FLock.io CEO Jiahao Sun on Blockchain, Decentralized Data Security, and AI Privacy Challenges

Hello, and welcome to the latest edition of the Techstrong AI Leadership Insight series. I'm your host, Mike Vizard. Today we're with Jay Haus Sun, and he is the CEO of Flocker io.

And we're talking about the implications of AI and data privacy, because, well, a lot of data's being collected, but we don't always know by whom and for what j Welcome to show. Yeah, nice to be here. A lot of companies are collecting a lot of data because they're using that to train their AI models, but theoretically, there's supposed to be these, um, end user agreements that we're supposed to read.

But somewhere on page 422, it says that they can use that data to train their models, but we don't always know how that's gonna manifest itself. 'cause it might be months or even years later before it pops out. In a way we didn't intend it.

Do you think that the rise of AI is eventually gonna force a, a deeper conversation about what privacy means and especially when it comes to data? Yeah, yeah, of course. 'cause, um, there are not only, I mean, we, we will see model, we see how much convenience they can bring us, right?

But then not every, uh, there are, there are places, there are applications where it just cannot be, um, um, applied. Um, when it comes to, for example, sensitive, sensitive industries, uh, regulated industries, banks, hospitals, right? Naturally, not even, not even between businesses.

Even the data within the same business cannot be shared between different desks, right? So in such scenarios, like even today, some of the banks or regulated business, they can't, they can't use charge BT just simply because there are so many concerns over these private privacy issues of data of of, of all those models that can actually, you know, uh, took away your, your, your, your information and with it, uh, and expose your, your, your users, uh, privacy and everything. So, um, that, that's exactly, that, the point that you mentioned, uh, I think, uh, it's, it's gonna raise a lot of concerns in the future.

How long do you think it will take for that to come to a head? Because today, it seems to me at least that I see people using these tools in their copying and pasting data into them with little or no regard for who actually owns that data, or whether they're just the custodians for it, or it's just loaded with personally identifiable information from customers or whatever. So do we have to wait for some sort of catastrophic event, or is there some way to get ahead of it?

Oh, I think, uh, those type of things, well, no, of course. No, we, we shouldn't wait for a CATA catastrophic event, or, I think in privacy everything's so important that even, even if it's, um, a small event that's already catastrophic. Um, um, and, and I think we already see lots of, uh, technologies, lots of companies trying to build this, you know, build models as open source build models that can be, you know, tuned on your local devices, private devices.

I think it's, um, yeah, it's kind of things that we, we've been super, um, um, um, uh, positive about and trying to, you know, tackle this specific scenario, you know, versus those centralized companies, you know, building those closed source ai. Mm-hmm. Will this get addressed as a set of regulations, therefore, and we're gonna have to just wait for various, uh, legal bodies around the world that kind of come to terms of what data privacy means?

Or is this gonna be maybe addressed more holistically from the top down at some point? I, I, I would say, yeah. Yeah, I would say, um, especially because the definition of data privacy, right?

How, how, how that can be defined, whether your data itself of course is private, of course, but then whether the derivable your data are also private or whether the derivative that's not even readable by any anything else are also your PRI privacy, right? For example, your data, uh, the model, the model had a, had lots of weights, and when those weights are still private, because it's a derive of your data, and also it's actually a collective di derive of lots of people's data, whether that's kind of like the collective IP then, uh, when it's being trained. So there's lots of grounds need to be discussed, but so far, under GDPR and many of the local, uh, data privacy, uh, legislations, um, the, the, the very obvious first priority is to make sure the raw data never got exposed.

So that's at least the first steps every company trying to secure. I feel like this issue's been kicking around for a long time, even before AI became up. So, is AI just really, um, forcing a discussion around this topic that's, we've kind of been postponing for many years?

Yeah, I would say so because AI is now, you know, uh, um, absorbing all the data across the world, right? Then people starting to realize, ah, it, it actually understands everything around me. It actually understands, uh, all the context I'm talking about, how, how that happened.

Like, I just bought a bottle of wine earlier this morning, then I actually received some marketing materials this afternoon. How come? Right?

It's, uh, lots of things happening around us. Well, thanks to ai of course. And also, you know, due to ai, right?

We, we we're starting to realize there are certain privacy concerns that's already, you know, being noticed by us. Mm-hmm. Do you think the general population will start to push harder on this?

Because, well, I think everybody's having that same experience lately. No matter what you do, if you're talking about a topic at your family dinner, the next thing you know, there's an email about it, right? Yeah.

Yeah, yeah. I think people will do, and many people will, will raise awareness of it. And that's actually back to the point I, I wanna, I bring up, uh, earlier.

Um, so there are scenarios where convenience really being bought by ai, right? But then the concerns only by certain people who are concerning their privacy got leaked. There are other scenarios where, um, it's only privacy.

AI can be deployed, as I mentioned, uh, regulated businesses, or for example, privacy, uh, uh, personal assistance, privacy companions, where you, you, you actually share data with some, with an ai share everything about yourself, because if not, then you will not have a good assistant for yourself, right? Um, then that's a dilemma. Whether you wanna share everything with them or with the ai or you, you wanna just share some surface information about yourself, then your AI wouldn't be good enough.

So there are always concerns and, and, and balances between the two. And eventually, I think eventually the killer application or the or killer solution for this is a total private AI solution where you don't need to worry about it where you are safe and confidently happy to share all your data with that AI that's actually helping your day-to-day life. So how do we go about building that?

'cause so much of what we're using today are public tools that go off to some sort of cloud service. What would be required to get to that private kinda architecture you're talking about? Mm-hmm.

