Quantum One Step at a Time EP-01: Chasing Quantum Advantage

Hey everyone, I'm Alan Shimmel, that's John Willis, and you're watching Quantum, One Step at a Time. This is a new podcast that John and I kind of cooked up. For those who aren't familiar with quantum, we don't expect you to be a quantum expert to come on here.

We're actually going to start at a pretty basic level and build up. But we're going to do this every other week and talk about what is one of the hottest topics. I'm not going to say it's as hot as AI is, because God knows what's going on with AI.

Yeah. But it's a hot topic. John, welcome.

It's great to be on here with you. You know I love doing kind of videos- Yeah, that's great ... and just talk tech with you.

Yeah. It's great. Maybe quantum will clean up all this AI craziness that's going on here, but definitely- Well, there is sort of a relationship there, right?

Yeah, there's definitely, oh, there's incredible opportunities, you know, of the hybrids. Unfortunately, I tend to think of it as you ever see those diagrams they draw of one star circling another star, and one star's gravity is pulling- Mm ... and that's actually a good thing to do in a quantum podcast.

Yeah. One star is pulling the star mass out of the other star, and eventually they supernova or whatever. I hope it doesn't come to that.

Yeah, no, I think it's certainly interesting, I think the first real breakthroughs in quantum are going to be these hybrid scenarios, and we can talk about some of those where, I think everybody's been looking for the elusive, what they call quantum advantage, which sort of is a nonsense term, but it's a marker, and it's this idea where a quantum can solve the kind of problems that classic kind of von Neumann architecture computers can't. And so instead of looking at a classic computer can't do this, a quantum computer can do this, it's going to be more like parts of it might be done by the classic computer, part of this might be done by neural networks, parts of it will be done by quantum algorithms, and I think that's where the really interesting first breakthrough is going to happen. Thinking of that model, John, I think back to my high school days where we had auto shop.

Remember when they taught auto shop in school? Yeah. And they explained to you how a V engine works with the timing chain and everything, so that the pistons fired at the right time and all that.

Mm-hmm. To me, that's the problem with that model. And I do agree that that is probably going to be the first thing we do, but I think the problem you're going to have is a timing chain issue, right?

Where the quantum computation gets done here, the classical computing gets done there, a GPU kind of thing, or neural processor does here. How do we combine that information in the right order at the right time to come out with a standard- Yeah ... something useful.

This is one of the more interesting things that's going on right now, the whole system engineering challenge. So there's really two ways to sort of break that up. One is when I talk about the hybrid, I'm talking about there are examples where IBM recently did this 1,200 atom protein thing that probably within our lifetime wouldn't be solved by a quantum algorithm.

But they took the math and they did some of the math with a CPU, some of the math with a GPU, and some of the math with a QPU. So that allows it to sort of segregate the algorithm, roll it up, and you don't have the sort of timing of what we learned in auto shop, right? Mm-hmm.

The second problem is the more interesting one, which is how you work with quantum algorithms that have incredible sensitivity, right? One of the biggest problems is the error correction processing. These things can run for 10 seconds right now.

So to solve the real hard problems, they're going to have to run for minutes or maybe hours, who knows? And the question is, there's so much interference patterns in the way a quantum computer works, that the timing becomes how do I take the output of a quantum algorithm and then first clean it up? So there's strange system engineering challenges, like the quantum processor, it might be cryogenic, might be superconductor or photonic, will then sort of have to offload to maybe an FPGA to do some cleaning and then to a GPU to use accelerators for error corrections.

So the assumption is right now that everything you do in a quantum processor has a lot of, in my terms, not a quantum physicist terms, pollution in it, noise, interference. Well, but this goes to the whole fault-tolerant- That's okay ... qubit kind of stuff, right?

And I feel, John, we jumped right in. Yeah, we did. We did jump right in.

Let's set a little table here. Look, we're all hearing a ton about quantum computing, right? The government is putting real money behind this.

We're hearing about breakthroughs coming every day. But I did a report on this two months ago. The quantum computing industry probably did about $550 billion last year.

That's pretty damn good considering we don't have a working model, a working computer, right? A lot of that is probably in the post-quantum algorithm security, and we could talk about that later. But what's holding us back, John, for people who aren't quantum experts, what's holding us back from having a real quantum computer?

