Webinar – Data, Explored #5 – The Blockchain Kill Switch

Welcome everybody. It's great to be here. I'm really thankful to all for giving me the opportunity to be one of the host of this Data Explored, sponsored by Actian.

It's a great way to discover fantastic people and have a very humane chat. So, my name is Jean-George Perrin. I often go by JGP, and I'm here working on data and AI strategy at Actian which is a fantastic job.

And today I've invited Karen and Karen invited Beni. And, you know, that's kind of a LinkedIn kind of thing. and we are going to first, in the first part, we are going to talk exclusively with, Karen about her book, Blockchain Tethered AI.

And then in the second part, we will have a discussion as well with Beni in there. So, but Beni, if you want to jump in during the first part of the book, you know, only if Karen says something wrong. Okay.

Um, if I say something wrong, you keep it for, you know, a couple of hours or something. Thanks, I appreciate that. Um, or you can say it in French.

Okay. So Karen, let's get started. Okay.

Tell me who you are. I am a software engineer. I've been programming since I was 12.

And I spent a long time as an e-commerce engineer, and now I specialize in AI and I have since 2016. And my company, Kilroy Blockchain, won the IBM Watson Build Challenge for North America in 2017. And so I have been at this a while.

And you're an IBM champion? I'm an IBM champion, yes. And that's how Jean-George and I know one another.

And I am also co-chair of the AI/ML Task Force of C2PA, and you can look that up, C2PA.org. It's the Coalition for Content Providence and Authenticity, which is how I know Beni. All right.

So that, we will dive deep into that, a little later because this is really fantastic. I think. So, we are going to talk a bit about blockchain, right?

But it's not Bitcoin. It's not Bitcoin, no, it's not cryptocurrency. Okay, so everybody that came to the talk hoping for Bitcoin tips.

It's okay to leave. Bye. So Karen, tell me how did you get there?

Okay, so something I love, I love people's story about how did you find out there was potentially a problem or an issue or something to do about this problem? Okay, well, let me tell the story then. I was coincidentally in Austin, living where you are right now broadcasting from in Austin, Texas, and South by Southwest was a big thing in Austin, Texas.

Every year they take over the whole town. And IBM had wrapped a building in cognitive kitchen, cognitive cooking. This is 2016.

And I was, I went to the cognitive cooking building and looked around. I was already interested in blockchain and enterprise blockchain. And I looked around the cognitive cooking display and they had a robot that you could move with your mind by putting this net on your head.

And it was sitting on a table, and you could move it forwards and stop it and move it backwards and stop it. And I looked at the guy and I said, how and the world did you do that? And he says, oh, just a couple lines of JavaScript.

And I said, as a web developer, I know that that's what I wanna do next. And so I started getting interested in the AI. At the same time I was going to meetups in Austin about blockchain, enterprise blockchain, which was coincidentally also, IBM was also heavily involved in that, and Hyperledger Fabric and IBM blockchain at the time.

And so at that same time in Austin, this is definitely an Austin Texas story, this couldn't have happened anywhere else, I was also coaching a Dragon boat team, which who knows what a dragon boat is? No, I don't know what a dragon boat is. Okay, it's a 41-foot long boat with a dragon head and a dragon tail.

And then you've got people that paddle the boat that are a team. 20 people in the boat. And you've got a drummer that sits on the front and keeps time.

And then you've got person that stands on the back and steers with a 12-foot long steering ore. And that's my job. I'm a Dragon Boat coach.

And in Austin, I had the opportunity of someone that I knew there through Dragon Boat, also worked at the school for the blind and visually impaired. And he was going to do an outreach program for the kids that were between 18 and 22 years old as their last chance to come back and learn how to live in the world. They're learning life skills and learning how to get out in the community.

And so one of the activities they did was Dragon Boat. And so my friend Scott said, would you like to get involved in this and help me coach? I'm like, great.

And so I spent months with these kids getting ready for the race, and I noticed that every time that kids got bored, every time we took a break, the phones came out and the headsets went on, just like any other teenager. And so at the time, because I was looking at the IBM technologies, I was thinking, you know, you could make something with the phone so they could see what's around them and see what the people, where their person is that left them there, so they're not just stuck and wondering. And so then right then IBM announced the IBM Watson Build Challenge, which I thought was just a crazy coincidence.

And I said, okay, now we've got an excuse to do this. And so we dove in and we kept winning and winning and winning. And one of the questions along the way was what happens when someone who's blind and visually impaired depends on AI, if that AI is just completely based on data, and you've got no way to prove that the data has not been tampered with.