I'm thinking, uh, I'll say there are many decentralized AI companies starting to build solutions or different solutions around different, uh, um, tech stacks, right? There are people trying to do encryption, meaning that they can encrypt your data while the receiver receive your data. Still encrypted data, meaning that your role data has never been, um, exposed, but same, same as, um, uh, the thing I mentioned, right?

Whether your data directly is still, still, still your data. It's a, it's a thing to, to, to discuss in the future, maybe by the legislation, but at least, at least for now, your know, data is secured. And there are other ways, for example, federative learning, that's part of what, uh, flock is doing better learning and blockchain.

That's why we call us ourself flocked. So, um, it's also one of the solutions where, well, we keep every data local, we deploy local models to your local devices, makes the training, makes the changes, and then update model makes the general model better. Um, that's, that's, um, that's also one of the solutions, um, um, that's been deployed so many years between different companies and centralized solutions.

For example, Google, apple nowadays, if they, if you install their, um, uh, typing, uh, well typing software, uh, input software, right? They're actually predicting your next words by using Federation Learning J just to make sure local data's local, um, um, like, because otherwise they're gonna have everything to type down, including your passwords, your secrets, everything. So there are, yeah, there, there are so many of such, uh, tech stacks and people being, uh, um, exploring, and they all leads to the solution where we keep the raw data as close as possible to the user.

We don't actually submit them over to any, um, other, uh, other servers or other, other clouds. Mm-hmm. That's generally how the solution works.

So it's more or less a decentralized approach. Um, do you think that, uh, oddly enough that the combination of privacy and AI might pull more, um, blockchain applications into the enterprise be rather than just, you know, everybody thinks about blockchain more or less is Bitcoin, but, uh, are we gonna see applications that go well beyond just cryptocurrency? Oh yeah, of course, of course.

Like, like, like for us, the, the, the, the own chain structure is one of the governance, uh, protocol that's being used to govern the whole feder federated learning training mechanism. 'cause if you are only leaving this training mechanism to a centralized company, right? They can be evil.

They can just send back the raw data because just easier and more efficient for them to train the model, right? You want a public governance over the training model. So that's how flock works as a mechanism that we proposed.

So I would say, yeah, there are lots of such innovations, not only just treating crypto as a cryptocurrency, but also treating crypto as, you know, one of the governance methodologies to help, um, to facilitate the, the, the transparency of the whole AI model training process. Do we have the expertise to execute on that? 'cause I think one of the issues you hear a lot from enterprises is they barely understand how to make AI work and their understanding of blockchain is probably even less.

So, um, what would it take to kinda realize that from an expertise perspective? I would say you, you, you can, uh, think of this as, for example, uh, Bitcoin mining process and, and then other vendors who create machines to mine Bitcoin, right? So there are, so, so there will, there will definitely be power users like AI engineers in the world who can actually help facilitate the whole training process, who can help, uh, um, evaluate the training models, the, the, the outcomes of the models, right?

But there are also the general public who actually, uh, who wanna invest or dedicate into the training process to make sure that okay, they can put their weights or more legit governors into the network. So the networks is safer with more stakes on legit, on legit governance. So similarly as a, like a POS process nowadays for, for Israel, meaning that yeah, you have power users, you have general users that they all can participate to secure network, What is your sense therefore of, um, who's gonna take the lead on this within an organization?

Is it gonna be driven by somebody who's a CIO or is there, um, the security folks or who's kind of gonna stand up and say, Hey, we need to rethink our approach to privacy and ai? Uh, I don't have the question. Uh, you mean, you mean in our clients, right?

Who gonna be the person to raise awareness of this? Ah, I guess that's CIO position. Yeah, that's a CIO position for it.

Um, and, uh, but for us, it's not just the, uh, just pitching to the traditional, for us, it's not just pitching to the traditional businesses who have their CIO to raise awareness for us, it's more of those companies who already had the pinpoint of this. So we are providing a solution, right? It's, it's always hard to just just pitch to a company saying, oh, I need to figure, find your CIO and let you know that you have a concern of your privacy.

It's always the other way around. While in hospitals, banks, they come to us and ask, okay, whether we can provide a private AI solution because they can't just use the centralized solutions now. Mm-hmm.

Do you think at some point companies themselves may come around and say, we need to drive some sort of decentralized approach, because if I look around the world, there's gonna be different data privacy regulations everywhere, and at some point, maybe everybody will just get tired of trying to figure all that out and just look for a more technical solution to the problem. Yeah. Um, tech, um, I think, I think, um, they're right.

Uh, I think during the whole process, especially for all this, um, um, developments, right? Um, if it's actually generating more, um, um, okay, so it's actually, if it's actually creating more business opportunities and generating more or benefits for their own business, right? So that would be a, like, like a smooth proposal for the internal business to actually, oh yeah, we should, we should focus or, or at least lean more on the private AI solutions versus we buy the wholesale solution maybe from a centralized provider that might cost us a huge fortune.

And then still all our business insights are on health of another company. So, Hey folks, we've been having this argument about centralized versus decentralized for a few years now, maybe even more than longer than anybody cares to admit, but it looks like maybe it's all gonna come to a head in the age of AI and data privacy, because we're gonna have to make some fundamental decisions about just who do we trust with that data and what are they doing with it? Hey buddy, thanks for being on the show.

Thank you. Thanks so much. Cheers.

And thank you all for watching the latest edition of the Textron AI Leadership Insights series. You can find this episode and others on our website. We invite you to check them all out.

Until then, we'll see you next day.