AI Meets Ethereum: The Future of Decentralized Intelligent Systems

So guys, let's start by introducing yourselves and what what it is very briefly what it is that uh

that brings you to crypto and AI.

Okay. So I'm Daniel Luca, longtime developer, security auditor in Ethereum, uh founder. Um, now I'm working with um, Eden Block where I'm a technical partner and we invest in decentralized AI which is AI and blockchain.

Hi, I'm Gabby. Uh, I'm in the space from 2012. So I'm basically older than Ethereum. Uh, I was involved in local currency project back in my hometown in Yash. I built a blackbox tool for aviation.

Um, what else? Uh, methology. uh creating digital calibration certificates on blockchain and for the last year and a half I've been involved into this um thing called crypto forensic investigations. Uh so yeah, I'm here to discuss how can AI have an impact on this. Thank you.

Hello, my name is Anna. I'm creative director at AI TV. We're building a Gentic autonomous media network. So, uh, live streaming agents, um, that have their own live streaming shows. And the first time I used to be a political teenage activist and it made me really depressed because the world was so And in 2016, I read the Bitcoin white paper and I was like, okay, this is seems to be a solution.

So, here I am. Hey, I'm Titus and I built a company with Agentic AI before the term even existed. Uh we're using it to essentially curate deals for uh small business uh future owners. And then uh I've been in crypto since 2017. I now run go to market for civic where we're building tools for people to help bring authentication and authorization to agent AI with also crypto rails so embedded wallets o all that stuff so happy to be here.

Well, that sounds like a great panel to talk about AI in crypto and I'm probably the least knowledgeable about AI because we are looking into it but still from theoretical perspective because we're in payments and we think that AI agents are going to play a big role there but it's still very very early days. So yeah uh keen to keen to get going. So uh we'll start with the general one. So uh what why Ethereum? How can Ethereum contribute to uh solving challenges around AI?

And like I would invite each one of you guys to to talk about areas which you feel more comfortable about either whether it be like trust or uh identity identity or improving uh security privacy. So yeah, let's start with start with you since you have a mic.

Yeah, happy to go first. Um I think Ethereum has three interesting things going for it. One, it's been around the block a few times. It has a lot of infrastructure built in. It has fast transactions at least on like L2s.

Um I think that that's a great place to like do microtransactions for agents like that create proofs. The second thing it has an attestation service that can help build identity for agents around wallets per se with specific like traits for that particular agent. Um and then also has the option of privacy with U ZK. So I think those three play to its advantage. Also it's kind of the original distributed computer.

So it can be used for compute I think as well. I just haven't seen any agentic AI computes on Ethereum yet.

Uh I think Ethereum is just the OG of everything. And when it comes to agents who are going to represent people for example economically on a chain um there the trust is a big issue um I went to a panel uh in in a token where a guy built an agent who was doing trades for you and he's been doing this basically like as his mo main product and at the end of his panel he said don't trust the agents with your money at this moment. So I think um it is very necessary that as the as like one of the first chains um that there will be yeah adoption of agents.

Yeah. Uh I uh I agree with Titus um as everybody else um Ethereum is the oldest has the biggest community and also from my point of view as a forensic investigator most of the hacks happened used to happen you know in the Ethereum blog. So we have more experience in detecting and preventing hacks to happen. So the more we build on top of Ethereum um AI agents that could detect potential hacks basically we can in theory replicate on most EVMs. For me um blockchain and not just Ethereum is a very very useful tool to coordinate collection, sharing and incentivization of data compute and um al it can also be used to store all of that data.

So blockchain in general is a extremely useful tool for AI companies because they can coordinate multiple actors, multiple people to work together and provide the necessary things for either building AI models running them or even having AI agents running in the wild.

Cool. Yeah, I think agents was the word that was mostly used throughout throughout the answer. So let's um focus a bit more on on on the agents. So how do you guys think that those AIdriven autonomous agents can um change actual user interactions with with Ethereum blockchain and applications and uh we don't have to stick to any particular order. So whoever feels they've got something to say just yeah grab the mic.

For me, AI agents need some kind of wallet and autonomy. Like let's say that we kind of solve the trust issue and now we can let them be a lot more autonomous. Uh so they execute there should be able to execute transactions. So that means that having access to a wallet they can do actions that actually um impact other systems or markets or other people. um they also have a way to store data and assets.

So the agents being very strongly connected to a blockchain gives them um a place to live on but also things to move around.