Well, there are some real quantum computers. The question right now, so there are a fair number of vendors that are actually selling quantum computers. The real question right now, well, there's sort of this enthusiasm, and it's intelligently based, reality based enthusiasm.

In other words, the math tells us that if we can get to this sort of fault tolerances, so basically in general that what people would sort of call quantum advantage, it varies, some used to call it quantum supremacy or whatever, is this idea that you can have a large enough number of qubits that are fault tolerant, that can run for a significant amount of time. And those are the three really hard problems that the smartest people on the planet are trying to figure out. Now, the reason there's so much money getting dumped into it is because the people who are dumping money into it believe when that happens, we're going to basically see things like, not only the crypto, right, which is a whole, but the sort of quantum proofing and quantum cryptology and all that, but the real interesting thing are going to be being able to design new types of plastics, being able to do drug discovery at a level, solving logistic problems that were just not even fathomable with classic computers.

Everybody in this space believes that will happen. And so that's where all the money's going because again, think about DNA. I always say the next one.

Think about in the early days of DNA signatures. There's going to be a patent frenzy when the first people can patent these algorithms, that literally design, it's like the Star Trek, the new type of plastic, or these things probably- Transparent aluminum. Yeah, right.

These things will probably, or here, a good example of being worked on right now by companies like Volvo and BMW is what if I can create a battery? What if I use quantum algorithms to design a battery that lasts ten times longer and is ten times less time to charge, or 100 times longer? But the problem here right now, which is really interesting, is it's sort of a race between what AI can do, but we haven't been able to prove.

Quantum is like, well, it can do these things. Yesterday AI couldn't. Now AI can do some of these things, right?

So there's an interesting sort of race of, and the question that still remains is there's no solid proof right now that a working quantum computer Can solve problems better than a working neural network. So that's the sort of catch-22 that's going on right now, but again, I think all that money's going in because the science sort of tells us that there will be a point where it will do things that a neural network-- Because remember, a neural network, even a GPU, is really a von Neumann architecture, right? Right.

It's a bit on and off, bits and bytes. Not to go too far into the physics of this, but the idea that you can have instead of two properties in a bit, you can have multiple properties in what's called a qubit, and it just sort of expands the ability to do compute in a way that is just- And that's the thing, John. Yes, you could take a neural network and a classical computer kind of thing, and if you have enough horsepower and enough time and enough GPUs and enough memory, you could solve some of these really complex issues.

And you may be able to solve them in a human timeframe even. But the amount of energy expended to do such a thing is going to be prohibitive. That's really the issue, right?

It's a cost function. I think it- Where, yeah, I mean, with quantum, you don't have that. Well, I think you do.

Yeah, I think you will. I mean, to be honest with you, I don't know that that's a solve. I've asked quantum physicists that question.

I've gone now the second year to the IEEE Quantum Computing Week conference, and last year when I went, I asked a bunch of them, and it's not clear. Because there's a lot of work in electricity and power that goes into that error correction and fault tolerance, and so there is this idea of quantum at the edge. Now, that I might be able to, that could sort of maybe solve the large data center quandary right now.

But I don't know that that's clearly that quantum advantage is that it's going to be better on the Earth. I'm not sure that's a clear picture. But again, in fact, there are mathematical type problems that people believe that probably can't be solved by a classic computer in a timeframe.

There's a lot of hype, right? Google supposedly sounding like they broke 256 Elliptic Curve Cryptology, that big announcement this year, which is scaring everybody because Bitcoin has a 10-minute lock and all that sort of stuff, right? Well, first off, it was a simulation.

Yeah. They didn't actually break it. But the fact that they proved that they could break it in under a minute, it means something, right?

And the sort of comparison is, again, I don't know the next number, but we're talking about thousands of years on a classic computer, like the math. And so that, again, that's where I think the real hype is solidly based, in the fact that right now the math tells us that the binary or the von Neumann architecture will take thousands, sometimes millions of years to solve some of these problems, and the math tells us that with a quantum process, it could be minutes or hours. So, the question is what are the useful things that get solved?

What are the things that AI can sort of do in the meantime? So again, if you look at quantum as just a computational force multiplier, let's use that term. Then you got to compare it to other force multipliers, to other brute force kind of methods, including AI and ever bigger machines and ever bigger data centers.

