Forging Private Agentic AI

Transcript

Uh so essentially like my name is Sandy Fatic. I'm the CEO of Cala Network and Calime Network is essentially an open source framework which allows developers to build self-s sovereign applications. So what is a self-sign application is an application that the user owns the data and today I like is going to be a talk about partially philosophical and partially technical and it's going to be my experiences and like why I joined deputy and why I think we should uh you know nurture like what are the values we should nurture in this industry and we forget forgotten many of them so disclaimer I'm not a philosopher uh I'm just an engineer uh with many years of experience working in big tech. So I've

been at Facebook, I've been at Google and several years ago I joined near protocol as one of their first infrastructure engineers. The guys who are interested in AI uh the founder of near protocol Ilia was the guy who invented transformers and um I really had you know good time working with them and three years ago we started this idea of like how do we bring privacy and data ownership to the users. Um and this talk came as like a combination of few things. So I am a guy who really likes to write really like sci-fi stuff. So sci-fi uh is one of part of like my life. Uh the second second I built really big uh infrastructure systems and I used ML before. So I bu worked on like automatic detection of traffic to

prevent EOS attacks. I worked on automatic releases which uses which use ML. I work on security analysis which use ML not AI. So I'm going to talk about like few things. First thing what is cippher punk? This book never roamer is like recommended many many time by many many people. Uh I don't like putting in slides because it's recycled and every time seen uh but I think everybody should read it because it shows us what future in a digital world look like could look like if it goes wrong. So what happens if big private corporations own our data, own our thoughts, if AI is controlled by

only them? It also talks about you know the hacker culture about high gadgets and gives us gives us like the blueprint of how to live our life and if you think about technology in essence how I would say like progress of technology and privacy technology is one word it's called rebellion email started as a decentralized protocol SMTP we centralized it governments wanted to read all the emails we said let's build encryption bitcoin started as an act of rebellion in 2008 2010 after the 2008 financial crisis and I think like we are at the moment today where we have to start another rebellion

and the new the rebellion is against centralized AI it's against corporate AI surveillance and it's about philosophically preserving our essence so what are the stakes the stakes are very high so there's a concept by a philosopher called a panopticon. A British philosopher Birmingham went to Russia and saw a prison with one tower and on the tower you had one light and the question was like the prisoners are behaving pretty well. So why are they behaving pretty well and they only have one tower and one guard and the thing was like there was a light the light into the prison. The guards could see everybody what was going on but the

prisoners couldn't see if the guards is watching or not. So the the thing is if you're not if you don't know if you know that you're being watched but you don't know if you're watched at that time you will behave by the rules of the game. And I behaved by the rules of the game. I joined big tech companies and one of the first things they told me is bring your true self to work. I said great bring my tra self to work. There's like values we are connecting people. We are doing this. But a friend of mine told me like, Sandy, if you bring your true self to work, you'll be fired the first day.

And I said, I mean, probably, yeah, I would be fired, you know, that's why I'm not there anymore. But the question is like, no matter like, you know, what your political opinion is, what is your view, whatever it is, having centralized power is very, very dangerous. So, I'm coming from Croatia, an excommunist country. One guys could have been the war or the other guys could have been the war and they would put the other guys in prison. like doesn't matter right so if it centralize all the AI every time when we click something currently 90% of the tra data we currently have flows to centralized servers your email centralized servers AI we are chatting about our deepest thought terrible right

so what are the current challenges I see like what are the towers of surveillance today in the agentic AI space one is the boom Boom. Everybody's using AI and it's a good reason why we are using AI because like it is awesome right like you can write code you can you know analyze financial analysis which you had to do have you know 20 people to do it over a months you do it in in in days but all of these things bring privacy pitfalls every time when we do something on the internet when we query a prompt we give a little bit about ourselves about somebody and I ask like my team we're building in the privacy space and they told us like send we are building peer-to-peer networks storing local data

of users on the device and I said like it's a lot of data it's not I asked everybody like take a centralized server go export your data and see how big is the data footprint you leave the data footprint you can see about yourself the things you actually care about and then take a look about the terabytes and terabytes of data which is stored about you why are they storing when I scroll why are they scrolling when I click Why are they serving me ads? I don't want to see because they make money. And then the third like I think danger we are sering currently in the world is government regulations. So we've seen in Europe chat control which in my opinion is a bizarre thing

like we should not be able to see other people's chat. I think you know we went through crypto wars one crypto wars 2 like we shouldn't be discussing this again. But as long as there is a possibility that we can be censored and surveiled, we will be. So I say the corporations own the towers, but we own the shadows. The shadows in the prisons. So what are the shadows and how do we get more shadows? By building technology. Because cyher punk as the movement gives you one thing, two things actually. One thing is quite dark. hackers in, you know, basement and stuff like not not not [laughter] the most fun thing, but it gives you another thing. It gives you a motivational thing to rebel against the system. It gives

you a way to build and use technology to solve things which encryption TLS, right? We did it many times. So can we use new technologies like maybe these people are against them for them peer-to-peer networks to actually solve and build shadows. So the ciphery philosophy is you know humans against machines, terminator whatever. I think like machines are just the best thing which ever happened to us, right? So like we should be using them daily but we should really think about how do we preserve the freedoms?

