sulfur. Just something as simple as iron and sulfur, we can't solve that exactly. Like iron sulfur molecules are too hard. So really small, small molecules are really, really hard. Too hard for classical computers to solve. People think we can solve a lot of things. It actually turns out we can't solve very much.
- [35:52]
A
You say solvent instance, you know precisely how that molecule works and is constructed,
- [35:57]
B
know precisely what the energy levels of that molecule is and how they come together and then be able to do that on a classical computer and compare it to a quantum.
- [36:07]
A
It would be really, really useful to know that specifically, because if you can
- [36:12]
B
have energy levels, then you can estimate reaction rates. If you can estimate reaction rates, you can see how different types of chemicals will react. That can then lead to better informing event how to build materials or even drug design. I just want to be careful and not say, oh, we're going to solve drug design or that. Because there's many scientific steps to make that so. And so what quantum gives you is a different tool to give you more accuracy and then lead to making the different methods work.
- [36:43]
A
You can subcontract out aspects of a problem quantum right now, and that just gets you further along than you would have been.
- [36:51]
B
So at the moment, even this result still does not beat the best approximate classical method. It's comparable. So the art of chemistry for the last hundred years has been about approximating. So what we've done is we have gotten very, very good at coming up with ways of approximating nature. And a lot of the things that we do and we exploit and we use to estimate approximations, they don't simulate nature the way nature approximator. And there's. I could list many different acronyms of different methods that go into approximating nature. What quantum gives us is to eventually get beyond that approximation and do it the way nature works. And so we aren't beating those approximation methods. And this is why, I think this is why it's still in the science. But they're getting comparable. Getting comparable with a new tool where the previous tool is a dead end, makes scientists very excited that nuance is where it is. And so that's in machine learning. Sorry, Hamiltonian. Then there's examples in differential equations. So can I actually come up with differential equations and solve them? And if I can solve them, you could look at things like Navier Stokes equation goes into weather. There's financial differential equations that you can better predict. So differential equations, there's many different examples there. And then I would say the two others are optimization. And then there's quantum versions of machine learning that are very exciting as well. Yeah.
- [38:22]
A
Cleveland Clinic, one of the organizations that you guys have worked with, why would the Cleveland Clinic be calling you up?
- [38:29]
B
Because that problem that they want to look at. So they've also done similar problem to the Rican lab. So they've taken that method now and they've looked at molecules that matter for drug design. So they're fundamentally looking at those molecules that matter for eventually replacing some of the steps. So they're investing to see how reliable it can be done. And so there's a scientist there that's done many iterations now using the techniques that were done first with the team in Japan. They've now replicated that for new molecules that are essential primitives for eventually designing drugs and things that may matter for medical.
- [39:09]
A
Yeah, and also there's some finance firms, banks, hsbc, Vanguard, and their interest is what?
- [39:17]
B
So that was the differential equation and optimization. So if you are doing very large calculations like risk portfolio, or if you want to model the Black Scholes equation or things like this that are fundamental for them to make better predictions, come up with better trades and things like this. That is a very hard computational task. And so rather than quantum replacing that whole problem, can quantum be a subroutine in there? And what HSBC showed is they showed they could take their real data, they could take their real classical method, and they just replaced a tiny part of it. They replaced a tiny part of it with a quantum subroutine that allowed them to come up with better predictions of the weights that then when they were to Compare trial A versus trial B, it was 34% better at predicting algorithmic trend. And that's a big deal for them. It's huge. Yes.
- [40:11]
A
Yeah.
- [40:12]
B
Now, do they need to do more trials? Do they need to see Is this a heuristic algorithm? Do we need to be careful? Is there other classical algorithms that go into these? These are great questions that are now being investigated. So think of this period of heuristic algorithms as really a period of scientific discovery using these machines, knowing that we want to continue and build the ones which have deterministic algorithms that can run.
- [40:39]
A
Do the people who would profit the most by starting to run quantum experiments realize that they would profit so much from running quantum experiments? In other words, does the world know this? You've given us a couple of specific examples, but generally speaking, there must be a very large universe of people who could gain from at least starting to play in the space.
