- il y a 23 heures
What’s Next for Quantum Computing
Catégorie
🤖
TechnologieTranscription
00:00Sous-titrage Société Radio-Canada
00:30C'est parti !
01:00So don't hesitate, this is how you do it if you're not familiar with the process already, you connect to
01:05the VivaTech platform on the app, select the tab in the drop-down menu, titled Interactive Sessions with Slido, and
01:15then you go to Select, Stage 3, and you're ready to roll.
01:18Put in your questions, your comments, and our moderator will make sure that you get some answers to your questions,
01:23so don't be shy.
01:24We want this to be as interactive as possible.
01:27So, having set the scene, please join me in welcoming Jeremy Kahn, Senior Reporter at Fortune Magazine, and our wonderful
01:35panellists.
02:05Thank you all for being here.
02:06I'm Jeremy Kahn, I'm a Senior Writer at Fortune Magazine, and I have with me today Georges Olivier Rommand, who
02:14is the CEO at Pascal,
02:15Joe Fitzsimmons, the founder and CEO of Horizon Quantum Computing, and Manjri Chandran Ramesh, who is a partner at Amadeus
02:24Capital Partners, and we're going to be talking about the future of quantum computing.
02:28But I don't know how many of you are actually familiar with quantum computing, so first, we're going to ask
02:33George to hopefully give us a 30-second definition, what is quantum computing, George?
02:4030 seconds?
02:41Well, you can take a minute.
02:43No, no, I will take 30 seconds, just to make it short, otherwise it will last forever.
02:48I mean, quantum computing is another way of doing computing, basically, leveraging the strange and funny effects of quantum mechanics.
02:56And it's not continuous improvement of our current technology, it will really be a disruption.
03:02I mean, some computing tasks that are today intractable, but we cannot imagine to implement them on our current technology,
03:09will become a reality thanks to quantum computing.
03:12And I think that quantum computing is the future of AI, in a nutshell.
03:16Very good.
03:17Joe, do you quibble with that definition, or is that all right?
03:21Not at all.
03:22I mean, ultimately, it's harnessing physical effects to do more general computation.
03:27So it's an extension of computing, and if we get it right, we can really have a next wave of
03:34computing that goes far beyond what's currently possible.
03:36But there's a long way to go still.
03:38Great.
03:38I think you'll have to keep the mic closer to you, Joe, to make sure everyone can hear.
03:42So one question is, what can you do with quantum computing today?
03:46There are, you know, quantum computers that exist.
03:50What can you do with them today?
03:51What practical purpose do they serve?
03:53And I know there's some difference of opinion among our panelists about this.
03:56But Joe, what's your view on this?
04:00Quantum computing today, the reality is we're not at a point yet where quantum computing is useful.
04:05So there have been quite a lot of efforts to build quantum processors, and it's very clear that there's concrete
04:12progress being made in that direction.
04:14We're getting to more and more advanced systems.
04:16And we've started to cross the line where quantum computers have become hard to simulate.
04:21So we're seeing systems that are hard to simulate using conventional computers, but to actually use them to solve a
04:28real-world problem that has real-world consequences for industry,
04:31we're not quite there yet, but we hope it's coming soon.
04:34Great. George, I know you have a slightly different view on this.
04:36What do you think we can do today with quantum computers?
04:40I have a different opinion, a slightly different one.
04:42It may depend on what we're talking about, but today at Pascal, we have implemented real-world use cases with
04:49real-world data,
04:50and we were able to benchmark the results against the classical set of the art.
04:53We are not yet better than the classical set of the art, but we are matching the performances.
04:58And what are some of those use cases where you've had that equivalent?
05:01It was in finance, this one.
05:04We implement something else also in the sector of chemistry, and I think that chemistry will be the first field
05:10to benefit from quantum computing.
05:12And, Mandri, when you're looking at this as an investor, what are you keeping your eye on?
05:18Because there has to be, I assume, to invest in these businesses, a sense that there's, within some period of
05:24years, going to be demand from this from customers.
05:27But if the systems are still too primitive to actually be used in many practical cases, that market might not
05:34be there.
05:34So, how are you kind of viewing this ecosystem as it develops, and how investable is it from your perspective?
05:41Thanks, Jeremy.
05:43Hi, great to be here.
