Defending Privacy with Cryptography – Pascal Paillier, Zama.ai

This is Techstrong tv. Hey everyone, welcome back here to techstrong tv. I'm really happy to introduce this next guest person we have here as a guest today, and I'm gonna do my best on his name.

But if you know me, you know, my French accent is not great. But, uh, let me introduce you to Pascal, PA. That's right.

Am I close? Yeah. All right, you're good.

Hi, Pascal. Welcome. Welcome to Text TVs.

Hi. Thank you for having me. Thanks.

It's, uh, it's our pleasure to have you, Pascal. You know, you're somewhat of a, of a celebrity within the, the world you play in. Uh, why don't, if why don't we start here?

Why don't we start people with telling them a little bit about you? Yeah, sure. So, I'm a basically, I'm 50, uh, French guy.

I live in Paris, and I've been working on fully amorphic encryption, which is the technology, uh, that we're developing at Zema. Uh, pretty much all my life, my professional life. I started, uh, working on that when I was carrying out a PhD.

I defended it in 1999, so that's like a while back, but it was already on that, on that subject, which is like super fascinating to me. Always has been. And basically I'm a cryptographer.

Um, I've, I've been working on theoric encryption, but not only, uh, I've been in this embedded security industry working for, uh, the smart card industry, more precisely implementing, you know, all kinds of cryptographic libraries for M B D platforms. I created my own company more than 10 years ago, called Crypto Experts. And, uh, and then at some point, um, I created together with my co-founder Randy Indy, we created zma, which is our new adventure.

And it's been more than three years already, so time flies. But, uh, it's, uh, it's really something that is still, you know, fascinating to me to work on a morphic encryption because it's now becoming a reality. It's not just a cryptographic dream.

Uh, you know, the concept existed like 50 years ago. People were dreaming about building this, but it wasn't until 2009 that it really became a scientific reality, kind of. And then, uh, but it was like terribly inefficient.

And in over a decade, uh, we've made such a tremendous, you know, progress on making it efficient. And this is what we're building as now, like, uh, fully amorphic encryption for everybody so that for developers, for businesses, so that you can actually process data encrypted without actually seeing it, which I think is absolutely, uh, disruptive. And probably it'll be some kind of a small revolution, maybe a big revolution.

I, I, I don't know, never knows, but, um, especially for AI, because, you know, with everything that has been told about AI and the privacy issues and, you know, charge g PT being banned in Italy and other countries, I think they're now revisiting that issue now. But yeah, there, there are so many, like big privacy concerns and, uh, flu morphic encryption is, is exactly the kind of technology that people are after. Uh, if they knew it, it existed because it is getting more and more popular now.

People are becoming more and more aware of the existence of this technology. Uh, but when you talk to somebody who's never heard of it, usually their eyes start, you know, sparkling. And they're, what is that?

Is that even possible? Mm-hmm. Is that a thing?

Is that even possible? It seems like completely magical. And so right away they can like start thinking about applications.

How can I use this? How can I integrate this technology into, you know, the data flow into, uh, you know, uh, to put that on the table to kind of bring some, some privacy guarantees to the users. Data is becoming like the, you know, these big privacy is now, you know, under Verge, I think of being, uh, resolved using fully encryption.

It's one of these like transformative moments and like, uh, where cryptography, uh, actually can, can create a revolution in our, like, digital lives. And it's, it's super exciting to see that happening and actually building products around that. And yeah, so it's, it has really been my passion working on that.

And, and, and look, I mean, for, for those, I mean, for those out there, you could go to, uh, Zama a I zma ai. You could read a little bit about Pascal and, and his kinda lifelong passion here. And, and look, you've, you've been, you've pioneered usage in, in crypto cryptography that are actually named right for you and everything.

So it, yeah. You know, you, you, you've done these things, and I don't mean to embarrass you or anything, but you Know what, question here Yeah. Pascal, you know, our audience is a technical audience, right?

And, and actually, as I mentioned off camera to you, uh, the end of August, we're doing a, a virtual event around digital certificates, P k i, uh, quantum ai and how Yeah. Really, yeah, we are on the threshold or on the cusp of a new, a new age when we talk about encryption. Absolutely.