- [41:02]
B
So the enterprises that use computation as key for their survival understand the limits of classical computation. And they're very interested to get started. The universities are very interested. Could we get more students doing more algorithms? 100%. Some of the limitations on the rate of algorithm discovery is because people are thinking through the classical way of writing algorithms. My belief is yes. So this is why we want to get more and more students and things, because it's just starting. But I would say in general, most people are aware of it. Could we get more? Could we accelerate it? Yes. Do we need to make better hardware? Do we need to come up with better libraries? Yes. Do we need better software? Yes. But it's all happening over the next few years.
- [41:49]
A
Is it hard to get someone who's spent their entire life thinking in terms of solving problems through classical means to make the transition to this new paradigm?
- [41:57]
B
It. There's a lot of examples when you approach something with the classical intuition, it's not the right way to do it when you approach it through the quantum. But if people are being taught to understand the fundamentals of the math, then a lot of the techniques carry across. I don't recommend people need to learn about entanglement or superposition because whilst the physicist will argue like spooky action at a distance and all these type of things, entanglement is the power. Yes. That's how physicists are labeled, how quantum is different. But I would say do we need some physicists really worrying, thinking about that? Yes. But we need more applied mathematicians that are realizing they can use this as a different way of looking at the problems.
- [42:43]
A
Yeah. I want to ask you one last question though. Is we're describing. It's more than a new technology. Were talking about a new paradigm, just a way of thinking about problems. Can you compare this to kind of previous technological paradigms? If I'm looking At the last couple hundred years, where does this rank in terms of a new field that we've opened up?
- [43:06]
B
It's a hard question to answer, but I often say the history of computing, this will be the first time that computation has branched between classical and quantum them. I like thinking, reading a lot. In the past, one of the things that I think was a way we changed as a society was the invention of zero. Before zero, math was limited. Realizing that numbers have a number as 0 allowed us to develop a whole set of new mathematics that then went on and defined, like, everything from waves to calculus to all of that. Yes, we can describe it with that same math, but when we describe it with that math, it gets exponentially big and gets impractical to do. Now we can actually work on it. I would say if I had to give you a quick answer, maybe going all the way back to when we were accepted, zero.
- [44:02]
A
I thought you were going to say like the airplane. But fact, you went several orders of magnitude beyond that.
- [44:07]
B
Yes, but. But I think it's so fundamental.
- [44:11]
A
This is absolutely fascinating. Thank you so much for chatting with me about it.
- [44:15]
B
Thank you for your time.
- [44:18]
A
Hey, listeners. So normally we'd end this episode here, but the Tech Week attendees asked Jay some really great questions, questions I wish I'd asked, so he wanted to include those here. Enjoy.
- [44:31]
B
Hi, Jay. Thank you so much for the great presentation. My name is Trixie Appiato. I work for Willis Towers Watson, an insurance broker. I help CISOs identify and quantify their cyber risk so they can prepare for threats before they happen. And so quantum threats keep me up at night. You mentioned so many good problems that quantum can solve. It can also break encryptions in our classical computer systems. So what safeguards or policies do you implement into your teams to build quantum capabilities responsibly? And what can we do for people in this room as builders and users to secure our data and systems before quantum computers become more energy efficient, cheaper, and more available? So it's a great question. So, yes, one of the algorithms for quantum computing is to break our traditional encryption. So at IBM Research, we were aware of this from day one one, we've come up with algorithms that we believe and have very strong evidence will not be broken by a quantum or classical computer, and NIST has selected them. So first, the scientific technical question, security is saved. There are algorithms that exist that we can implement that neither a quantum or classical computer can break. So the technical answer is, we're all okay. The more complicated answer is a social and society answer. Encryption was Built in classical computing in a way that was never thought of, of being upgraded. It's mixed everywhere. Some of it is downstream, some of it is like software that you may use, some of it is software that you've developed. And I get that. If you've got a product and you want to have it secure for the next 10 years, you probably want to think about how you're going to upgrade it. Or if you have data that needs to be secure for the next 10 years, it needs to upgrade to new encryption. So the real challenge is more of a social business problem of how do we actually transition from old encryption to new encryption knowing this is going to happen. So we at IBM have been very proactive