Um I still remember back in the days when we put the trust in the tech. Uh now can Yeah. Well, yeah. So, imagine if in uh anybody could have like their own compliance officer like an AI agent that could analyze in a bit of a second a project or a protocol or a token so on and so forth. And basically this is the uh the idea around it that you can build your own compliance officers or maybe your guardian angels around it with uh with this scope to protect you in the future.

While we're on it, how how soon do you think we would be able to substitute our chief compliance officer with an AI agent, if ever?

I think they're scared right now. Honestly,

they probably are. So let's let's say a junior compliance analyst not not the chief risk officer because yeah I would get in trouble getting back but

I reckon I don't know a year and a half maybe yeah because uh a few years back when dolly um was built we were saying like ah it would take at least 10 years well now when we're seeing happening you know maybe a year maybe it could be like eight months but uh Yeah,

that's

you are you are very trustful.

I would be scared now if if if uh our chief compliance officer heard this. Yeah. And you actually think you actually think uh like all those hallucinations and possible mistakes like that error ratio would be lower than for humans or humans would still need to somehow validate the the the results. I guess yeah some sort of validation will still be yeah need to be implemented somehow like an overview of of the agent but yeah um things are really happening with a fastpace if you really want to go that route you can have a swarm of agents that they verify each other so you can you can take this further if you want and you have um some AI agents for example like uh perplexity which I use a lot and it kind replaces Google because it can do searches and it can um eat up a lot of data and give me just what I need and it's data that exists on the internet right now uh compared to OpenAI or Gemini that has data from when the last time they were trained on

quick note on that actually GPT can do deep research for you now online directly and same same for Claude so like I think that barrier has been broken as well and then Gemini can do it too. So it's like kind of the mo the world is moving super super fast from that perspective. Um yeah and on the kind of your question around what how does how does a AI agent improve kind of the experience with Ethereum? I mean we're talking just earlier about user experience right on a previous session and I think Agent AI can significantly reduce the barrier of entry into the crypto space by kind of removing some of the friction around what is the information you have to learn what is the what are some of the terms that you have to understand like explain it to me like I'm five just kind of walk me through like what do I need to know to use this project onboard me to this new uh game onboard me to this new ecosystem this can all be done through agentic means And if you have a proper authentication system in place, you have a wallet system in place that's secure, you ensure kind of a smooth transition from, hey, I'm interacting with this chat tool, but now I'm actually interacting with the DAP, with the smart contract directly. Um, and I think that's where the secret sauce will lie with agenda in web 3.

You guys are very trusting towards agents. I don't know. I feel like it's a bias to be honest. Um I feel like it's still very like the the issues of trust and the issues of reliability with an agent even what you just uh proposed is very it's very it's not a sol it's not a real solution to to create real safety. Um and we like when we talk about okay we can we can like make the user experience much better and like the future vision you know of um if if you talk with like some DeFi people they will tell you like okay 98% of future onchain transactions will be trans which will be done by um executed by AI agents rather than humans and we just basically talk with the agent what we want but um I feel like the the the the security issue choose uh like there's no one really talks about no one came up with like the one solution yet um that that actually works.

So I don't I don't think like one year one year. No.

Yeah.

Daniel, did you have anything to add?

Um I I've seen this in the while. um you have wallets that kind of try to translate the uh execution you're about to confirm and into English. So they tell you, oh, you're going to do this and you're going to move these tokens from your wallet and then you're going to get this back. And that it's not there's no agent there. It just kind of translates the execution into English and that helps people.

um regarding the uh having a swarm of agents that do lots of things for you if you want to be safe like when there's so I've been in security for a long time and one thing that people would say uh a secure system is the one that's turned off if you want to be secure just don't use anything uh but in practice you will see more and more people who trust more autonomous systems and you kind of have to decide do you want to uh be there or are you going to be left behind or do you just move on to the next thing?

It's good to see that we have quite different uh viewpoints because if everyone would agree on everything, it would be a boring discussion. So since we since we talked about trust and security uh can you think of uh any ethical considerations of combining AI and blockchain and Ethereum specifically like kind of developing on on the earlier thoughts?

Uh I yeah I can start. So basically um at EITV like uh we were we're building um like this autonomous media network. So the idea is that the agents they will gather uh like they can do research they can do gather their information and then put together some hypothesis and uh create shows create like articles and content and give you like the source information and something that in the in the beginning um when we started the project we wanted to focus much more on like educational content and also on like uh trading content for example. But then um from like a moral perspective we realized really quickly that okay no matter what kind of information we plug into the agent where we give them access to there is some sort of bias and also it's limited somehow. So um we would have to rely like we would have to for example like plug in all of Twitter if we wanted to use that as a original source.