05:44So, I'm a partner with Amadeus Capital Partners, and our quantum portfolio includes companies like Riverlane, which is doing error
05:52correction for quantum computing,
05:54New Quantum, which is doing quantum networking,
05:56and Photonic, which is a quantum computing company, as well as a French company called CryptoSense, which got acquired by
06:04Sandbox AQ.
06:06And I started investing in quantum technologies in 2017.
06:11So, even before, it was actually cool in the investor world to look at this space.
06:15And it was hard, you know, my first investment memo had a lot of, this is a moonshot, guys, so,
06:22you know, let's be very clear here.
06:25But the advantage, you know, when you invest in deep tech is that you're expecting a longer time horizon.
06:32Not to be confused for a lack of commercial validation.
06:37Yes, we are patient, but equally, at every stage, we like to see what the customer pull can be.
06:43So, a lot of my due diligence conversations include speaking to people in the chemistry, you know, materials modeling,
06:51or the drug discovery area, or indeed the banking sector, to really see if a quantum computer exists,
06:59if there was quantum advantage, to explore with them what might be the art of the possible,
07:04and then build out a market opportunity.
07:08Right.
07:08And I know a lot of companies are looking at the technology, monitoring it, trying to figure out some way
07:13to start learning more about how they might use these systems.
07:16And I'm curious, you know, today, if you're a company and you're looking to do this, how do you make
07:21that easy for them?
07:22Joe, how do you talk to customers or potential customers about how they should be thinking about exploring this technology
07:29today,
07:29so that they're sort of ready once we do have systems, you know, that are capable of some of these,
07:35an advantage in some of these use cases?
07:37Yeah.
07:37So, that's a great question.
07:39I mean, for us, we're a developer tools company, so we're building tools to make it easier to develop for
07:44quantum computers.
07:46As we see it, there's basically two barriers to quantum computing.
07:49One of them is first building useful quantum computers, but you also need the tools to be able to take
07:54advantage of them.
07:55If you're looking at this from the perspective of a company, you know, what can it do today?
08:01Already, you need to be moving towards that direction, building up talent in that direction, building up skills in that
08:08direction,
08:08if you're in a situation where quantum computing is going to be highly disruptive to your industry.
08:13If it's only going to lead to marginal gains for you, then it's not that important to be leading the
08:19pack.
08:19But if it's something where it's going to truly disrupt your industry, then it's something where you need to look
08:24sooner rather than later,
08:25because it's not something you can do instantly.
08:28You can't instantly gain expertise in quantum computing.
08:31You can't instantly put that team together.
08:34There's a really limited amount of talent in the area that's available.
08:39So, yeah, I mean, it's a time where there's progress happening.
08:44So, really, if you're exposed, now is the time to pay attention.
08:48Right.
08:48And, George, how do you view this?
08:50And when you're talking to prospective customers, how do you get them, you know, thinking about the technology,
08:55how they might use it, or, you know, are they setting up POCs?
08:59What are they doing?
09:00Well, I think the main message I'll convey to my customers is that they have to start their quantum journey
09:05now,
09:06because we have the first quantum computers and the first applications.
09:09We were able to implement them on our processor, matching the performances, not yet beating them.
09:16But it's a matter of time.
09:17We have a strong R&D program to improve the performances of the processors, meaning that in two years, we
09:25should beat the classical technologies.
09:28So, the quantum journey starts now.
09:29They have to learn.
09:30They can start learning how to use the computer with the current technology, with the current tool we are also
09:36providing them.
09:37And this time it's useful for them just to train, to understand how it works, to build up the use
09:42cases,
09:43and think also about the scarcity of the resources, the talents and the machines.
09:49Right. Interesting.
09:50Obviously, one of the big problems for use cases today is that today's quantum computers generate a lot of errors,
09:57and the errors, you know, make the systems less practical to use for many use cases.
10:02How far away do you think we are from error correction or, you know, reducing that error rate to the
10:08point
10:08at which we really see the advantages of this technology over classical hardware?
10:13I don't know.
10:14Mandri, maybe you have a view on this, or have you seen from front of your start-up?
10:18So, I'll come to George in a minute.
10:20So, I'm not a technical person.
10:21I have my PhD in machine learning, not in quantum physics.