And so I asked this question to top experts at IBM in AI and machine learning, and I actually asked it at one of those big conferences where you stand in line to ask on the microphone, and they huddled together, and they came back in a second and they said, if you think that's bad, think about the algorithms. And I'm like, holy moly, they don't have anything to make sure that this stuff doesn't get tampered with. So you could trust this AI, like if a farmer was gonna plant crops based on what an AI told them, and that data could be all messed up and wrong.

And so that's already have been exposed to being exposed to blockchain in Austin at these meetups. And I would, I wanna mention Mark Anthony Morris, who was a leader there in Hyperledger Fabric and IBM blockchain, he really did a lot to bring it to the community. But I started thinking, well, you know, you could use AI to track and trace this.

And so IBM did some research.

Google did some research. And then in 2021, I started getting the feeling, I was like, I told my business partners, we need to really nail this down and write a book. And at the time, I was doing book reviews for O'Reilly.

Yep. And so I did a proposal to O'Reilly. It was just a gut feeling.

I'm like, this is the time. And so, we wrote for the entire year of 2022. And then our book was released in February of 2023, Valentine's Day, right along the giant surge of LLM onslaught.

And so that is the story of how Blockchain Tethered AI. So I was a reviewer in the book. I'm even quoted at the back.

And, yes, I like to play it humble, And I, so in order to prepare for today, I reread it. Um, and to be fair, Karen, the thing is it's still very relevant today. So it's because even if it came out just around the same time, as you said, as the LLM, think you mentioned a little bit of ChatGPT somewhere in it, but I think it's still very, very relevant.

Okay. So, but what will you change now, in the book, or are you planning a second edition or something, that is. Or how did the LLM and, the massive interest in GenAI is, can affect your book?

Well, after, right after the book came out, I started to do a few presentations at local areas. And it, you know, I can't say that anything had a lot of people at them, but the people that came were really, really good and really involved. And one of the discussions we had at the library was about Adobe Firefly and how Adobe Firefly had to release something where you could tell that the picture was made from AI.

And so as a reminder for everybody, it's kind of a Dali image generator made by Adobe, right? Yeah. Made by Adobe.

Yeah. And they had a way where you could look inside the file and see that it was made by Adobe Firefly. And so that discussion came up at the library and Adobe Firefly was brand new at the time.

And so we said, well, how did they do that? And we went down the road of, of saying, well, this was done with content credentials and we found content credentials, and we found the C2PA.org, and we thought, wow, you know, this is an interesting idea. Maybe this is how you could implement most of these things that we talk about in the book.

And the identity pieces there and the provenance pieces there, and the cryptography is there. And so all of those types of things were very interesting to us. So we went down that road of C2PA.

Okay. Alright.

And so if I could change the book or write a new book, because okay, so CTPA has introduced me to a whole new world of standards bodies that I was not familiar with before. And they can be parts of the Lennox Foundation like, C2PA or DIF, which handles identity, or it can be an industry vertical like SMPTE, Society for Motion Picture and Television Engineers. And these folks have been working for years on standards.

And in the, in the case, Beni is nodding because that's the industry he comes from, and these folks are really far ahead. And, you know, I was really astounded at how far ahead a lot of their technologies are. Their knowledge is so deep, and they're so humble.

Just like you, Jean-Georges. But they are, no, they really are. They don't even realize how far ahead they are.

Okay, Karen, I feel like you really want to go to C2PA. I don't want to go there yet. Okay.

So I'm still. You're still on the book. Well, you asked me what I would do differently.

No, no. Okay. Okay.

Includes standards No, no. It's great. So just as a reminder we've got a Q&A.

Do not hesitate to write in a Q&A to ask questions to Karen. You can already ask question about C2PA, I just won't tell them to Karen right now. So before we go there, because I know you really want to go there, so I'd like to understand the thing.

One thing is that... okay so the book comes with his own repository where you've got code samples and that's pretty great. It's a lot of work.

It's a lot of work to do a book. I know a thing or two about that. And it's also a lot of work to maintain a repo with the code that is actually relevant to a book.

Um, so tell us a little bit about what people can find in the repo for the Geeks and among. Okay. The repo has some docker containers that you can fire up, that give you an example of an AI training event and a blockchain setup.

And where you can you can pop up an application that's also in the book. You know, there's lots of screenshots of it in the book. And where you can emulate a workflow of a team that has, that's responsible for training an AI and then, you know, approving it.

Proving that it's been trained properly. And then track of all that on blockchain. Yeah.

And one of the developers I believe is probably here in the chat. Jenish is Jenish was the lead developer, and if anybody wants to reach out to him and talk to him, they're welcome to do that, through me and I'll connect you. Awesome.

Awesome. And do you know if anyone has taken this code and put it to production or things like that, or, No, and, you know, Jean-George, it wasn't intended to be put into production because it's just a really simple example of a basic AI system that I think we used, I can't even remember what we used now. It's been so long.