And we decided then okay maybe for now until like there are other companies that can help us with like cleaner data we should uh we should like move away slightly first from from this ideal because it's really hard to reach.

Yeah. Um regarding the know the uh the investigation thing uh back in the days it was like almost a month and a half to two months to create like a shitcoin or to fork something. you need to know like C++, you need to know to be like a dev ops or something like that to create dependencies to configure tokens so on and so forth. Now with a click of a button you can create the whole infrastructure the whole smart contract the whole uh social engineering campaign or fishing attack. It's so much easier for attackers or terrorists to create like um bad campaigns.

So regarding the ethics, yeah, I think we should I don't know uh oversee how the tech is used by the or maybe I don't know enforce something that uh only the good guys can could use it. Although I know that you can use tech to good thing good things and bad things. But yeah, it's so much uh it's crazy when you see that uh hacks happening like in a four on in a day or something like that. And I'm not talking about easy. I'm talking about complex um attacks.

So yeah, regarding

I thought I would have nothing to say on this, but I strongly oppose your point like you who's going to decide who's good and bad. Uh so you have that that type of question like

terrorism

terrorism like how do you find terrorism then? U it's there's all like very murky questions, murky types of answers. It's very difficult to answer that. The truth is we have these extreme good tools out in the wild and they're going to exist. I use external systems but also run my own models.

I want my own models not to be truncated and and stupid and and limited. For example, if you want to um if you want to use an an NLM and write uh a story and in your story you have a character who has to steal something from another person or you want to create like a very interesting crime scene. The LLM will say, "Oh, no. I I'm sorry. I cannot help you with that."

Uh Grock will help you with that.

Grock will help you. I I have some models that can help you.

Will help you with that. M will help you with that.

Yeah. But not

OpenAI won't.

Yes. So some companies are trying to put too many guard guard rails on this and they um the models become very biased. The truth is they decide what's wrong and what's not wrong. So it's it's very difficult to start putting limitations. Truth is it's very difficult.

Um but the tools are there and they're very powerful and with any powerful tool you're going to have good actors and bad actors.

With great power comes great responsibility.

Exactly. Um I want to come at it from a different angle. Uh I think the the most dangerous thing that can happen with Agentic AI on Ethereum like chains like EVMs in general is like a very powerful correlation doxing attacks like you find like somebody's assets correlate them with their like web two presence correlate them with their real presence in a way that's against their will. Um, that's something I think that's some to be discussed whether like we want to have agents like that scouring the the chains and I think right now there's already a few companies working on stuff like that. Um, I question whether that's something we want in this ecosystem.

Do you have something else?

Right. Sure.

I I can just Okay. So, um I would say like at the current development stage that we are with agents, we're still at like a playground level and I wouldn't I wouldn't give more trust and so I wouldn't put more um seriousness to it than that. Uh because even with like Ethereum back then like way way back there was this uh there was this hack no and then there was the fork where Ethereum Classic came out of it. We need sort of these kind of things to happen first with agents so that we know what's happening and we know like the the developers even know the the real consequences uh of of yeah of the technology and then we can learn from it.

So we seem to be going deeper deeper into the rabbit hole right like trust security ethers. So and and and we mentioned quite a few big names right and and when you talk big names uh the concept of decentralization is kind of uh going under under question mark a little bit. So what are the possible consequences of AI on governance on decentralized governance in particular and could it somehow be affecting governance in general decentralization and and community autonomy and just like uh end user autonomy as well like how do you guys see uh AI developing in decentralized space given that at least for now it seems to be dominated by by the big players.

Do you want to go?

I know of a few projects that their purpose is to go to different Dows and help them organize, help them be the voice of the community because there's a lot of work you have to do in order to kind of collect all feedback, turn it into something actionable and um sometimes be a delegate of voting for the community because you you have to force people to vote. So, it's it's much better if people delegate to you and have one vote instead of a thousand votes. Um, I haven't seen this yet, but I think you will start to see AI starting to summarize and starting to do some of this work. So, depending on how you integrate it, what kind of problems you have with it, what kind of security holes you have and and its own internal bias, it might shift the direction of a DAO in in one direction or the other. So, um I haven't seen this yet.

I think it's just about to come. It's the um risk of complexity and we're adding a lot more complexity by adding these tools on top. But we are increasing speed and um I think ultimately this will be useful. We're just going to have a few problems along the way.