When we talk about, you know, cryptography, um, you mentioned you got your PhD in this back in 19 99, 24, almost a quarter over a century ago. Oh, yeah. And, and when we look back over that timeframe, right?

I remember when, you know, the first, uh, SSL socket for Netscape came out a couple years before that and everything, right? Yeah. Um, there really hasn't been, I mean, we've been using those, these encryption, the standard kind of, you know, we went from, what is it, 56 to 1 28 to 2 56 or what have you, bid encryption.

But beyond just doubling the, the, you know, the algorithm if you will, there hasn't been a fresh, you know, reset of, of this. But now because of things like quantum and because of ai, and because you know, of, of where we are, we are on the, there are new technologies that are poised to really Absolutely. Yeah, Yeah.

Change, change the game, right? Yeah. For, Yes, for me, I mean, it's been the same game for my whole career.

Mm-hmm. Now we're really changing, but I think we need to start off, how do you define what is fully amorphic encryption? How, how would you define it for our audience?

So, fully morphic encryption is this dream technology that allow, so let me get back a little bit and, and tell you what encryption is. So when you encrypt, so basically with what we call symmetric encryption, you and I are sharing the same secret key. So I, I keep mm-hmm.

I keep it secret. You keep it secret. And so I can encrypt a file, data messages, whatever, send them, encrypt it to you, you're going to decrypt them, and we can communicate privately, um, and nobody can understand what we're exchanging.

Uh, but the big problem with encryption is that, um, unless, I mean, basically what, the only thing that you can do with encrypted data to is to store it somewhere or to send it to somebody. Uh, there is no alternative. It's, it's just, it's just a blob of basic type data that you send.

And the only thing that you can do with encrypted data is basically to decrypt it. So this is like regular encryption and, uh, morphic encryption is a new type of encryption where not only can you encrypt for data, uh, in transit and data address, you know, in, in storage, uh, you can also compute over encrypted data. So it seems a little bit illogical or paradoxical in a, in a, in a way, but it's actually the beauty of it.

It's like, I can encrypt. So assume that you're a server in the cloud, right? Web two or web three, you can be like a smart contract running or whatever.

Uh, what I can do is basically to, so I have a, a, a bunch of data and I need you to process it because maybe you have some value added algorithm, like machine learning or whatever, and you're capable of taking my data and creating new information out of this data. And I would like to buy this new information that you're capable of extracting out of my data, but I don't want to reveal the data to you. So what do I do?

So I use polymorphic encryption. I basically, uh, generate a key for myself. So it's a symmetric, uh, uh, key, but I, I'm not sharing it with you.

You're the server, so I don't trust you, right? Right. So what I do is that I have a symmetric key that I keep for myself.

I take my data, I encrypt it. No, now I send it to you encrypted. So not only is it protected like in transit and addressed on your server, but also it allows you to, without ever decrypting the data to actually process it.

So you're going to process my encrypted data, uh, every step along the way. Every intermittent variable, uh, will be encrypted to you. You have no means to decrypt it.

So you, you, you cannot view, you cannot see the data, but you can still apply functions on these data. You end up with a results, which is itself encrypted, and you send that back to me. And now what I do is that I type my secret key, uh, I just decrypt it, and I get exactly the same thing as if I had sent you the, the PLANEX data in the first place, and you had applied your algorithm on it.

Only the big difference is you never saw anything. So there's, it's, it gives, it gives me some kind of a privacy guarantee, uh, privacy in the sense of data confidentiality. You don't see the data, uh, but same time you do best, and you're, you can, because just, it's the same thing to apply the algorithm of the plant data or the encrypted data.

It's guaranteed to give the same results. It's just that you don't see the data anymore. So, um, it was, you know, 50 years ago people were dreaming about building this primitive.

And, um, it stayed like that for decades. I remember when I was doing my PhD and, uh, the years that follow were trying to actually build it was, it was like super, uh, super intriguing in the sense that it couldn't prove that it couldn't be done, wouldn't build it. We couldn't, we couldn't, yeah.