10:25I'm not going to make a guess on when we're going to have the quantum advantage.
10:29I don't think that would be the right thing.
10:31But what I find very interesting is that a lot of corporate teams have gone through already one cycle of
10:39building out a huge quantum team
10:42because, as Joe said, this is not going to be something that companies can take advantage of
10:47without an in-house, you know, person to actually understand the technology.
10:52But equally, we've seen where quantum teams have been let go of in the larger corporates.
10:59And then again, you know, as the progress happens in the technology, these teams are coming back into play in
11:06the corporate world.
11:08So, I think it's fair to say that, you know, everybody, whether it's the investor group or the larger corporates,
11:14we're all very closely tracking the progress.
11:17And I think, you know, we are going to see some really interesting advances very soon.
11:23But I don't think it's a good idea to actually say this will be the date.
11:26And, George, I know you wanted to get in there.
11:28And I'll go to Joe in a second.
11:30What's your prediction on when we get error correction solved?
11:32I have no prediction.
11:34It's really hard to guess.
11:35And I will not do it.
11:36One of my co-founders, Alain Spey, who was awarded the Nobel Prize in Physic last year,
11:40is stating that he will not see a quantum error corrected processor in his lifetime.
11:47That's his view.
11:48But what I know is that we do not need quantum error correction to deliver business value today.
11:53And this is a fact.
11:55This is reality.
11:56And why is that?
11:58I mean, because a lot of people do think, oh, you know, for so many things, you know, these systems
12:02are too noisy.
12:03Why is it that you think actually today we can actually do this?
12:06We can deliver business advantage even with noisy systems?
12:11Absolutely.
12:11And four years ago, at Pascal, it was a bet.
12:14I mean, we had no evidence that it will work this way.
12:17We thought that by programming the processor in a different way, we could mitigate these errors instead of correcting them.
12:24And actually, mitigation is working pretty well.
12:26I think at the end, when we compare our experimental data against theoretical one, compute on an errorless quantum emulator,
12:36there is no differences, meaning that we are able to keep and mitigate these errors.
12:40Interesting.
12:41And Joe, what's your view on this?
12:42How far are we from error correction and does it matter?
12:46So, I mean, what I should say is that error correction is already happening.
12:50We've seen results over the last number of years getting closer and closer to a fully error corrected system.
12:57We've seen demonstrations of repeated error correction.
13:01We've seen all of the building blocks of fault tolerance demonstrated.
13:04And since 2016, we've started to see coherence gain in qubits.
13:08When I say we, I mean the field as a whole, not horizon.
13:12But we've started to see the first coherence gain, qubits that are living longer because of encoding.
13:19And, you know, there is now quite a lot of progress on that front.
13:24There's certainly a lot of momentum building towards it.
13:27And to be honest, although error mitigation is extremely important, things like dynamic decoupling, decoherence-free subspaces,
13:34give a way to avoid some of the overhead of active error correction.
13:39I think, ultimately, to get to really useful applications, we're likely to have to really reduce the error level in
13:46the systems by a lot.
13:48And at the moment, current thinking is that that probably requires active error correction.
13:52Because for those applications, we know there is an enormous advantage.
13:56Whereas for the applications that are designed for near-term systems, NISC systems,
14:02these rely on variational techniques.
14:05So it's a little bit like training a neural network or something like this.
14:08You come up with some parameterized program and you tweak the parameters.
14:13It's very hard to prove whether they work or not.
14:15So, yes, you can get good results out of them.
14:18But how do they scale?
14:19No one really knows.
14:20And we won't know until we have larger scale systems.
14:23So if they work, we might see advantage with noisy systems.
14:26But we know we see advantage with error corrected systems.
14:30And we're certainly on the way towards that.
14:32Great.
14:34I want to first of all thank the audience for submitting questions.
14:36We have a lot of very good questions here.
14:38And I'm going to get to them in just a minute.
14:39But before we do, one thing I wanted to ask the panelists was,
14:42right now there's still a debate over what's the right architecture going to be.
14:47You have all these competing ways of creating qubits out there,
14:50superconducting qubits, photonics, trapped ions, neutral atoms.
14:54And I think a lot of companies are afraid that if they start learning about one of these things,
14:59they'll sort of commit to the wrong architecture.
