TechRepublic’s Karen Roby spoke with Christopher Savoie, CEO and co-founder of Zapata Computing, a quantum application company, about the future of quantum computing. The following is an edited transcript of their conversation.
SEE: The CIO’s guide to quantum computing (free PDF) (TechRepublic)
Christoper Savoie: There are two types of quantum-computing algorithms if you will. There are those that will require what we call a fault-tolerant computing system, one that doesn’t have error, for all intents and purposes, that’s corrected for error, which is the way most classical computers are now. They don’t make errors in their calculations, or at least we hope they don’t, not at any significant rate. And eventually we’ll have these fault-tolerant quantum computers. People are working on it. We’ve proven that it can happen already, so that is down the line. But it’s in the five- to 10-year range that it’s going to take until we have that hardware available. But that’s where a lot of the promises for these exponentially faster algorithms. So, these are the algorithms that will use these fault-tolerant computers to basically look at all the options available in a combinatorial matrix.
So, if you have something like Monte Carlo simulation, you can try significantly all the different variables that are possible and look at every possible combination and find the best optimal solution. So, that’s really, practically impossible on today’s classical computers. You have to choose what variables you’re going to use and reduce things and take shortcuts. But with these fault-tolerant computers, for significantly many of the possible solutions in the solution space, we can look at all of the combinations. So, you can imagine almost an infinite amount or an exponential amount of variables that you can try out to see what your best solution is. In things like CCAR [Comprehensive Capital Analysis and Review], Dodd-Frank [Dodd-Frank Wall Street Reform and Consumer Protection Act] compliance, these things where you have to do these complex simulations, we rely on a Monte Carlo simulation.
So, trying all of the possible scenarios. That’s not possible today, but this fault tolerance will allow us to try significantly all of the different combinations, which will hopefully give us the ability to predict the future in a much better way, which is important in these financial applications. But we don’t have those computers today. They will be available sometime in the future. I hate putting a date on it, but think about it on the decade time horizon. On the other hand, there are these nearer-term algorithms that run on these noisy, so not error-corrected, noisy intermediate-scale quantum devices. We call them NISQ for short. And these are more heuristic types of algorithms that are tolerant to noise, much like neural networks are today in classical computing and [artificial intelligence] AI. You can deal a little bit with the sparse data and maybe some error in the data or other areas of your calculation. Because it’s an about-type of calculation like neural networks do. It’s not looking at the exact answers, all of them and figuring out which one is definitely the best. This is an approximate algorithm that iterates and tries to get closer and closer to the right answer.
SEE: Hiring Kit: Video Game Designer (TechRepublic Premium)
But we know that neural networks work this way, deep neural networks. AI, in its current state, uses this type of algorithm, these heuristics. Most of what we do in computation nowadays and finance is heuristic in its nature and statistical in its nature, and it works good enough to do some really good work. In algorithmic trading, in risk analysis, this is what we use today. And these quantum versions of that will also be able to give us some advantage and maybe an advantage over—we’ve been able to show in recent work—the purely classical version of that. So, we’ll have some quantum-augmented AI, quantum-augmented [machine learning] ML. We call it a quantum-enhanced ML or quantum-enhanced optimization that we’ll be able to do.
So, people think of this as a dichotomy. We have these NISQ machines, and they’re faulty, and then one day we’ll wake up and we’ll have this fault tolerance, but it’s really not that way. These faulty algorithms, if you will, these heuristics that are about, they will still work and they may work better than the fault-tolerant algorithms for some problems and some datasets, so this really is a gradient. It really is. You’d have a false sense of solace, maybe two. “Oh well, if that’s 10 years down the road we can just wait and let’s wait till we wake up and have fault tolerance.” But really the algorithms are going to be progressing. And the things that we develop now will still be useful in that fault-tolerant regime. And the patents will all be good for the stuff that we do now.