It was either a traffic we talked about traffic, we talked about images, but I'd have to, I'd have to open my book and look. It would be very rude. But we did a basic event and then, but the main thing wasn't the event so much as tracking the workflow and what we did during the event.

And when the, like the, the custody of that data when it was introduced, who had it, you know, where did it come from before that? When was it introduced to a machine learning event? Uh, what type of, what type of machine was it trained on?

What are some examples of tests that were done before and after the training. Those types of things. And then they're all tracked on blockchain to where you can go back and reverse what happened, have some kind of a trail of breadcrumbs of how, how an answer might have been arrived at, right?

You might say, well, this worked fine if we go back before we introduced this last training data set. And by the way, where did that data set come from? And so, those types of things.

And the idea was to use blockchain because of the fact that we predicted that so many of these events wouldn't be done by people anymore. They could be done automated. And now we're into the age of agentic and we know for sure that it can be agents doing this type of event.

And so, really the code that we included with the book, the Blockchain Tethered AI code, was meant for developer, you know, an engineer to set up on their local machine, and say, okay, this is how this could work, and how does this apply to my system and my ecosystem, my environment, and how do I pick out the blockchain touch points that are important to me to be able to track and trace then to go back and determine the origin and to even be able to reverse the events. Yeah, yeah, yeah. And make sure that there's no tempering in the whole process.

Yeah. So, alright, I want to transition slowly to C2PA, but tell me just in a couple of words or how much time you want, actually, what is NYX NoCode? Oh, NYX NoCode.

Okay. Well, NYX NoCode is an invention that I made while I was working on an AI chat bot. And I had it, it was one that was gonna go on a website, a little AI chat bot.

And so I was working on a blank page, and I was looking at that blank page going, I bet that I could make something come up on that blank page. You know, I've been doing web development for so many years, I thought I bet I can give it a command and make it make something right there on that blank page that I'm looking at. And, and I did it right.

And so I realized then that I could create any kind of page as any kind of web development that I knew how to do, and even a lot of things I didn't know how to do just from a prompt. And so I created a product called NYX NoCode. And it started out I was gonna make it just for anybody.

And I found out that it was really hard to get people to try things. And it really truly is. And I can't, you know, that would probably be another hour episode trying to figure that out.

So what I decided to do, I kept asking, well, what do I do? What do I do? And, what I ended up arriving at was to do a hackathon for K12, for actually it's middle school through high school, because I already have customers in that area with other, with blockchain things that I've done.

And so the, so what we did was I worked with the superintendent of, of Mena school district, Dr. Lee Smith. Mena, being in Arkansas, not in, and not in U of AE, it's the one that's here.

And it's a city. Anyway though, the what we did was we came up with a three week hackathon, no code hackathon, where the students I built this, this platform with all these fun things and, you know, made it look fun for students. Interesting.

Lots of things to try. You could even create a picture from a doodle, you can, yeah. It goes on and on.

Now you can do it from music, and MIDI, and guitar, and it'll make the AI pictures and, and all that. But what they did was they had a hackathon with a purpose. The first, the first hackathon last year was 3, 3, 4 weeks long.

And they they came up with ideas for the bike trails that they have in Mena, which are mountain bike trails, which is a really big deal for them. They, there's a tiny little town, beautiful clean air, clean water, you can breathe there and surrounded by mountains, but that's all that's there, right. You know, and they have the mountain bikes, mountain bike trail, and they got state funding and federal funding to build the mountain bike trails and actually build a lift up to the top of the mountain.

So it's a big deal for this town. So last year, the kids competed and I got sponsors to give them eight mountain bikes. So there were big prizes, and they and they made webpages promoting ideas of what to do on the mountain bike trails.

And then, but every, all the AI that I built into this, any kind of image that they produced, and this year they did video, but it has content credentials where you can tell where it came from. And, uh, and so that's kind of a lead into content credentials. But we're teaching, what NYX NoCode does, is it teaches the, um, teachers how to show the students, guide the students, in AI fluency and help them to look at it with a critical eye.

And so they're not just accepting everything. They know how to look at something and examine it and say, Hmm, you know, this might be Ai, but you know, it's kind of useful for this reason. We're not gonna worry about it, you know, but at least they know and they're aware.

And so they had a hackathon again this year for three weeks and it went great. They came up with, and this is 200 kids all at once on my system. So it's a good stress test.

And this year they promoted the events at the school that they wanted to see more people participate in. Now this is great. I love how you are, um, you are in, are you were involved with the next generation.

I know we can continue on this because I know some of your project as well. And just in a word, okay? You are also, you are a learner.

You are, you are an all ever learning kind of philosophy as well, right Karen? So yeah, tell me, what you're doing right now. Well, right now I'm going to, I'm going to college.