15:00And that will not be the one that ultimately proves most advantageous
15:04and kind of becomes the standard around which the future of quantum computing is built.
15:10So my question actually, and it's also a question for you as an investor, Mandri,
15:14how do you try to sort of spread your bets or hedge your bets, as it were,
15:19across these different architectures?
15:21And then, Joe, I know you're trying to come up with software
15:24that essentially would work across a large number of these architectures.
15:28I'm curious how that works as well.
15:30But Mandri, why don't we go to you first?
15:31How do you recommend people sort of think about trying to mitigate the risk
15:37of going with the wrong architecture?
15:40So I think the entire landscape is quite nascent.
15:45Again, I don't think people can sort of call one architecture
15:49and say this is definitely going to be the winner.
15:51And indeed, you know, it's going to be the winner in a particular time frame either.
15:56I think silicon photonics looks really interesting
16:00because the scalability factor is almost a given, right?
16:05So given that quantum advantage,
16:07most of the simulations suggest that you need a million qubits or so.
16:11It makes sense that you want a scalable path.
16:15But in terms of technology readiness levels,
16:18it's probably not at the same level as superconducting or ion traps.
16:22Those are further along.
16:24What's been really interesting for us is, you know,
16:27both our investments in Riverlane as well as new quantum
16:30is these two companies work with pretty much any of the hardware architectures.
16:35We've not had to make a choice on which hardware architecture we actually work with.
16:40New quantum especially is looking at distributed quantum computing networked.
16:45So it actually is in a position to solve the scalability problem
16:49for things like ion traps and superconducting and indeed cold atoms.
16:54So our approach has definitely been,
16:56let's try and work with as many of these players as possible
17:00as the landscape matures before we take the final bet.
17:04Right.
17:05And Joe, how do you look at this problem
17:07since you're trying to work across different architectures?
17:10Yeah.
17:11So I think the most important thing to understand here
17:14is that the basic use cases of quantum computing,
17:17the fundamentals of quantum algorithms,
17:19how you make use of these systems,
17:20does not actually depend on the hardware.
17:22The hardware is all representing the same type of information
17:26and manipulating it in almost the same kind of way.
17:31So at the moment, there is not a strong argument
17:34to bet on a particular hardware platform if you don't have to.
17:39Of course, if you're a VC and you're investing,
17:41you need to decide where you're going to put your money.
17:45If, you know, if you're making a hardware play yourself,
17:48of course, you have to focus on the technology you think is most likely to succeed.
17:53But if you're an end user,
17:55there really isn't a reason to be locked into a particular technology at this stage
17:59and not be able to move in future.
18:02So, I mean, I guess the most obvious advice is don't sign a 10-year contract.
18:06Right.
18:06Anyone, and, you know, like, keep your options open beyond two years.
18:12Right.
18:13Yeah.
18:14Interesting.
18:14So there's a question from the audience, which I think is an interesting one,
18:18which is how long or difficult is it for someone to learn to use a quantum computer
18:22as opposed to a classical computer?
18:25For instance, doing something like building a machine learning model.
18:27How much harder would it be for someone to learn to do that on a quantum machine
18:31than a classical machine?
18:33George, you have experience doing this, so what's your view?
18:36Absolutely.
18:37There is a good news.
18:38It's not that complex, actually.
18:40I mean, I love saying that our first quantum solutions were developed by physicists,
18:45so people with deep understanding of quantum and of our architectures.
18:50But now I can see that classical data scientists, mathematicians,
18:54people from machine learning are now their hands on that.
18:57We have such science in our team, and they learn just, you know, in a couple of months.
19:02And a mathematician among our team developed a new algorithm, I mean,
19:07for optimization just in a couple of months.
19:09So it's not that complex to learn, because, I mean, we are providing the tools to help you
19:14and to support you in the journey.
19:17But at the end, you know, a good mind, smart people, and they know how to handle that.
19:23And just, Joe, I'm sure you have a view on this.
19:25A couple of months still sounds like a long time, probably.
19:28So, I mean, I would strongly disagree with this.
19:31There's basically two types of quantum algorithms.
19:33You can kind of divide the field into two parts.
19:36The first are kind of variational techniques that are used on near-term systems,
19:41and those are extremely accessible.
