Gianluca Di Bella - Trust in the Age of AI: Deepfakes, Challenges, and the Web3 Solution

Let's give a a big applause to John Loca and uh let's have we show a good speech.

Thank you so much Yonut. Thank you. Okay guys, so uh thanks Yut for the intro. Just a few words about me will be really low because I don't want to take so much time about myself. Uh today we are going to talk about an important topic.

Everybody of us uh we are using CHBT and EI in general in our life but anyone is actually talking about the issues of this technology, right? because as any innovation we have the pro and the cons. So um a few things about me. I'm Jaluca. I'm from Naples but I'm been living in Amsterdam from around four years where actually I build uh three startups.

The first one is MGS is a web 3D firm. We work with Ethereum Foundation and to implement permissionless smart contracts. Then we have next labs that is a crypto index uh startup and there is also napulit I'm one of the founder one of the major shareholder of this amazing event. Uh so in a nutshell today we are going to discuss about you know the issues of the AI right why uh actually like we don't understand what is happening with the this word this sea of actual content generated by EI what actually what is the main issue okay we still trust in what we see and what we hear actually so let's start with a small game this a bit easy okay EI or real tell the friends. I think we can all agree that on the right the one generated by AI.

Okay. So this is kind of like easy. So on the left is a real person. On the right we have BI content generated. Now we will go a little bit harder.

Okay. Okay guys in your opinion which one is generated by AI or by actually it's a real picture any takes right. Yes. So on the right we have the one generated by I and on the left we have the the real picture. So as you can understand the difference is really really hard to tell right.

If we see actually another example this is even harder in my opinion like okay in your opinion which one is EI and uh or the one that is the real picture and it takes left EI actually is right. Right is the one with the eye. So as you can understand it's really hard to discover which content are generated by AI and if we talk about picture about food maybe it's not really the issue but uh if people or maybe scammers or people with bad intention start to create fake documents or fake contents that that's might be a big issue right so uh I like to to call this big wave of problems that we're discussing deep fake right Deep fake is actually for me the dark side of AI and is actually like um the the fake content generation about contents that actually you might think are real but are actually not. So let's discuss about the associated dangers of BI. Okay.

So we have as first the spread of misinformation because you might check that actually you think that contents are actually real but again there is a misinformation because contents are not real anymore then we might have a issue about identity for theft and fraud actually and I'm going to tell you an example of this and also uh there is of course an issue of manipulation of political and social opinions. So you can see like you know we take this example of this funny name that is always everywhere the name of Mario Rosi as you as Italians or maybe international people we always use Mario Rosi to do examples. So uh the funny thing is that this guy uh call it David Aurora made a test to create a fake u patent. Okay a fake driving license. This fake driving license actually passed the control.

So the police that was controlling the actual uh uh the actual document to check if it was real or not. The crazy thing is that this fake AI generated driving license pass the control. So you can understand that immigrants or people that are not actually from that country can fake documents and can fake to be citizens. Do you understand what we are going here? Like it's a really big issue.

So what we are trying to do and that will be the main focus of this talk and how we can actually fix this issue right once there is like let's say there is the Nike brand or the IDAS brand or whatever we always see the C as copyright right so as blockchain nerds we are trying to create something similar we are trying to create a flag a page a a symbol or whatever is actually representing that content is generated by I and by content I mean that could text, could be video, could be music, could be everything, right? What we need to do, we need to implement a page or a system that actually explain that that content is generated by artificial intelligence. So why this is something that you know people are actually not discussing about, government is actually not caring because you know we live in Italy, most of you are Italians. We know that you know I always love to say that innovation goes as a Ferrari and government goes as a panda right because every day we have like a new vibe coding tool we have a new uh we have a new tool actually as uh you know that actually can do everything while on the government side the legislation and the law is always low. So the main issue about that is actually that the government dependence has always delays.

They have always bias and and of course there is also a lack of global interoperability. So governments are not communicating each others right. So we have this issue that centralized platform are really vulnerable. So for example like the most secure thing to verify identity or is some sub with KYC verification or is like the recapture you know once you log in to a new website you have the recapture but guys like an EI can can manipulate the recapture in a few seconds now but we understand that the recapture is a technology from 2000 and we are in 2025 and we still have the recapture. Why is that?