I'm going to the University of Arkansas, and I am specializing in two areas. I am taking independent study and piano, which I've done for two and a half years. And I'm proud to say I can play Jean Jeanie and Killer Queen all the way through.

After our episode. All the way through, yeah. I'm getting good.

I'm getting, so I can do it. But we, you know, we're progressing. I'm getting, I'm getting, so I can play, and I'm really excited about that.

But at the same time, I've been talking to that instructor about AI. And, Dr. Miroslava Panatova, and she's amazing.

And so, but then this year, I was really excited. Dr. Zach Steelman reached out to me and he was announced as an AI fellow at the university.

And he teaches a business systems infrastructure class at the Walton College. And I was able to get into that class. And, you know, at first he said, well, you could come to our class and speak about C2PA.

And I'm like, well, why don't I just come to your class? And so I'm taking the class and I love it. And, um, but that's what I've been doing.

And now they just started the AI club, okay. So I went to the inaugural meeting of the AI Club at the University of Arkansas. There's about 150 people there.

Nice. And then we're also having our own three week hackathon, AI hackathon. So I'm going to get to collab with my teacher, with my piano teacher on a music invention and compete.

Yeah. And you, and you told me a little bit about it. And I will leave the cliffhanger for people who follow you on LinkedIn.

So that was great. Okay. Beni, you've been super, super patient there, I think.

And I can feel that Karen is, wants to talk so much about C2PA. So, Beni, come on in, unmute yourself and tell me in your best French, who you are. You want it in French, Jean-Georges?

No, I'm just kidding. So, I'm Beni Beeri, and I'm very happy to be here and to discuss with you C2PA in a few seconds. Uh, by education I'm a philosopher of sciences and I am a builder by necessity for the last 25 years.

A self taught coder. I was active in different verticals starting from gaming, advertising, technology, blockchain, although the concept of blockchain is a problem today for a lot of people. But I want to reassure you, DLTs, and not only, you know, crypto, we're gonna discuss that in a few seconds.

And for the last almost decade, very active in AI. I was, I think employee number seven of Metaphysic, that did bring to the world deep Tom Cruise. I don't know if you remember this viral reel on a TikTok and on Instagram where Tom Cruise was deep faked in a perfect way.

So this is what my company did as a pardy. And to showcase the dangers of this technology when you can impersonate people. And then after this, we brought a AI and most a likely genetic video AI to Hollywood.

Most of our work can be seen in a lot of movies. We went to American Got Talent, we even made to the final by impersonating Elvis Presley almost perfectly on stage. And then we were bought out a year ago.

And now what we do with this new company that we have formed together with a PFT and an Indian company, we are bringing Hollywood quality to the enterprise world. And meaning what we do, we are building digital doubles at the level of Hollywood, but for audio and video content at the level of enterprise. So this is the wrong way.

And in this constructive, a little bit more than 1000 workers, I am in charge of trust and data governance. And I think that this is why we're talking today. We can talk about data governance a little bit.

Yes. There's a few things. I know a couple, but that's super exciting.

And Beni, you are so you're working with a lot of people around the world, but you're based in Tel Aviv, right? Exactly. Perfect, perfect, perfect.

I like we're very global scene today. So tell me, I don't know if we want to take that. Okay, but what is C2PA?

We mentioned it, but what's the origin of it and what is it? I will jump in. Karen, please forgive me.

Yeah, go ahead, please. I don't want to use big words, and I don't want to give you a kind of definition of what you can read on your, on the website, et cetera. I think that the best way to define a C2PA, it's a digital birth certificate for any data point that is evolving at the past of you, you modifying the data point.

So if you want to, when you buy food, you have the nutrition label, you know exactly what was what from what was made. The nutrition. C2PA is offering the same kind of service for any digital, if you want to say media, by providing the following information, who did that with what kind of technology, why it did that, and easy authorized to proceed with this kind of stuff.

In another kind of language, if you want to, it's a technical standard, coming by the way, from the Linux Foundation. And I know Jean-George, that you are a little bit aware about this kind of stuff. I've heard of it.

Yeah. Together led with Adobe and represented by our dear friend Leo Waal, which is the father of the PDF. So it's not anybody, it's the father of the PDF.

And they've created four or five years ago this open source standard that is embedding temper evident metadata into digital content. So this is for the jargon, if you want to think of it as a manifest that travels with every image, video, audio file, or text together with you. That's what is C2PA.

So a bit, a bit like the exif data you found with pictures on your digital, or you take with your digital camera. Right? Definitely.

And it's, it's funny you shall mention that Jean-George, because I can tell you that Nikon camera, is the new generation of Nikon cameras, they are C2PA compliant. Yeah. Meaning when you use Nikon, you are gonna have this metadata embedded in every single picture.