But there are just certain issues, problems, solutions waiting to be found that just lend themselves to a quantum architecture type of solution. And what I hear you saying, John, and if I'm wrong, correct me, is a large part of the overhead here is not going to be on the quantum computer itself doing, or the qubits itself doing the computation. It's going to be on the cleanup and making it digestible.

That's right. Yeah. Yeah, I mean it's just- It's different.

No, you're spot on. And here's the sort of the caveats, if you will, is that quantum is not brute force. It is literally this idea that you kind of simulate, or you don't simulate, you actually, depending on the technology, the sort of superconducting electrons or photons for photonic, you are literally creating quantum states.

And the way I like to describe it is if you were going to write an algorithm, even sort of neural network based or just the smartest computer programmers on the planet in parallelism, and you wanted to figure out the proper way to get out of a maze, it would still be instruction based, right? You could paralyze it, you could literally say, "I went this branch, okay, let me backtrack. I went this branch," and then eventually you calculate which is the proper way to get out of a maze.

In a quantum algorithm, it puts the algorithm in a quantum state, like a wave, and the wave has the interference. And so what happens is the end state, it's like one fell swoop. It's like pushing a wave through the maze.

And what you're getting is the highest probability of the wave. Think about throwing a couple pebbles into water. The one that went the furthest, that the other ones sort of canceled out the waves and some of them increased, then that one, and then it's all probability based.

But the math works. So it isn't a brute force, and that's where you get this incredible power to be able to solve these, again, that even in your fastest paralyzed compute could take thousands of years. Theoretically, if you could do the error correction, get rid of the noise, the interference, and all those things for a length of time, you remember, some of these algorithms run for 10 seconds, and they deteriorate to a point of that the data's no good, right?

Mm-hmm. And here's the thing, it's not just dust or vibration, it's things like dark matter. When you're in the quantum state, there are unknown things.

And so what has to happen is, there's this idea, it's kind of interesting. I've been learning a lot more about this, but basically, people talk about bit qubits, "Oh, we're going to do the 100 qubit. " The real question is not really the qubits, it's the modality, right?

The photonics or the ionics or the sort of superconducting electron. It is that there's a ratio between how many physical qubits do I have to have to do this algorithm that can be error corrected by logical qubit. " But basically, you would probably have to have, in some cases, 100 physical qubits to one logical qubit.

And a logical qubit is basically saying, I'm making a blanket assumption that a high percentage of this is noise from the physical qubit. So maybe I could've got away with it with 10 physical qubits, but I'm going to need 100 of them. And because there's a high percentage of them are going to have all this interference.

So then I've got these crazy algorithms that run on accelerators or FPGAs, depending on what you're trying to do, to try to get the real signal versus... And for the quantum people out there, feel free to yell at me, get mad at me. I'm not a quantum physicist, so I'm using my words to describe these things.

But basically, it's that hard math Which might even be harder than the quantum math, but the hard math is how do you accurately decipher out all of that interference or that error to get the real value? And remember, there's a window of it just starts, again, my word, decaying, maybe 10 seconds, 15 seconds. It was too much noise.

Right? No, there is. That's right.

We certainly need breakthroughs there. Well, and this is where there's tremendous-- So here's the thing, right? I think there's tremendous opportunity for engineers right now, because think about all the science of Nvidia and what we've done with neural networks.

This just opens up an incredible world of a different pipeline of timing, latency, network communication. It's an unbelievable opportunity for engineering and for operations, because once you have a data center at scale that has CPUs, GPUs, FPGAs, and QPUs all in the same structure, think about let's use the lowercase D DevOps around all the work that it's going to take to operationally manage that. Oh, absolutely.

But you know, John, when Jensen Huang and his two co-founders got together in that Denny's in Silicon Valley to come up with a graphic processor and they went-- I don't know how much you know about this story, but I did it as part of my research for the book. They originally came up with a graphics processor that thought that the dominant form of the shape of the graphics would be rectangular, not triangular. Okay.

And so their first chip was a disaster because it worked for rectangles, not triangles, and the gaming industry was moving to triangle-based design. And they were literally 30 days out from not making payroll when they- Yeah ... switched to triangular.

It's the same kind of thing here with quantum. Well, it can it be the same kind of thing with quantum where we sort of almost stumble into the real payoff by doing some just basic science, if you will. My issue is, and I think it's a great way to kind of wrap up the last five, six minutes here, is the amount of money, I'm going to return to the money, the amount of money we're pouring into this right now.