It was, was really a hard problem. And mm-hmm. And in 2009, um, Gentry, uh, another cryptographer found a, the first example of a construction that actually works and is secure, but it was like super slow.

And today, the, among the challenges, the challenges that we're facing with, um, you know, developing or morph encryption products around the energy, uh, efficiency is probably our biggest problem. It's, uh, it's still slow. Like, uh, it's still like, uh, it, it depends on the computations that you want to do moor quickly.

Uh, so it depends on your algorithm. If your algorithm has certain characteristics, uh, today you can do that like quickly. Like it's, it's, uh, it, it can be fast enough for some kinds of applications, but for others like machine learning, uh, we still have like a factor 1000 x 10,000 x and so to address that, there are plenty of ways how we can address that.

But, um, I would say the, probably the most dramatic way to address that is hardware exploration building as asics that can actually process data hoor way and give you a big boost. And, um, this is what is happening right now. We are partnering with companies who are actually developing these, um, hardware designs, and we see that happening by 2025.

So it, it, wow. It's, it's happening and it's actually happening quicker than the quantum revolution. The quantum computer is still like, uh, to be desired, it's coming, but people talk, you know, 10 years, maybe 15, something like that.

We don't know exactly. Uh, for a full he morphic encryption, it's way faster. Because basically we understand the science now, we understand the engineering behind it.

We understand how developers can actually use it without having to know cryptography, this kind of tools that they need we're building right now. So, yeah, it's, I mean, it's, it's a big puzzle. We need to assemble all the pieces together.

Uh, hardware acceleration is this big piece that, uh, we will be needing for, you know, global adoption. Uh, but in the meantime, there are plenty of things that can be done even without it, like, uh, applications to the blockchain, you know, smart contracts, these kind of things. So this is where we are, and it, it's been fascinating, uh, honestly as a cryptographer to see that happening.

Like to see this technology, you know, not only emerge and, uh, as, as a scientific problem, but also emerge as a bunch of engineering problems to crack. And, uh, we're now like providing, I mean, everything we do at ZA is open source, right? So you can go and get GitHub and already play with our compiler with our, or Morphic encryption libraries.

If you're a developer, you're like passionate about, you know, privacy and you want to create these new user experiences using F H E, it's something that you can do today. Like by just, you know, going to our GitHub account and download our tools and stop playing with them. Got it.

Pascal, I wanna mention a few things about fully homomorphic encryption and why it is sort of a, you know, a dream kind of solution. Number one, quantum may not be here yet, but this type of encryption is quantum proof. Yes.

So it won't be rendered obsolete when quantum comes, right? Yes. Number two, just by the very nature of how it works, you don't have to have, you don't have to decrypt, if you will.

You don't have to have access to the data, to, to, uh, to work it from a privacy point of view, where we are worried about governments and or, you know, nation state actors or, or other entities mm-hmm. Gaining access to your data. This prevents that, that takes it off the table.

So if your data was in a Facebook or a, or an Apple device or a Google device or whatever, and the government or some entity said, Hey, we want you to decrypt this for us third party, the third party can't do it. They don't have that ability, cuz it's not based on that system that we have now. Yes.

Where, where they can, quite frankly, right? And, and so this takes that off the table. So from, you know, for everybody who, who's using, whether it's WhatsApp to, uh, some of the other, I forget off the top of my head, some of the other applications that people use now to send, you know, uh, communications in an encrypted matter, they're still subject to, to government orders, entities, you know, being forcing them to decrypt this takes that off the table.

And that, that's a really important distinction Absolutely. For a lot of people. Now let's talk about zma.

We don't have a lot of time, but what kind of products are zama coming out with? Now you are working with hardware vendors, it sounds like, to kind of build this in from an asic, you know, point of view, but Ab Absolutely. So what's, so they're like a few.

Um, I, I would say the, if, if you look at the scientific community around Phil encryption, it's, it's very small. Like it's just a bunch of a few experts, and a lot of them are actually working at Zema with, like with my co-founder Rand, we've put together like the dream team of Phil Encryption and right from the beginning, so the Zema is, uh, more than three years old now. Um, what we wanted to do was to do everything to, to make f g a reality for developers, because the world now is about software development.