Another, you know, tool, device that has embraced it. It's Samsung, the S25. When you take a picture or video, it's already C2PA, you know, embedded.

And the greatness, because I'm certain that a lot of people here and a lot of people around this initiative, and by the way, my side, I'm a volunteer there. I'm co-chairing also working group like Karen about audio watermarking, believe me or not. This is really complex.

And we met there, but we are volunteers. There is nobody that is paid to create this kind of stuff. But we have a major win.

We have a major win. So the Linux Foundation doesn't pay their volunteers. No, No.

And it's okay. We work for humanity. You know, when I, when my mom is asking, what do you do?

I'm a mission for humanity. This is what I'm doing. Next time I see my mom, I tell her that too.

Uh, humanity, I write that down. Does humanity bring food to your table? No.

No, but it's, it's spiritual. Exactly. I told you, you don't education Screaming.

But, but what I want to say about C2PA, I mean, if you walk in an isolating mode, I mean, you bring this manifest, it's amazing. You have cryptography, you have timestamping, everything, but nobody's using that. You are not going to be successful.

Yeah. The greatness initiative beside the technology, that is amazing by the way, I never met such quality people working on this kind of, you know, problems. Oh, yeah.

It's amazing. But the fact that we managed to bring Meta, TikTok, Microsoft, OpenAI, Nikon, and almost every giant big tech in the world adopting the standard, it's our major need, a major win for the initiative and for humanity, because it means that we are going to be able really to say what is true and what is not true using C2PA. Yeah.

Yeah. So Karen, we talked about a practical view of it, right? You can see the result of C2PA, for example, on LinkedIn.

Yes. If you, um, you may have seen the CR logo, which represents content credentials. And one of the places that you commonly see this is on LinkedIn.

If somebody puts in uploads an AI generated image or video, it'll, it'll automatically put that in, which means you could peek inside that file and see more in the metadata. You can also see them if you, if you generate an image, if you're in Google, in Google Docs now, and you go over to the AI on the right and you say, make an image of something, that image will have content credentials in it, and you can play with them and inspect it and dig in and see what that's all about. Can a C2PA certificate or whatever can be, can it be altered?

Because if I keep my comparison with, exif in photos, for example, it's super easy to edit and change whatever I want into my, in my exif. Okay. Uh, is it the same thing with C2PA?

It would be possible, and that's why you might wanna add other, um, other things to it, to fortify it. So if that's, if that's one of the concerns that that's gonna get altered, you'd want to think of that ahead of time and do some things like watermarking or soft binding where, um, if, if it is altered, it could be reunited with the original content credentials. And then also too, you can add another layer of protection like blockchain to it, where you can go back and prove that, prove what was originally there.

I would slightly answer differently, Karen, from you on this. Uh, I think that C2PA, and by definition, every single piece of technology is hackable. I mean, yeah, No, I mean, Yeah, it's all hackable.

Everything is hackable. And with the quantum gap that is waiting for us all, what we are talking about in terms of cryptography will be completely kuk in the next couple of years. But what I want to say is that this is the most advanced piece of cryptography when it comes to content identification and provenance.

It means that instead of having just, you know, a vision of metadata and logging, or if you want to, you know, keeping a record of metadata, there is a layer of protection that we call cryptography. And it means that every single manifest is doubly signed once by, obviously the creator, and second of all, by a certification provider that will say that what you have put in the manifest does reflect the reality. So yes, it's possible to hack.

I'm certain that, and by the way, we are trying, obviously to red team what we are building. And I can tell you so far that there, it's very complicated, but I count of my friends the non-ethical, I would say, hackers to attack it and to make sure that we are going to be even more bulletproof thanks to their attacks. But at the moment, we are in a pretty good shape in order, you know, to confront a possible cybersecurity issues.

Uh, are, so, what's, what's the target audience for C2PA content when we're, okay. I guess you're going to say content producer, but what kind of content producer and consumer? I would focus on my church here, will really work for my church, entertainment.

Think about how every single lady to the entertainment world, my company's coming from Hollywood. So you can just imagine what is the kind of paranoia that we are living in when it comes to data leakage or to the fact that our, our work would be, you know, used by malicious users and gain money from it without in our permission and consent, et cetera, et cetera. So the maker would say, you know, use case, from my humble opinion, is the entertainment world where people are creating IP using obviously, you know, a lot of tools and they want to protect the IP and mainly the visual ones.

But you can definitely think about everything you want in terms of health, for example, when you want to see that there is a list, I would say a chain of command that was definitely, you know, approved by somebody. So this, this is definitely where C2PA could play a role. But the best use case is what we try to do, you know, at Brahma AI at the moment, it's to make sure that when once a user is adding a piece of content or creating a piece of content, we start the origin if you want, you know, chain to make sure that at the end of the chain, he will still be the owner and will be able to defend his ownership or her ownership on this.