Yeah, and again, I think- Versus the return. We don't even know if it'll actually give us the returns we want. We don't even know where to look for the returns.

No, I think they know where to look. They do know where to look. But again, they know that the math works.

Again, these vendors, including IBM, but also IONQ and PsiQuantum, and the whole damn list, they all will say probably 2030 is this quantum advantage. What that means is some form of fault tolerance that works, that solves the problems that classic computers can't. Now, they're all vendors, but again, they're pretty smart scientists.

When you listen to somebody who talk, it isn't all marketing hype. There's a belief. Now, again, can it happen?

I think the thing that what really sort of... I lost my thought there for a second, but what really has to sort of pan out is the engineering to make that. And this gets really interesting because the engineering isn't just error correction, it's how do you put a refrigerator in a data center?

How do you get networking from a sort of a FPGA to a quantum refrigerator based some are room temperature, like photonics are room temperature. How do you do the networking? How do you do the memory management from a QPU that might be in sort of, almost sub-zero or absolute zero, sorry, absolute zero temperature, like 450.

Down to absolute zero. Yeah. Right.

Yep. And how do you sort of do a networking memory exchange between an FPGA and then an FPGA to a QPU all within a time window that is acceptable where right now the QPU- Well, that's my timing chain issue right there. Yeah.

No, that's what I was saying. Yeah, no, that is exactly where, but this is the engineering challenge, and there's vendors that now are focusing on that part of the stack. So now what's getting really interesting is that a fair number of vendors are like, "That's what we're doing.

We are basically decoders. " You got to remember, Google's not really selling quantum computers to anybody, right? Amazon, OEMs like Rigetti and stuff like that, right?

IBM definitely has quantum computers, and then there's a bunch that are selling quantum computers. " And they've got patents on it. So then these guys come in and say, "Well, wait a minute.

" And there's some success, some wind tunnel stuff, some battery, nothing again. I didn't get one person, including vendors, to say that quantum has definitely solved the problem that a classic computer can't. But there is interesting activity going on with manufacturers, particularly auto manufacturers for batteries, wind tunnel simulation, jet engine stuff.

They are playing around with those kind of algorithms. And again, I think back to what I said earlier, there'll be some hybrid examples that we will see some interesting... And again, if you follow some of the marketing hype there, it sounds like...

I put it this way, when you go to the quantum conferences that I've been to, Nvidia is all over the place. Mm-hmm. Boeing, car manufacturers, chemists, chemical engineering is like all over- Look, big pharma- Yeah, big pharma ...

is all over this. But John, here's kind of the thing about it. In many ways, no one's put their finger on the killer app for quantum.

And the other thing, and I'm not against it, don't get me wrong, I'm not trying to come off as a troglodyte here or something, but a lot of the gee whiz of quantum is almost like pure science, right? Really, it's almost research for research sake. And can it help us predict the weather better?

Yeah, probably can. Can we use that? You bet we can.

But when we look historically, who funds pure science? A lot of it gets funded by the government. A lot of it gets funded by, we used to have a university, well, we still do, we have the university system here in the US, where a lot of this pure research, pure science gets done.

We seem to now have moved in quantum to a different state. See what I did there with quantum to states? Yeah.

There you go. But we moved to a different state where it's private companies that are driving this basic research. That always scares me a little too, John, because then someone is going to own the patents to it, and someone is going to own it.

Yeah. I think all the quantum, again, there's the crypto, right, which is important. Everybody has to be investing in this.

I know you guys just wrote a really good paper and an analyst covered it, it covered exactly what you need to think about. You need to be quantum-proofing. You do all that stuff right now.

But again, the sort of gold rush is going to be the patents on optimization And again, I think that's why the private sector is putting money into this because I think-- And then the other side of this is there's a fair amount of government investment here, right? Because DoD, and by the way, China is, last time I checked, almost 10X the US in funding quantum research, right? So we have a quantum gap?

Big time. And it's just sort of a disgusting gap of what the difference is. So, yeah.

But again, I go back to the-- I think the belief is, so let's take the AI scenario. If we went back to 2015, AI was almost dead. It was sort of ML was herking around.