How do we preserve our digital thoughts? For me I I think I already you know gave away too much about myself. I think for everybody in this room they also probably given away but like what is for the next generations? How do we build systems so we don't end up in a pretty bad place in the future? And then who are the true, you know, visionaries and heroes of the digital revolution? They're hackers. Like every single person here who is either supporting a hacker or thinking how do we build this or activists or people building this software, they are the heroes and we should not put them in prisons. You know, we should uh we have many cases like tornado cache and others. People write code. Probably

these guys didn't even know they were building something. They're just building cool stuff, right? And we should build cool stuff and we should build stuff which helps the humanity. And then the last thing we cyber punk is motivational as I said. Yeah, it's let's reclaim the power rebellion. Um so like what we do at Calimero, we built the shadows as I would love to say it. So, Calro is a peer-to-peer layer two, layer zero, I call it. It's not a blockchain. It's a peer-to-peer network which allows you to store your data on your device.

Your device can be your server, it could be your laptop, it could be your mobile phone you have and then you selectively share peer-to-peer your data with peers. So, for example, instead of storing all your data on Slack, what majority of the companies here do working in Web 3 and you know, praising Cippher Punk, you should have a peer-to-peer network of your employees where your data is encrypted and stored on your devices. If you're using Google Docs, for example, maybe we should use a peer-to-peer replacement for that. If you're building agents, you can use peer-to-peer networks to store, you know, uh, shared memory between them instead of storing them on a centralized server. And then this philosophy as I always said like we

are architects which are embedding values into code. Like really what web 3 was about and why I joined web 3 was because I saw that blockchains could solve my data not being resold to other people. Like that's what I cared about. Honestly, I didn't care about so much about the financial part. But I think the financial part is the critical component to make it happen. So what we have to do I don't know blah blah blah. [laughter] Sorry, I I lost the thread. But essentially like we have to build shadow agents and there's like many tools we are currently building. So today we had like X42 is a great protocol. This was HTTP code which nobody used. Coinbase said let's use it to have micropro payments so the agents

can interact with each other. Problem it's not private. If you make you know payments on a blockchain and it doesn't use private payments you have a problem right? Like still everybody can track you. um encryption really important other technologies etc. So I wanted to say like um X402 meets private payments. So we should build open standards like this where we can use USDC other stable coins even tokens to essentially on Ethereum or other chains enable you to be do web commerce between the agents because the agents have to pay for the compute between themselves. Then they have to be autonomous so they can you know have a wallet and interact. We have seen that

people have built already AI agents which trade or make money and it has to be cipher funcommerce. So it has to be peer-to-peer. It has to be encrypted. It has to be available to everybody. And how I see the agentic web is very much of how we built the internet in the first days. So like if you think about like DARPA for example or SMTP it was you run a central you run a cert service not for other people but for yourself and then you peer-to-peer like you send an email to him and he gets the email he pulls the email right and like why did we get to a point where somebody's hosting for us this stuff very simply because it was cheaper the user experience was easier and like we

at this point in time should go back to that so like you should have your own agent which is maybe you know a TE if you trust a TE or you should run your own agent on your laptop device and the agent should be able to just go directly to the other agents without ever depending on a centralized party. So the vision is like how I call these technologies and how I see it is like every piece of this technology I call like liberation protocols. So TLS was a liberation protocol. X42 is a liberation protocol. Peer-to-peer networks is a liberation protocol. Bit torrent was a liberation protocol.

Encryption is a liberation protocol. I have how much time? Seven more minutes. So I have like one last thing which I want to ask to everybody like I I didn't want to make this talk about what we do and shill you on the technology which we have but I wanted to tell a story of what got me into web 3 and what I think are the values we forgot because web 3 today is you know memecoins investors investing in a project which goes up and dies and all the fundamental values disappeared.