It's like forensic evidence, you know, the more evidence that you have, the more likely you are to be able to trace something back to its origin and prove, prove you, yeah. You're building a forensic trail for people to be able to track that. Yes.

Works in both sides. Meaning it'll help people to claim their rights if today you need to show that you have created using, you know, I would say manual work, with C2PA you just opened the manifest and you show it to the judge and you will see, or she will see that you were the first, et cetera. But it's also helping us, thanks to the fact that a lot of platforms, editors, are using now C2PA to identify the malicious players and to bring them to court because, you know, responsibility, our accountability on both sides.

When somebody will do something wrong, not bad, and more, you know, an ethical guy than a moralist one, he will be able, you know, to be brought to court knowing that he has done something wrong. Interesting. Do we um, how do you see that applied to AI, if it does?

Thank you for asking that question. Yeah that's the focus of the AI/ML task force, is we are looking at ways that those, um, for instance, if you're, if your data set that you're using for training all has content credentials, and that's already been done at the origin, let's say, by taking them with a camera that puts them in at the time that the picture's taken, that's just one way you can get them in. But, you're training an AI with this training set, then how can you carry that through to then be able to show an output that those credentials were used to train that model and, had some influence on that output.

And then, or that the original content was used, a way to use the content credentials and carry it through. And then also too, we're looking at credentials for the model itself. Uh, you know, like the model cards and things like that.

And, uh, and so, there's been some yet to be released recommendations, written by some very big companies in the task force. And once those are released, I'll be able to talk more about it. But I will say that I'm gonna be testing it at the university.

That's gonna be fun. It's all, it's all looping, right? Yeah.

I've got, they've got a GPU cluster, okay. Uh, and, uh, and so they're gonna let me run a test on, I found, I've got a local artist, Zach Morris, that I'm working with that will give me access to his collection. We're gonna credential everything, and then we're gonna try to train a model, an image generator that will, generate something in his style and see how much of the credentials we can carry through That's, that's very interesting.

So, you can actually, so the whole idea, if I understand correctly, is you can work with an artist or a production company. You give it to a model. Some,to the LLM or the GenAI process can actually take inspiration from it, but trace the inspiration in the derived work.

Exactly. Bingo. And so that is, in a sense, the same thing that I was trying to get across in Blockchain Tethered AI without the trained blocks, I mean without the chained blocks, you know, that's the, that's really the difference.

And that's why you could take something like this and add blockchain to it, you know, if you wanted to. It's just one more layer of tamper evidence. You've already got tamper evident media because the cryptography is used on that as part of C2PA.

So you've already got a layer of tamper evidence there that can be checked, but then you can fortify it even more, with blockchain. So, if, and, people, listeners, don't hesitate to use the, the Q&A there. So, if I draw, if I draw a parallel to what I'm doing with the Linux Foundation with data contracts and data products, um we we're technology agnostic, correct?

You, you guys are as well, right? It's, we don't care. You mentioned Nikon, which is a best camera brand ever anyway.

Uh, and um, but it's not, and Samsung can take pictures as well and et cetera. So the thing is, you don't care from the, from the hardware perspective, or you don't care from the software perspective what's coming in. Correct?

I think that you're, it's spot on because, and I'm gonna draw a a parallel also coming from my world, from the philosophical world. The Swiss philosopher Jean-Jacques Rousseau who wrote the social contract sometime in the 18th century. And this is exactly this.

I mean, we don't care about the individuality in the collective. What we want, and this is what C2PA is doing, it's creating an, a collective standup that we are go all going, you know, to follow. Not because we have a direct, and so I would say individual benefit, but because it's benefiting us as individuals, a part of a collective.

And that's exactly what C2PA stands for. It's a new, if you want to social contract to manage or to preside when it comes to the destiny of data points in a new world where you cannot say what is synthetic and what is real Impressive. Exactly.

It's needed. Oh, oh yeah. It's definitely needed.

It's definitely needed. Um, so now, as a, if I want to try that, so I can go, if I go to, open it to ChatGPT and ask an image, the image that it will generate will contain this credential. Yes.

Yes and no Karen. It's not, it's not systematic. I mean, we need to be Got to ask for it.

You need to ask for it, that's for sure. And I'm not certain that, you know, ChatGPT, with all due respect, will execute. What I know for a fact here is that they are embedding that at every single level of their, you know, technology, not because they want, I mean, let's also face the truth.

They are not, you know, willing, ready to play the game because they must. There is a very important, I would say, weight for C2PA in the EUA Act. And in the NIST RFM, C2PA is mentioned a couple of times the backbone of the future regulations for AI.

But it's not systematic yet because there are some issues, and this is I think why we are discussing today, C2PA is not the perfect world. I mean, it's not wonderland. I mean, please do not think like that.