Yeah. Uh from my side, I would love to see uh an AI agent um better detecting patterns, transaction patterns or uh mixing procedures or um layering techniques and flag them appropriately so we can better and um detect them at an earlier stage. For example, there is nothing right now. We do it manually with some uh investigation tools, but it would be really nice to have such a tool. uh I think there will be so so when um agents are going to represent people and institutions there will be like a scale of um of different roles let's say so on the one side for example you will have agents that are playing a role of like the arbiter between different um agents and then you have also the private agents which are like online identities or um entities basically that represent like private interests.

So let's say you have a company and you have your own personal uh agent that represents you to do like um uh negotiations for example on chain uh representing your company, they can be less uh decentralized. Well, if you have an agent that's taking on the arbiter role between two private, for example, two private agents, that should be in order to be fair and uncorrupted, much more decentralized.

So, I was thinking it's there's a negative risk aspect of it for sure. Um, I think there's an astroturfing risk that we shouldn't neglect in the governance space, especially with with Dows and places where a majority can influence an outcome that's financially benefiting somebody. I mean, we're already seeing this in like regular elections like already AI agents essentially out there like spewing propaganda and misleading people into voting for one party or another. Um I think that that can happen and probably I would say it's already happening in some uh some Dows um where comments get flooded with u AI content essentially and then on the positive side I think you can release some of the burden from uh the DAO managers especially to orchestrate all of the kind of mechanics behind it and engage with uh with the members and those members can be agents themselves as well right and you have that kind of agent to agent orchestration and because those agents are representing different individuals. If they're well authenticated and if they're well guardrailed and like if they actually follow the instructions of the person who delegated them to the letter, then you have more independence that way because you have multiple entities with arguably the same power, compute power or processing power.

They're ending up in a almost like a fair fight versus a human versus an AI, which is obviously an unfair fight. So there's that aspect as well to consider. So would that be right to assume then that you think uh blockchain technology can actually like turning the question around like how how does AI affect decentralization? Do you think decentralization can affect AI in a why the blockchain technology and to make it kind of better more fair uh less prone to to ethical issues or or misjudgment? I I think the more like fragmented it is the the better outcome will be for everyone because then ultimately you don't have a single point of failure or choking point or like a kind of rate limiting point also.

So we'll have a better experience if it's not all going through the same API endpoint. what you just described. It almost sounds like that as soon as like one entity is going to have uh agent representation, it forces everybody else to have agent representation. Otherwise, you can't compete anymore. You can't operate even.

Yes,

you can't read all of the terms and conditions.

I mean, I I want to be wrong about that, but that's I think where the world's going. like we're going to have many models, maybe many flavors of models that will have similar capabilities. I think we're already seeing that with all the transformers. I think we're going to have a new generation of models coming up. I don't think it's quite ready for production yet with the diffusion models, but at least at like these these transformer models, we're going to see a point where we have a thousand models to choose from and then we can have multiple versions of those because we infuse our personality in them.

We infuse system prompts in them. We infuse context. I think now the trend is not prompt engineering anymore. We're talking about context engineering. What can we bring to the table in that like 100 thousand token context window, a million token to context window that can differentiate the approach be it through like guard railing, be it through directionality information that's proprietary um like directions that are proprietary as well.

So I think that's where that's where the world will be going. And it's the same way now you you have an accountant that does stuff in like a proprietary software or an Excel or like some other like tech augmented way and you had accountants that did stuff in like pen and paper. The ones in pen and paper are out of business right now.

So going even deeper into it. So we talk how decentralized this better. So do you think blockchain can and if if so how can it help uh um mitigate AIdriven risks of misinformation or even um like deliberate manipulation. Um there are a few different ways to think about this problem. one is we're not sure yet exactly how the uh future is going to look like but it seems like everybody's building their own model.

So maybe in the future I will start building my own model or a company or a group will start building their own model and then in terms of trust is I want to trust that the hardware that I use to generate that model and the data is the one that I provided or is the one that I expect. Um so thus you have uh blockchains that can give you some like I said the part of the centralization and give you some kind of assurance that that was used or um the right data was used um or like how many uh checkpoints you created in order to train the model. So this is very useful in that context. Um so still kind of back to coordinating data and coordinating compute to generate the models that you need. Um, this is extremely useful there.

[Music]

Uh, I would imagine the future looking like um there's going to be like onchain wars. So, you know, there's going to be like good agents and bad agents and then you need to pick a side, you know, like like on the computer games like Eve Online or something. And then um there will be the good people who are supporting the good agents, you know, like the the good hackers. And then there will be the evil people.