Because there is always like you know the limitation around the centralized platform. So we need like something from the government you know of course the innovation is faster but we need something from the the you know the the regions of the governments as like European Union as US and different alliance that needs to come to businesses and to people to say hey let's implement this beige or this EI verification system okay now in words we are all good but in fact how we can actually implement and verify that those content those contents are actually generated AI and that's why uh we love blockchain and we love open source technology because actually thanks to blockchain we have the possibility to verify if a content is generated by EI or not. So the blockchain actually like you know people once I'm in the crypto industry from 2010 in the beginning was like okay bitcoin was a scam, crypto was a scam now you know banks are buying it, Trump is sponsoring that you know like we have so many endorsement worldwide but actually like I was never uh so much interested in the speculation. I always loved the tech part right as a real nerd. I always love like the part that actually was everything transparent, everything trustable.

So we can apply those value and those fundamentals of the technology as immut immutability transparency to actually also this issue to verify EI contents. So this will be a little bit more technical. So we'll be a dive in the technical part and let's let's explain a bit like what actually uh we are going to discuss. So we have the blockchain that is a data layer and then we have AC contracts as smart contracts. So we can think as smart contracts as algorithms on top of the blockchain.

This smart contract that we are going to discuss is called SC 77 that me and our company in Amsterdam we help at the Ethereum Foundation to rewrite with the Ethereum improvement proposal 707. So just in a nutshell if you don't know how it works if you want update a code in blockchain because it's not decentralized because it's not centralized you need a DAO right so every time there is a new commit a new update you need to submit a proposal to the DAO those submissions are called Ethereum improvement proposal EY so once you see EY is a proposal once you see RC is already accepted. So one year ago this 7007 was a proposal now thanks god is accepted so we can use it. So we have the SC707 that actually is this model this algorithm that might help to actually check content verified and generated by I. So how it works?

So the SC uh is actually a model on top of the traditional 721 that is a traditional model for NFTs right nonfgeible tokens. So you can use NFTs to represent real world asset as art, music, real estates and so on. In this case we are going to use this NFT to represent identity. So this SC 721 will be customized to represent digital identity. Then actually how we can verify an identity without having troubles with confidential data with this example.

I always love to say the example of the bouncer. Okay. All of us goes to clap, right? We all of us love to to dance to disco to clap and so on. But have you ever thought every time you go to the bouncer why the bouncer is asking the ID card, right?

actually which type of data he needs to check just the date of birth right because he needs to check that I'm over 18 while once you give the ID card to the bouncer he's checking my name where I live all my personal data that actually I might be confidential you know I don't want to actually share those type of data so that's why that we enter with the technology as integration of ZK proof zero knowledge proof technology this type of technology is amazing because we can show on chain some data and we can actually not highlight other types of data covering confidential data. So in this case of the example of the bouncer might be great that I show just a document or a digital identity just with my data board without sharing all those other types of data. So we integrate ZK in this model to avoid to share confidential data. So I have to prove my identity right but I don't want to share where I live how many how how much money I have in my bank account and so on. So the whole concept of DK is to show the exact data that provider needs to check without showing all the other confidential data.

Then we have optimistic machine learning but actually what we do we do a onetoone match to match the identity to see if actually is existing right the identity. So we are going to check uh government databases that I don't know if you know but are public and you can access them with an API and the flow is really is really is really simple. So we tokenize the the identity with SC 721 standard the identity becomes a digital identity tokenized in form of NFTs. Then we have ZK to actually highlight the most important data to don't show confidential data. And then we we have the optimistic machine learning to do a match one to one to check if that data actually exists and makes sense.