So if you get only one thing from this talk is first join the rebellion and second don't forget the values of web 3 and the true cipher punk movement. Thank you for the time. Uh build a future where data is not a currency it's an armor. And if if you want to learn more about what we do as a company or if you just want to spin ideas or like talk about cypher punk um let me know. Yeah. >> Okay. I I feel very inspired right now. One, two, microphone. Yeah. Um, okay. We can start the Q&A session now. So, if you have a question, please raise your hand and our volunteer is going to pass

you the microphone. [snorts] >> Vent, [laughter] do you have a question? Okay, I guess that's it. Let's give a huge applause to Sandy. [applause] >> Now, there is a question. >> Oh, sorry. Sorry. We have a question right there. >> Almost forgot about me. Um, can you explain a bit more about how uh you tackle in Calimero uh the issue of X42 private payments or like private X42?

>> Can you repeat the question? I'm asking how do you guys at Calimero are tackling the issue or like the the thing that you call like private x 402 how do you tackle it >> okay so like uh exo x2 so essentially the way calro works is that you have a node a vasom node which you install on your local device and it runs a vasom binary whatever the appear is and then let's say the applic binary is a select chat you have a select chat I have a slide chat just to explain like high level what we do right [snorts] so the way it works we use we don't use a consensus like blockchains do but we use a CRDT a conflict free replicated data type essentially every time I make a

change on my local device I broadcast it to the network and based on the CRDT everybody knows how to apply the changes I made to their local storage and verify if the changes are correct so for example if I send a message I'm I'm the only one who is able to edit this message and you can verify it on your side so what What you can do is is a general purpose framework. You can build different things like a knowledge graph for example. You could have an AI knowledge graph which shares the knowledge knowledge graph between the AI members for example, right? And then if you every time when you make a transaction between the agents, you could just call from the node to you know make a payment or make microtransactions or make a subscription for a certain number of requests. So what it could be it could be like I just

get I had dump my knowledge graph in a peer-to-peer network and every time the agent wants to access this knowledge graph it pays like that could be for example one one example um or like you could have like one of the things like somebody built on a hackathon the guys were building like they wanted to do federator learning but without ever like sharing the data but sharing the weights. So what they developed on Calimero was a weight sharing uh weight sharing like algorithm where I train using keras my model locally I upload I just upload the weights to to the network then somebody else continues they run on their data and upload the like upload the the updated weights to

the network as well and they specifically build it for medical because like medical institutions have a lot of data about patients but they don't want to share it. So for example like you can have just bunch of nodes across the medical institutions and they just run a model locally and I mean how X42 ties into that like you could have you know paid models for a TE like instead of you running the the model you could just pay a TE or you could run your own model or like have I don't know inference whatever right uh but like you need a way to pay these things right like we could go to a subscription and integrate Stripe but kind of defeats the purpose right like blockchains are really good at u making payments They're better than anything we have out there.

They're cheaper. They're faster. Uh so I think like that will be any agent can have a private key and a public key, right? It cannot have a bank account. Uh so I think it's like just uh a natural thing to go with it, right? I mean I hope I didn't go too long, but okay. Okay. Any other questions? >> Yeah. In the right row. Okay. So, thank you for your presentation. Something I was thinking is just like Linux is becoming more and more common by normal users, let's say, not technical people as well as is

Android getting more and more popular. Do you think it will be possible to have some kind of AI that is also open source in a way or is not keeping all the data or is it inherently centralized in that way? So if I understood the question is open source model are becoming more popular. Do we be do I believe that we will have a model which >> is as good or what? >> Having some kind of LLM or some kind of AI agent that >> I mean we have open source models but most of them are built by big tech companies which open source the older model and keep the better one right. Uh

so we already have it. Uh the question is like there is a bunch of researchers who are trying to build you know data like decentralized data enclaves enclaves where you could like build a federated model um from the open source data right uh I think like just it's a matter of time when you'll get to a point currently if you have more data you have the advantage right if you have more compute you have the advantage right so it's harder to compete with the latest whatever is from a centralized company which has a lot of data uh but I think like we can build systems where we can build markets of data to train on without exposing user data, right? Like having encrypted in place where you pay uh and then you run it inside an execution environment or like I mean if

becomes really really fast at some point you could run on top of in the future. Uh so I think yes yes I think like if we stop using centralized services and optin into like data marketplaces which actually care about her data then we could build a model where you can we can not even build agents but actually build a foundational model which is decentralized. I think that will be much more interesting in my opinion. uh like if you can get to like a foundation decentralized foundational model and then you know have some types of racks to access different types of information for certain things uh to operate on like that would be something which I would be very cool but like I mean it will take a lot of money resources time and it's probably like much harder than to build an agent.

>> Thank you. >> Yeah. >> Okay. Okay. Any other questions then? Let's just give Sandy a big applause. Woohoo.