19:43So it's very easy for people to experiment with these.
19:46There's lots of frameworks.
19:48It's virtually impossible to prove anything about their performance.
19:51You can try them against particular data sets, see if you get good results.
19:55Generally, people don't even time how many resources there are used in it.
19:59But there's poor benchmarking for that.
20:02If you want to get to the kinds of algorithms where we know there's a large speedup,
20:06so like Shure's algorithm, which we know poses a major threat to cryptography,
20:10that is extremely difficult to do.
20:13So the number of people that have that level of expertise that have successfully done that
20:18is well under 200 people worldwide.
20:20So if you're looking to hire people for that kind of task, it's extremely difficult.
20:25Now, we put a lot of effort into trying to build tools to make this easier.
20:29Our ultimate goal is to make tools to automate this process for you.
20:33But it's an extremely daunting challenge.
20:36There's a lot of mathematical techniques, a lot of algorithmic techniques
20:39that are used to gain advantages on the way down.
20:41And really, you need a very deep background to make progress in this.
20:47Or at least that's been the experience we've seen so far.
20:49Right.
20:50George wants to get back in there.
20:53I mean, you have two different points of view.
20:55Yeah, yeah.
20:56And we have facts.
20:58But in our experience, I think it really depends on how technically deep you want to go, right?
21:05And as Joe said, there are a lot of companies giving you that sort of interface
21:10so that you don't have to go very technically deep.
21:12But what I find very interesting is a lot of computer science degrees and indeed, you know,
21:18even school syllabuses are looking at trying to include at least the basics of this
21:26so that it doesn't feel like a big hurdle to the next generation coming up to actually try and program.
21:33Great.
21:33We have another very good question from the audience, which is, besides quantum error correction,
21:39if there was one other problem in the industry that you could sort of wave a magic wand over and
21:43solve today,
21:45what would it be?
21:46So maybe we can just go down the line.
21:48Mandri, what would it be for you?
21:49If it's besides leaving aside quantum error correction,
21:52is there something else that you think is sort of holding back the industry at the moment?
21:55There's so much that needs to happen.
21:57I mean, first of all, we need proper, you know, sort of enough number of qubits that actually work,
22:02that are actually entangled together,
22:04that the hardware is not really sorted for us to be able to see the quantum advantage.
22:09There is error correction.
22:11We need, you know, sort of cryptographic algorithms to be able to handle that.
22:17We don't absolutely know how quantum computing is going to be delivered.
22:21It seems like it will be delivered over the cloud.
22:23It's not like everybody will have a machine just yet.
22:27But who knows?
22:28The landscaping has a lot to...
22:30Joe, if you had to pick one thing besides error correction, what would it be?
22:33So, actually, one thing I would like to see happen in the field
22:38is I'd like to see more of a move towards kind of componentization.
22:42So at the moment, pretty much all of the companies that are active trying to build quantum computers
22:48are basically trying to build monolithic processors.
22:51And the reality is various technologies have various strengths and weaknesses.
22:55What's good for long-term storage of information doesn't necessarily give you very fast operations,
23:02things like this.
23:03And I would think if we can get to a point where we can more standardize
23:08and more enable interaction between these technologies,
23:11that's what we'll need long-term.
23:13It's hard to do today.
23:15There's all sorts of problems around wavelength conversion and so on.
23:18It's really a massive challenge.
23:20But I think that would be an amazing step forward.
23:24Right.
23:24And, George, what would your one thing be if you could wave a magic wand?
23:28It will be a very done-to-earth answer.
23:31At Pascal, we are also engineers and industrials.
23:34We are building products that have to be operated 24-7.
23:37And as of now, we are controlling our qubits with lasers.
23:41And lasers are very a pain.
23:43Believe me.
23:44So my magic wand, I will have a laser, an operating laser at the industrial level,
23:49and just one, not tens of them.
23:52Very good.
23:54There's a question here that I also was interested in,
23:57which is what industry do you think is going to really see the advantage of quantum computing first?
24:03You know, what's going to be the first killer app here for quantum computing?
24:06Why don't we just get on the line?
24:07George, what do you think it's going to be?
24:10My answer would be twofold.
24:11I mean, first, there is a sector.
24:14I would say chemistry in general.