Then uh for the more technical I have also added a flowchart that explain a bit how actually those uh uh uh those flow actually works. So we have the model node that is divided in the machine learning model that actually has a an infinite loop of prompts that generate the check if the identity exists in the government database and then we have the ZK machine learning prover that actually proves that the data exist and hide the confidential data of the person because I don't want to share my address in blockchain right I don't want that people comes to my house or whatever then uh we All a web socket where you see those type of wait where you see those type of white thingy we can call it as a as web socket like this is like a web socket while those are the nodes. So the nodes are the machine learning and this is ZK machine learning improver this is all a web socket. A web soocket is just a docker that runs in loops the uh the algorithms. On the other end, we have the smart contract that actually uh verify if the data are making sense again and then we mean in form of NFTs.

So why all these guys? Why we actually need all those nerdy complicated things? This is no worries is everything in back end. So you know the nice thing of technology that everything is in back end you don't have to actually check nothing on front end everything will be in back end in a few clicks all these projects will be run what you see in front end is actually a dashboard where you can upload your text where you can upload your picture where you can upload your video or your music or whatever and you are going to have a message to check if that uh text is generated by Yi or not as we have right now with the X right there is like CH GBT or there is GBT zero that might check if your text is generated by Y or not. Imagine the same concept but applied everywhere for every type of content and also we are going to add this beige on top of the content.

Then we talked about in theory right but uh I really like tangible things. Okay. um which type of real world application we can we can have right so in a tangible way in my opinion one of the most important um practical use cases is the securing the journalist journalistic integrity because I remember when there was the um in US that there was like you know all the all Trump against Camala race and so on online were circulating a lot of fake videos about Trump or how many times s you are getting contacted from Indian scammers or whatever uh strange scammers on Instagram that you're like oh win uh $100 with the fa with the face of Ellen mask or whatever that's all the issue about the fake right so the concept here is like to try to securing journalistic integrity so all this flow of the SC707 to verify EI contents can be applied for journalistic integrity so will be really nice to check that that journal or that data that we're reading is actually is actually verified and actually is a real data. Then we have of course realable social media verification. How many influencers now are doing EI contents?

Come on. We are already having AI influencer you know we already having like people that were faking their picture modifying them. Now we have also AI influencer guys this is too much you know. So we really need to understand how prove that that data is real. Uh then other practical use cases I have just five minutes so I will be a little bit faster.

Uh we have uh protecting intellectual property preventing academic plagaries. All of us we use judgbt for our essays but you know in the long term I don't know if it's really good that so we need some something to pro to prove that those contents are real. And then of course enabling trusted voting systems because now EI bots are used to vote. If you want to buy a new shoes, let's say a new drop of D Jordan, we are going to have thousands of bots EI that are going to spam the the actual form. Also like not just for voting system for every time of of system that actually we need um to buy something online or any type of of interaction that we need to do online can be flagged and can be manipulated by AI bots.

So we love VI agents but we need to try to moderate them. That's all the point. So all this uh to say that we need to implement you know those type of technology to increase trust in digital transaction so we can verify that a transaction is made by human and then of course to try to make more responsible and uncountable AI contents creations. The conclusion is that um you know we need more trust in the in the era of digital age because guys in five years trust me if we don't start to use blockchain to verify EI contents is not in five years like in two years we will be not able to see what is EI or what is not TI. Okay.

So that's why we're are pushing with governments especially in Netherlands and I'm trying to do my best also in Italy to try actually to to start to implement those type of implementary technologies to try to verify those type of AI content generation because otherwise sorry for the word in a few years we are completely and we will know which content is generated by AI or not then um you know we need to support the adoption of this RC707 that me and my company helper to Right. And of course like it's important to understand I just made this uh state funny thing with Chibbt. So I I use a DI to verify this content. Uh you know blockchain is just technology is a trust engine in the age of VI. Now what we need to come next with integration, collaboration and civic empowerment to try actually to uh implement those type of VI content verification.

Now uh my talk I think is arrived to an end. Um and uh I don't know which type like I actually I'm really happy because last year I didn't have so many people in the in the in the stage. It's a bit crowdy. So GG all. And uh we have different type of people here.

Okay. We have the designer uh we have web three newcomer or like uh newbies we have crypto founders and we have web three industry people. So all of us if you have any question we have like a few minutes and I'm happy to answer you. Thank you so much.