So, thinking that, “OK, this is a 10 year time horizon for those fault-tolerant computers. Our organization is just going to wait.” Well, if you do, you get a couple of things. You’re not going to have the workforce in place to be able to take advantage of this. You’re probably not going to have the infrastructure in place to be able to take advantage of this. And meanwhile, all of your competitors and their vendors have acquired a portfolio of patents on these methodologies that are good for 20 years. So, if you wait five years from now and there’s a patent four years down the line, that’s good for 24 years. So there really is, I think, an incentive for organizations to really start working, even in this NISQ, this noisier regime that we’re in today.
Karen Roby: You get a little false sense of security, as you mentioned, of something, oh, you say that’s 10 years down the line, but really with this, you don’t have the luxury of catching up if you wait too long. This is something that people need to be focused on now for what is down the road.
SEE: Quantum entanglement-as-a-service: “The key technology” for unbreakable networks (TechRepublic)
Christoper Savoie: Yes, absolutely. And in finance, if you have a better ability to detect risks then than your competitors; you’re at a huge advantage to be able to find alpha in the market. If you can do that better than others, you’re going to be at a huge advantage. And if you’re blocked by people’s patents or blocked by the fact that your workforce doesn’t know how to use these things, you’re really behind the eight ball. And we’ve seen this time and time again with different technology evolutions and revolutions. With big data and our use of big data, with that infrastructure, with AI and machine learning. The organizations that have waited generally have found themselves behind the eight ball, and it’s really hard to catch up because this stuff is changing daily, weekly, and new inventions are happening. And if you don’t have a workforce that’s up and running and an infrastructure ready to accept this, it’s really hard to catch up with your competitors.
Karen Roby: You’ve touched on this a little bit, but really for the finance industry, this can be transformative, really significant what quantum computing can do.
Christoper Savoie: Absolutely. At the end of the day, finance is math, and we can do better math and more accurate math on large datasets with quantum computing. There is no question about that. It’s no longer an “if.” Google has, with their experiment, proven that at some point we’re going to have a machine that is definitely going to be better at doing math, some types of math, than classical computers. With that premise, if you’re in a field that depends on math, that depends on numbers, which is everything, and statistics, which is finance, no matter what side you’re on. If you’re on the risk side or the investing side, you’re going to need to have the best tools. And that doesn’t mean you have to be an algorithmic trader necessarily, but even looking at tail risk and creating portfolios and this kind of thing. You’re dependent on being able to quickly ascertain what that risk is, and computing is the only way to do that.
And on the regulatory side, I mentioned CCAR. I think as these capabilities emerge, it allows the regulators to ask for even more scenarios to be simulated, those things that are a big headache for a lot of companies. But it’s important because our global financial system depends on stability and predictability, and to be able to have a computational resource like quantum that’s going to allow us to see more variables or more possibilities or more disaster scenarios. It can really help. “What is the effect of, say, a COVID-type event on the global financial system?” To be more predictive of that and more accurate at doing that is good for everybody. I think all boats rise, and quantum is definitely going to give us that advantage as well.
Karen Roby: Most definitely. And Christopher, before I let you go, if you would just give us a quick snapshot of Zapata Computing and the work that you guys do.
Christoper Savoie: We have two really important components to try and make this stuff reality. On the one hand, we’ve got over 30 of the brightest young minds and algorithms, particularly for these near-term devices and how to write those. We’ve written some of the fundamental algorithms that are out there to be used on quantum computers. On the other hand, how do you make those things work? That’s a software engineering thing. That’s not really quantum science. How do you make the big data work? And that’s all the boring stuff of ETL and data transformation and digitalization and cloud and multicloud and all this boring but very important stuff. So basically Zapata is a company that has the best of the algorithms, but also best-of-breed means of actually software engineering that in a modern, multicloud environment that particularly finance companies, banks, they’re regulated companies with a lot of data that is sensitive and private and proprietary. So, you need to be able to work in a safe and secure multicloud environment, and that’s what our software engineering side allows us to do. We have the best of both worlds there.