24:16And the second part of the answer is which kind of tool we will use to address this use case.
24:22And I'm a strong believer in machine learning.
24:25I think machine learning plus chemistry will be the first killing app.
24:30Interesting.
24:30To my point of view.
24:31Joe, do you agree?
24:34This is challenging.
24:35So chemistry and materials, for sure, will be the earliest industries to see an advantage from this.
24:41That's almost a given at this stage.
24:44Machine learning, we know there are quantum algorithms that give enormous advantages for machine learning.
24:50But the best of those, to get the largest advantage, you need to solve an input-output problem.
24:56So you need to be reading from a quantum memory, and those don't exist yet.
25:00There are variational techniques that are used for machine learning, but these are harder to prove an advantage over classical
25:07techniques.
25:08So beyond chemistry, beyond materials, I think the next thing that we'll see an advantage is more like computer-aided
25:15engineering.
25:15It's not going to be machine learning for a while.
25:18At least that's my view.
25:19And you think not finance either, because there's a lot of talk about, you know, risk calculations and finance.
25:24You think those won't be the first things to do?
25:26I think it depends how you count it.
25:29I'm not sure they'll be first.
25:30So, I mean, if you're looking for something that comes right after chemistry, I'm not sure it's financed directly.
25:36Again, I'm not a big believer in variational techniques.
25:39Right.
25:39So, of course, there are some for optimization that you control, portfolio optimization and so on.
25:44But if you're looking at Monte Carlo methods for pricing and things like this, which we know there are advantages
25:50for,
25:50I think those are a little bit further off.
25:52You probably, you know, a lot of the time you're going to need larger data sets and so on.
25:56I think probably it's engineering, but there could be some finance applications.
26:00Manjuri, how about you?
26:01What do you see as the kind of first killer app here for quantum computing?
26:05So this has been an interesting debate that I've had since, you know, 2016, 2017,
26:10where initially, you know, sort of people said it's going to solve everything from the traveling salesman problem to, you
26:16know, finance to chemistry to drug discovery.
26:19It's, in fact, going to, you know, actually replace AI for drug discovery.
26:22And I think everybody's now calmed down a bit.
26:25Our hypothesis is similar to yourselves.
26:28You know, I think it'll be chemistry and materials modeling in particular.
26:32We are seeing a lot of interest from the drug discovery space.
26:39I don't know as yet.
26:40We'll have to see whether, you know, the quantum advantage is actually going to make a significant difference.
26:46I know, you know, sort of AI has had its effects, but equally there are the naysayers.
26:51They are, you know, sort of saying even AI has not really made the huge dent in drug discovery that
26:57we thought it'll make.
26:58So I think it'll be some way off before anything else actually comes about.
27:03Great.
27:03And, Manjuri, there's a specific question here for you from the audience.
27:06Someone's asking that, you know, VC firms typically have these investment horizons.
27:10They're just sort of five to seven years.
27:12Their fund structures are often set up that way.
27:15And given that this is such a long-term bet and such a long-term technology, you know, how is
27:20Amandadeus able to invest in this area?
27:22How are you able to do this?
27:24Because we've done deep tech for 25 years.
27:29No, you're right.
27:30Most VC funds would struggle.
27:33That's the reason why we're specifically deep tech.
27:35We are a 10-year timeline.
27:37Having said that, we know how to manage the portfolio in a way that we invest so that the company
27:44has the most amount of time for us to give them the time to be able to realize the enterprise
27:50value.
27:50And, yeah, we've got LPs who understand that requirement as well.
27:55We have stayed in companies for a lot longer than the 10 years.
28:00And our LPs have understood that, you know, you do need to stay with some deep tech companies.
28:05It does take that long.
28:07But it's not for every venture investor, for sure.
28:10At the early stages, I mean, the industry is actually becoming a lot more mature.
28:17So, you know, the later-stage VCs can now look at this and sort of see a time horizon when
28:23they can realize their investment.
28:26Interesting.
28:27There's been a lot of discussion, I know, in the field about whether quantum-inspired algorithms are providing an advantage
28:35today even on classical hardware.
28:37And that's maybe one reason people should start to become familiar with some of these quantum techniques, is that even
28:42if you don't run them on quantum hardware, maybe they provide some advantage, again, for things like optimization, even on
28:49classical hardware.