It's a layer one, meaning an infrastructure layer, and we need to build on top of it a lot of other things that are missing, like identity access management systems, where blockchain, by the way, will play a central role if you ask me. because we need this kind of stuff. But because it's not, you know, fully cooked, if you want to, you need to add your touch.

So it's not systematic deployed at OpenAI at the moment. Oh, okay. Well, if you have a model that you know, does it, then you want, you can get them consistent.

Yeah. Okay. Yeah.

You've got, I don't know if, you know, they change their models and, and if you use the chat interface that if you use their APIs, you choose your model. And so, uh, I know that I've consistently gotten content credentials out of, out of Azure's implementation of DALL-E Okay. And also out of Sora, which is pretty cool.

So, you know, it's probably, you know, whether you are using the public facing client or not, you know, I can't see how that would be hit and miss. Yeah. Depends on what model they, farm it out to when you make the request.

Poor DALL-E. I'm gonna miss DALL-D, DALL-E's being retired. It's like a old friend going away.

It's like, I'm gonna just use it right up to the last minute and make images. Is it? I haven't seen that.

Is it going? I've not used DALL-E for a while. I was playing ChatGPT recently, so with the 5.0 model 5.2 models.

Uh, so, um, yeah. Okay. But, so one thing, you know, Ôtez-moi d'un doute as we would say in French, remove my doubt from me.

Uh, so any content, text, my my, the book I'm currently writing for O'Riley, um, illustrations, anything, Everything. And I'm gonna be even, you know, we work on text, obviously PDFs and texts and stuff like that. We work obviously in AI and machine learning that Karen is leading with uh, technical working goal.

Uh, but we work also on audio. This is what we are trying now to solve and, and to tackle. It's not an easy one.

Uh, and there is also a live video task force. So yes, to answer your question in a variable way, we tackle every thing. So, I could take, you know, I work a lot on data contracts, so I could take a data contract and certify it with C2PA?

Yes, definitely. Okay. That's, I see those wheels turning.

Yeah, it's a nice trick to have. And, you know, one word that hasn't come up in the conversation is agentic. We might be the first You must be in a tunnel.

I didn't hear you right. Agentic. Agentic.

Oh you can hear me. Can you say, okay. So yeah, agents, um, seemed to be the thing of the day.

And that was something that, you know, I knew that it could happen that, you know, we'd start dealing with computers that would then go deal with other computers on our behalf. Yeah. I knew that was coming for a long time.

But, what I didn't foresee are the power of the agents that people are building. And, and it's, you know, because when I wrote Blockchain Tethered AI, you know, I'm thinking, well, I need to get this into the hands of the top engineers, and probably nobody else will even read it, but it's gonna be in the hands of the engineers. So they'll have, somebody will have suggested this, right?

A way to, to keep this from getting out of control, because technically AI could get out of control. You know, we haven't spent a lot of time talking about it, but there are definitely aspects of it that could get out of control. And, and, you know, like I said, I was thinking of it from the terms of a great big company losing the ability to call an agent back.

And I thought about things being digital twins and massively parallel. But what I didn't think about is my brother-in-law is doing it. You know, I don't know your brother-in-law, but, but yeah, it's becoming very popular for sure.

You know what I mean? And, and so it's when someone, it's like, when I was a kid and my parents bought me a chemistry set, and first thing I did was wait until everybody was gone and make the stink bomb that I wasn't allowed to make, I had to bury it in the yard. I had to actually bury it in the yard.

But, you know, what does somebody do if they make the, the agent equivalent of a stink bomb now? And, you know, what do they do? They, you know, if they haven't built any controls and ahead of time to be able to stop that and call it back and say, We, we're still a bit in the wild West.

Okay? And that, that's also why I think that the whole idea you had with tethering AI with blockchain, allow us to actually safeguard us. There, and, and now with agents being the new wild west, I think you're, you're totally right.

We, we need to be able to have, we can't, I think it's very difficult to control the whole thing, but at least to make sure that what's coming out is what we expect is coming out from the, from the real guy, right? From, uh, from the producer we expect it's coming from. Uh, I'm always, you know, I'm always very surprised about the, the dynamic when it comes to eventually debugging, okay.

All this agent to agent communication because it's, it can go, it can go crazy very fast. Okay. And, um, and monitoring all those things, I think we will have to have some monitoring agents at some point to make sure that this is not completely crazy or going completely out of control.

So I think, so we can build also, you know, if you want to, some, some rules if you want to, of our social contract using explainable AI, that will force the agents to go transparent. I mean, I think that what you just mentioned here is that the black box era is over. Where you are putting everything you wanted in a training model, and you could do whatever you want.