on this. We've developed tools where we can determine where encryption is used. We've developed tools which can show you how to replace it. And we early on have made sure the mainframe when we made these algorithms. So I think it was Z16 that was the first version of the mainframe to have these quantum safe algorithms implemented. So my answer to your question is yes, there's a real problem, but it's not a technical problem, it's a social and business problem. And I'm not minimizing that. I understand that that is a lot of work. You need to start now, you need to come up and do a, you need to make it part of your IT transformation. You need to get onto it. And I realize, I realize it's not going to take zero time because it's not an easy problem to do. So the short answer is one, we developed algorithms that we can't and we're developing tools to help you in that transformation. Thank you so much. Thank you. My name is Emma, I'm a product manager at Expedia working on software side of things. My question is around the non technical roles outside of the researchers, the mathematicians, the builders, how can the rest of us, whether it be policymakers, those in the legal fields, those thinking about what use cases Quantum can solve for in the future. What should we be thinking about and how can we prepare for that? It's a good question. I think this is part of the requirement of the scientists to being able to articulate where they are. We need a forum for those type of discussions. I think a lot of this can fit within the forums that we already have for classical and AI and I think we need to just be asking how do we actually bring them into them? Because I don't think of Quantum as a replacement of compute, I think of it as an accelerator. That expands what is possible. And I think we can ask those questions in those forums. Are we doing enough now? I think I agree with you. No, I don't know the answer to it. I think it's a really interesting perspective because those existing forums do start to bring in those other fields as well. So it could warrant the same sort of discussion and perspectives. And I understand those forums right now. AI is probably dominating, and it should be like we are going through a period of time where AI is impacting society, technology is impacting society in big ways. So I totally understand that most of their focus should be on AI, but we should start to ask, where is Quantum in that as well. Hi, I'm Gobi and I'm a graduating PhD student at Northwestern and also a member of South Park Commons, which is a fund here. You mentioned earlier that some problems are best solved by classical versus some problems are best solved by quantum. When we're thinking about this, if we're not experts in quantum, but we're thinking about this from an AI perspective, could you just clarify, when we think about quantum, what is deterministic and what is not deterministic? I think the future of computing we've got to get our heads around is that not everything is deterministic and it's much more going to be probabilistic. How do you handle error bars? How do you put confidence? I think a lot of those questions which you're referring to in AI are going to completely apply in quantum. Quantum. I actually think it's a mistake to compare AI versus quantum. I actually think of quantum as much. It's quantum versus classical compute, and AI is going to come across on top. So as we go forward and we get a better understanding, I'm not going to say quantum is going to replace the classical compute that enables AI, but I think some of the math you do in AI will be able to go to both. So what can we formally prove? I can come up with a problem where I take a circle and I color half of it red and half of it blue. And then I say I'm going to apply an operation that takes those dots. Make it say, let's say 10 dots over here red, 10 dots over here blue, and I'm going to wind them around many, many times. I can then show you that if you feed that into a classical computer, it's a classical random number generator, you can give yourself as much data as you want. You will never be able to say, did the red come from the left side or the right side? You would take infinite data like you would have to break a classical random number generator. I can show you a quantum algorithm that can do that deterministically. So where we're thinking is when the data appears to be completely unstructured or you looks essentially like a complete random number are to the classical methods, there are quantum methods that can actually potentially find that structure.
- [51:44]
A
That's it for this episode of Smart Talks with IBM. If you haven't already, be sure to check out my conversation with IBM Chairman and CEO Arvind Krishna. And stay tuned. Another episode is coming soon. Smart Talks With IBM is produced by Matt Romano, Amy Gaines McQuaid, Trina Menino and Jake Harper, engineering by Nina Byrd Lawrence mastering by Sarah Bruger music by Grammascope strategy by Tatiana Lieberman, Cassidy Meyer and Sophia Durlon. Smart Talks With IBM is a production of Pushkin Industries and ruby studio at iHeartMedia. To find more Pushkin podcasts, listen on the iHeartRadio, Apple, Apple podcasts or wherever you listen to podcasts. I'm Malcolm Gladwell. This is a paid advertisement from IBM. The conversations on this podcast don't necessarily represent IBM's positions, strategies or opinions.