28:50I'm curious if you have a view of that and whether that's something that companies should really be looking at.
28:55George, I mean, do you have a view on whether that's useful?
28:57I don't.
28:58I don't.
28:58I would judge the question.
28:59Yeah, you're like, no, you need the machine.
29:01You're selling the machine.
29:02Yeah, you don't want to tell people it's okay to run it on classical.
29:04I can understand that.
29:05Joe, how about you?
29:06So we focus entirely on quantum algorithms.
29:10I guess what I would say on this front is that one of the things you run into, one of
29:17the issues that makes near-term quantum algorithms, the variational techniques, so hard to quantify is that they produce different
29:26output distributions than many classical algorithms.
29:30So what's hard for those algorithms and what's easy is just fundamentally different than what's hard and easy for some
29:37of the classical algorithms.
29:39Now, that's led to a situation where there's a lot of POCs that go on at the moment.
29:46The only way to find out whether this is really useful for you or not may be to engage in
29:50a POC to try to do something.
29:52But it's also led to the situation where we've discovered ways to de-quantize quantum algorithms, where we've figured out
29:59how to make these run on classical hardware efficiently.
30:03In those cases, you still have the same phenomenon where some problems that are hard for previous algorithms are easy
30:12for these, but some that were easy become hard.
30:14And so there can be room to see an advantage from these algorithms, but it's very hard to tell where
30:20you're going to see that advantage.
30:22Right. And, Mandre, are you looking at anything in that space, like the companies that are saying, oh, well, you
30:26know, we're not going to build a machine, and we're not necessarily even going to say you have to run
30:30this on someone's quantum system in the cloud.
30:32But we think just sort of taking these techniques from quantum computing and applying them on classical machinery provides some
30:39advantage.
30:40So we have looked at a few of them.
30:42I think where it's kind of fallen a little bit for us is we've not been able to build out
30:48that market opportunity.
30:49So as a deep tech investor, we have to invest in things that we know is going to fundamentally absolutely
30:57change the way the world works.
31:00That's, you know, given the technical risk that we take, that's the reward.
31:04Now, we appreciate that not everything is going to work that we invest in.
31:07But if it does work, it's going to make such a phenomenal difference that it is a winner.
31:12And so, you know, there are a number of businesses that pitch not just in quantum, but even in other
31:18areas.
31:19Perfectly valid businesses, you know, will definitely be a success.
31:22It just won't change the way.
31:24It'll just make it, you know, sort of two, three times better.
31:27Personalization algorithms was a big fad.
31:29You know, like we had so many companies pitch to us perfectly valid businesses, but it was going to make,
31:35you know, things 2x better.
31:36It doesn't really work for our model.
31:38So at this point, I haven't found something that excites me enough in terms of a market opportunity.
31:43Right.
31:44I know often at quantum computing panels, there are questions around cryptography and the fear that quantum computers will break
31:52all, you know, existing cryptography.
31:54You know, I want to ask the panels, how real are these fears?
31:56And, you know, is this something people should be concerned about today or in the near future?
32:01George, what do you think?
32:02Well, I think we are pretty far away from that.
32:04We are decades away, probably.
32:06But back to the previous question of quantum error correction.
32:10So, but I assume there is a real fear and I understood since I created Pascal, but it takes time
32:17to change the way we are encrypting the data.
32:20So it's better to anticipate for the players.
32:22But that's my only take on this situation.
32:24I mean, basically, we are not working on cryptography at Pascal.
32:27Right.
32:28Joe, what's your view on this?
32:30Yeah, okay.
32:31So this is tricky.
32:32It's certainly some time before they pose a threat, before quantum computers pose a threat to modern cryptography.
32:39There's an excellent report that's done each year for the Global Risk Institute that tries to estimate the timeline to
32:45quantum computers posing a risk.
32:47And, you know, they interview about 50 experts in the field.
32:50I have to admit, I've been one of them in some of these, but it's often people like Peter Shore
32:54and people working on the hardware.
32:56So the guy that invented the algorithm that poses the track, the people working on error correction, people leading the
33:02hardware efforts.
33:04And they ask you to estimate roughly, well, estimate your confidence that there will be a quantum computer that can
33:12break RSA 2048, which is just a particular current standard, within 24 hours.