You had the financial, by the way, the banks are working really hard to try to figure this out because, like what's it called, quantum networking, where the speeds of a confirmed transaction could be, think about high frequency trades at an order of magnitude confirmation faster, right? But the banks are putting-- But you didn't, in 2015, '16, it wasn't until sort of the 2017 attention is all you need that things started rolling downhill positively. There was a big thought about, I mean, we hired a woman at Docker when she was an ML/AI person making $65,000 a year, saying she had a hard time finding a job, right?

But it wasn't, and now, all of a sudden, now the AI experts are cashing checks, right? Right. So I think there's that type of-- Now, the question is the window.

Is it five years? Is it 10 years? Or is it 20 years?

" Like this IBM, where they did a 1,200-atom protein simulation, and they- Yeah ... again, even a quantum computer, probably wouldn't happen in the next 20 years. But they figured out they could split the math up into things that were CPU bound, GPU bound, and QPU bound, and again, a simulation, but, and that all of a sudden, they were able to solve it in, I think, minutes or, you know.

Right. So I do think, the splitting up and then the reassembling is where we're going to spend a lot of our efforts in the years. I don't think anyone wants to hear 10 years or 20 years.

I think we've heard five to 10 years for so long. I think people, IBM's planted the flag for 2029. I think Google- Yeah, but that's the running joke.

But the running joke is quantum advantage is always five years away. I mean, that's everybody, every time somebody talks about '29, '30, but like I had one of the chief engineers of IonQ and they just had a great announcement where they can do error correction on a CPU, right? That could be huge.

That solves a lot of networking, memory bandwidth problems, because right now, most of the vendors are having to do error correction on GPUs or FPGAs. " And again, he's a vendor, but brilliant quantum physicist who believes in his heart of heart, it is probably going to be around 2030 where we see the first couple of advantages. But if even- 100% ...

so within 10 years, it's still, I know you say nobody wants to hear 10 years, but again, if you could get the patent on a new plastic, what would that be worth? What if you can figure out how to land planes? Well, you got a patent on a quantum algorithm that lands planes at the same angle, and saves billions in fuel costs.

What if you own that? You have now a toll booth on every airline that tries to land a plane. And I think this is what- The problem is a patent's good for 17 years.

If it's 10 years out, I only got seven years left. Yeah. Well, there you go.

There's that too, right? Well, but there's tricks there, as we both know, right? No, absolutely.

This is patent stuff, my dear. Yeah. Hey, but John, we're over time.

All right. We're going to try to keep these to 30 minutes because we understand audiences' times are valuable. But you know what?

This was a good first thing to just throw some stuff on the wall, John. I want to come back. Let's talk next time more about what we mean with this fault-tolerant cubic qubits.

Let's talk about the timing chain issue, as we called it here today. Let's talk about what are the real, or what we believe to be the real things it solves. I'm looking forward to the discussion.

And I just feel like we need to mention, John, you're working on a book here. Yeah. Well, that's kind of why I got involved in this.

About 10 years ago, when my son was going off to an engineering degree, I wanted to see if quantum was something he should get into, and at the time, everybody told me, "Don't even send him there. " Last year, I started seeing all this hype, but interesting hype where, okay, something's happening here, and I've just finished my AI book, history of AI book, and I thought, let me try writing a book about the history of quantum computing. And I'm pretty close to having my first draft done here, maybe hopefully in a month, maybe by the end of the year, definitely.

So, yeah, so that's got me sort of whole-- And just to add one little bit icing, there was interest faded, then I found interesting, started to write a book. Now I'm looking at like, is there business opportunities to talk to CIOs about how much they should be investing, and a small piece. But don't not ignore this space.

And I saw a couple of banks this week that said exactly the thing. There was a large financial bank that said they've hired two quantum students, quantum computing students out of universities, and didn't put them in crypto and put them in the business side, right? So that's the kind of thing that I think is really interesting, is start, make the assumption that it will happen.

Start putting some talent in your roadmap. Don't spend out, but like just enough talent so you're hedging your bet. And I think there's a good CIO conversation to be had.

And so I'm exploring, does anybody want to listen to me on this? But yeah, that's from some notes. Very cool.

All right, hey, we'll be back in two weeks, but right now we can call an end to our first episode here of Quantum, One Step at a Time. There you go. I'm Alan Shimmel.

He's John Willis. We'll see you next time. Thanks, Alan.

Thanks, everybody.