I think that now, especially because people are not trusting 100% AI, a transparent, you know, box here, glass box would be way better for us to advocate our cases that what we do is the right thing to do. And we are moving to a more mature phase where people are asking those questions, how can I trust you? I cannot trust it.

And C2PA is a part of the answer. And blockchain could be also a part of the answer because it's immutable logging. And when it comes to identity, if you connect data to identity, you need both things.

You need a system to track a provenance and authenticity, meaning that this is a real picture that was taken by a real guy in the street, but you need also to manage the ownership of this data point. And at the moment, C2PA does not solve it. And we need definitely to bring an element of decentralization and DID or even even more self-serve identity to cover, you know, this, uh, this gap and to make sure that we are doing the right thing.

And that's the next challenge for us, I think. Yeah. And it comes to, you know, there's, a line of work which I I'm working on here at Actian.

And, and I'm not the only one. Ole is contributing and others about what is AI already data, okay? So I'm in Austin because there was Data Day Texas here, and we had a discussion with practitioners trying to understand what is, what is data and how is it really to, go to, to be consumed by AI and especially enterprise data.

Okay. So there's, it's a, it's a wool challenge. Uh, and, and especially when, when the architectures are, as you said, Beni getting more and more distributed, more and more decentralized.

How do we ensure that there's some something, that it's not a complete BS when it comes to, to the CEO's table. Okay. Or, uh, so that's, but that's, that's a very, that's a big challenge.

Um, we're almost at time. I wanted to let you, anything you want to say, Karen. Where can we meet you in person?

Where can we meet you online? Um, what's your next future, um, in the next six months, one months, whatever? Well, you can, uh, I'm easiest way to get ahold of me is on LinkedIn.

And, um, so I'm LinkedIn, you know, linkedin.com/in/karenkilroy, all one word. And, so that's the easiest way to find me. Um, and, I am just, I don't think I have any big events coming up or anything big planned.

Uh, I am open to coding. And I am also looking for school districts that are interested in having hackathons. Also youth organizations, uh, any type of group that wants to introduce their community, that has instructors, teachers that they can keep at the center of the instruction.

That type of work is what I'm looking for. And I'd also like to read your quote off the back of my book, if I could. Do you mind?

No, I don't. Okay. I was really happy, really thrilled to, to have Jean-George, review my book.

So, and this was his, his comment that got picked, it got chosen for the back cover by O'Reilly's editorial team. He says, AI is a superhero toddler. Without meaning harm its laser eyes could reduce the city to ashes.

The solution? Make sure the toddler does not have a tantrum in Times Square. This hands-on book describes this solution precisely and pragmatically.

And I still mean it. Thanks , Karen. Still mean it.

Yes. So, you know, this isn't light reading. The first few pages are, the first few chapters, are kind of stories.

They're fun. Anybody could read them. As you get down into it, you might wanna flip through, but I think anybody could learn something from it.

And, uh, for engineers trying to figure out how to implement the workflow that involves traceability for AI, this is your baby right here. because we figured out all the pieces. Yeah, it's not, it's not, I would agree it's not always an easy read, okay?

But it's a very useful read. And, uh, and if you are, if you're not too technical and you don't care too much about, but trying to understand the big concept, that's the first part of the book. And then if you are really, if you, if you really like it, then the second part is all the hands-on part, which is just fantastic.

Where, where you've got all the details to make it actually work and the, and the repo as we mentioned. So yes, Beni, your turn. Oh, you can find me on LinkedIn.

Okay.

Uh, I think that my next adventure is to definitely, you know, bring trust, uh, at the level of the enterprise, you know, need for AI and to try and manage to do what we did with Hollywood. Before our time in Hollywood, no AI. After our time in Hollywood, AI is everywhere, but in a responsible way.

And we want the same thing for the enterprise world. And I will quote to end, maybe one of my top philosophers again, his name is Tan Lee. He has created Spider-Man.

Uh, and he said that with great power comes great responsibility. And this is really power. Indeed.

Indeed, indeed, indeed. Well, and place you'll be Beni in where we can have a coffee or something, Uh, I will be in Washington, DC in March if anybody's there. I'm a fellow at George Mason University in America Center, dealing with political economics.

And I will be in Paris also for the race summit in July this year about AI, and you might even see me on stage. Okay. Alright.

Fantastic. Well, Beni, Karen, this was a fantastic chat. I'm really super grateful.

I knew a lot about, about what you were doing, but it's the same every time I had a discussion with Karen, I learned something about her. And we didn't even mention David Bowie this time. Okay guys.

David Bowie. Yay. We can't do one without mentioning, I did mention Jean Jeanie, which is a David Bowie song.

So thank you for mentioning David Bowie. Thank you everybody. Alright.

Alright. So thank you very much. Really appreciate being here.