33:18And at the 10 to 15-year point, the expert opinion tips over from majority against to majority thinking it's
33:27as likely as not.
33:29So that's, you know, maybe 13, 15 years, something like this.
33:34If you look at the roadmaps from some of the larger companies in the space, IBM and Google, they're a
33:41little bit faster than that.
33:42But I guess you would anticipate a little bit of rose-colored glasses there.
33:49But what I would say, when I talk about the limitations of NISC algorithms, and when I say that quantum
33:56computers haven't reached an advantage yet,
34:00that's, you know, that's the majority view in the field among people that work on quantum algorithms and things like
34:06this.
34:06What I'll say now is not a majority view, it's very much a minority view.
34:12There's an effort at the moment to standardize what's called post-quantum cryptography.
34:16So to come up with a kind of classical form of cryptography that is secure against quantum attacks.
34:22And my own view is that this isn't going to succeed.
34:26So I don't think there actually is classical cryptography that's secure against quantum attack, but it's very much a minority
34:34view.
34:35So at the moment, certainly the hope is that we can transition to something like kyber crystals or something like
34:43this,
34:44that will give us cryptography that can't be attacked efficiently by quantum computers.
34:48But what you need to keep in mind is that the number of people that work on those attacks,
34:53coming up with quantum attacks on these crypto systems, is less than 10.
34:56It is not many people.
34:58And they're not working full-time on it.
35:00So what happens if we get to a world where there's a thousand people working on it full-time?
35:05What does that look like?
35:06And I'm not sure, we've already seen a bunch of the candidates fail, have attacks discovered for them.
35:13So I don't know how...
35:15So we've got 10 years, maybe 15, then we're in real trouble.
35:18So I mean, we have that long to solve the problem.
35:22Maybe it gets a good solution, maybe it doesn't.
35:24And Mandri, I want to get to you, because I know that Amadeus, you guys invest in a lot of
35:28companies in the cybersecurity space.
35:30So what's your view on this?
35:31I think there's a lot of nuance here, actually.
35:33So if we speak to a lot of the CISOs in various companies, all data is not equal, right?
35:41A financial data for a company is absolutely crucial just before results are announced.
35:47After results are announced, it's really not confidential anymore, right?
35:51So I think what a lot of CISOs are also looking at is data at rest, data on the move.
35:57of what happens if data is harvested now, but retained for when there is actually a quantum computer working.
36:08So I think it is fair to say that it is important to look at the frameworks available
36:13and put together in place something for the various types of data.
36:18But yeah, until the quantum computer is here, I guess we're not going to...
36:22Fantastic. Just 30 seconds, Joe, if you can...
36:24Yeah, so I just wanted to jump in and say that actually it's not just a problem for companies.
36:29So it's not just about securing your own data.
36:31The problem with some of these attacks is that you can undermine all sorts of infrastructure
36:36that the companies rely on.
36:38So like the DNS infrastructure that underlines the internet or Windows Update.
36:42You can sign new updates, send them off to someone,
36:45and their computers will all update with malicious code.
36:47So there's all sorts of things to be worried about, and not all of it's in your control.
36:51Exactly.
36:52So it's all about getting the right framework to protect yourself.
36:56Yeah, it needs to be a standard.
36:57So standards seem to be a theme running across the whole conversation,
37:01and we're not quite there yet.
37:03We need to obviously do some work on standardization across the whole field.
37:06I want to thank you all very much for listening, and thank our panelists.
37:09This has been a fantastic discussion.
37:11And if you stick around, the next session will begin in about five minutes.
37:15Thank you very much, everyone.
37:16Thank you. Pleasure to be here.
37:17Thank you very much, Jen.
37:20Thank you very much, Jen.
37:23Thank you very much, Jen.
37:23Thank you very much, Jen.
37:23Thank you very much, Jen.
37:24Thank you very much, Jen.
37:24Thank you very much, Jen.
37:24Thank you very much, Jen.
37:25Thank you very much, Jen.
37:25Thank you very much, Jen.
37:26Thank you very much, Jen.
37:27Thank you very much, Jen.
37:28Thank you very much, Jen.
37:30Thank you very much, Jen.
37:32Thank you very much, Jen.
Commentaires