And even though AI now is going to change the formula, you know, in that industry, uh, but it's still people with a lot of creativity. Um, many of them have, uh, you know, a lot of concerns for privacy. The, you know, what you mentioned about like, uh, messaging apps and, and whatnot.

Uh, data today with, especially with ai, we, we, we see now that more than ever, uh, that data is actually really the goal of the, of the third millennium. And there's a lot of value in data. So our vision at Zama has always been to make this a reality, um, fully morphic encryption has to stop being complex to use.

It has to stop being inefficient. Um, and it has to stop being like, too, too complicated. And you shouldn't like require a pH to have a PhD in cryptography to actually instrument the technology, pick up on the technology and use it and integrate it.

It should be like effort class. And so the, the, the things that we're building at ZMA are basically software tools that, um, so we have civil, um, uh, I would say civil ingredients in what we're building, but essentially there is what is called T E dot S. So it's a rust implementation of, uh, a cryptographic mechanism, which is called Taurus, F E T F H E, which is the technology that Zema is, um, actually implementing.

And if you're a cryptographer, you can already use TG dots, you can, uh, look it up on our, uh, GitHub repo and instrument that, and instrument a use case, but you have to be a cryptographer. And so you have to, you know, uh, basically build your homomorphic application on top of our library, which even for cryptographers is already complex. So this was kind of the first step, but our vision has always been to, uh, basically create a compiler, which is called Concrete.

So it's a, an entire technology stack actually. But the biggest ingredient in there is our compiler concrete. And what concrete does is that you basically, it's a compiler that texts the algorithm, um, given by the developer, uh, basically in a, in a familiar programming language.

So we have the front end with, which is using Python, uh, called Concrete Python. It's just the front end of the compiler. So basically as a developer, data scientist or whatever, what you do is in Python, you would like write down, describe, you know, implement the algorithm that you want, but it's a planex one, uh, an algorithm that works on, you know, planex data.

And now that goes into the compiler, and the compiler does all the magic, meaning it's going to compile and trans transform your algorithm into an normal morphic equivalent that you can then deploy. And so it removes all the complexity of using fully morphic encryption in the first place. You just have to be a developer.

You know, your staff, maybe you're a data scientist, you know the government that you want to do. Maybe you're a machine learning scientist. You can use our framework and basically push the button and have a homo equivalent compiled automatically with optimization of parameters and, and all the strategies, you know, to, uh, basically, um, uh, split everything into small pieces that can be like, um, uh, converted locally into ho morphic operations.

And then you can just deploy that. And so it's suppose, I mean, the vision we, we have is to make it easy, to make it super easy for developers to like adopt this technology. And so that's, that's why we're a B2 d company business to developers as opposed to the b2, b2b or b2c.

Uh, yeah, but it's the way things are now. I think it's awesome to see like, uh, the communities of open source, uh, uh, software, how they work and how people are, there's, there's a lot of drive, you know, people have dreams and we've come such a long way with software development over the last, like, uh, decades that now it's more about like the motivation, the tooling now as uh, is, is in such a state that it's, it's it's way easier to develop created applications. You just have to have a good idea.

And, uh, now you can just, with what we're doing at ze, we can just add fully morphic encryption on top of that. You can just, you know, develop homomorphic apps and, and put them out there into the world and, you know, or integrate them into preexisting applications. And it's going to be a big, a big change, I guess, for privacy in the next years.

Thanks. Absolutely. Absolutely.

Pascal, we way over time, this was the, We are 15 Minute interview. I've never done, I Told you I'm very targeted. You have to stop me, otherwise, I, no, that's what say It was delightful.

And I, I think it's important that our audience Is aware of, of these kinds of things. Yeah. I wanna thank you for coming on here.

Sure. My pleasure. Um, I mentioned that, uh, digital certificates and Quantum and so forth event, we'd love to have you speak there.

Sure. We'll get you more information. Best of luck with Zama and keep us posted.

Thank you. Thank you. All Right.

All righty. Take care. We're going to thank you.

We're gonna take a break here on Textron tv. We'll be back in a moment.