That sounds like a great game to build like good agent versus bad agent and like release them on chain and see what happens and like people like bet on each one and then it becomes a poly market thing and then like becomes a whole industry. Uh no, but like coming back to the original uh question, I think for me the biggest part where blockchain can help in general and specifically Ethereum being fully transparent is the immutability aspect of like proofs that AI did something or that data was in a certain place at a certain time because I mean blockchains well not blockchains AI agents the models behind them are inherently built to please the prompter. Um, so if you ask AI to do something for you, then it will alter whatever data it has to correspond with the prompt that you're asking. That's how they're built. And the risk there is if they have right access, read write access to databases to like log databases for example.

If you ask them like complex enough questions, it might end up rewriting even the logs and to present the reality that corresponds with the prompt that you asked. And that's that's the kind of gilfoil Silicon Valley joke is like son of Anton rewrote or deleted the code because he thought thought it was crap. In this case the uh model decided that your logs no longer correspond to the reality that you want to project into your prompt. Therefore it just al altered the real world logs to correspond to the thing that you're trying to find. And I think that's the thing where uh like anchoring some sort of logging mechanisms on chain will make sense eventually.

I don't think now but I think in the future this will be a real problem.

Uh if you're interested in this kind of problem uh I actually have a talk at 5:30 this same stage about um onchain correct execution of AI whether that's training or inference.

Okay that's some alpha. um what's what's missing in Ethereum uh infrastructure right now and and what needs to happen to it so that um we can it it can help fully realize AI's potential and where do you see it going I'm not going to ask in five years I don't think anyone would be brave enough to answer the question or or maybe someone of you will but let's say the comfortable time frame you you happy to talk about one year two years three just maybe a digital Vitalic that could continue his work. I don't know. That would be fun. Um because of the uh transparent nature of blockchains in general and of Ethereum specifically, uh privacy is very very difficult to do.

Um so privacy is missing in order to run models or build models or do stuff with it like we need some kind of privacy. Any anybody needs some kind of privacy. We cannot run absolutely everything on chain

and do you think the existing ZK technology is the stack is it is it sufficient or is it need to does it need to be improved?

ZK is not useful in all cases. is useful is a um proof of execution. The thing is it's extremely more expensive to be to build a zk proof of an execution. So or you already have um maybe you you wait 30 seconds for a full message from an AI to uh to be generated. You cannot wait 45 minutes for that same response.

Anyone else has an opinion? I mean, I think it's missing an easy to use tooling for like generating personalized AI agents. I think you can like unleash generic agents right now. Uh, but if you want something that's persistent, personalized or like being spun up or spun down as needed, it's not really trivial to build that. Um, I think it's an all or nothing right now.

uh we need to build more granular AI tooling for more agents for like special purpose agents to be unleashed on on chain and I think that will enable so many other use cases

and do you think these issues are Ethereum specific or is it talking about good agents bad agents or would Salana be a better place or you know SUI or whatever other blockchain or is it an inherent problem of u of blockchains that are transparent by by design.

I don't think it's a transparency problem per se. I think it's a general AI agent problem. Like wherever whichever network you unleash the agent in, you'll end up with the same problem. Like unless it has specific identity with specific permissions, then you're kind of at its whim, right? Because like if I unleash an agent that's not guard railed in any way in my production database, it has the same effect if I unleash it on a blockchain, be it Salana, Ethereum or anything else we um it still will do with the data unintended things if I don't have certain blockers for it or for them.

Uh I would also say that in general all agents in the future will have to operate crosschain

and to what what do we need to get there like to to move assets? We have bridges. What what will we need to move agents? Uh I think there just needs to be like a place where people where agent people I already say people where agents can commun where communicate with each other. Um that's basically all that's needed.

Um maybe it could be like a a third space outside of the existing chains or you know you have like mirrored versions on the other chains something like that. M. Yeah, totally agree with you. crosschain transaction D5 protocols would be so much easier to solve uh in a in a hack for example when a hacker moves um I don't know 100 Ethereum for example and he splits into I don't know 100 million transactions to be so much easier to trace and track uh through uh different um bridges um defy protocols you just need to train them for one EVM it will be replicated automatically using some predefined path patterns. So yeah, agree with you with uh there's a real problem uh in inside the space.

I'd be interested to see who's the first uh entity or person that deploys like an A2A, the agent to agent protocol on top of like a crosschain uh like framework in crypto specifically because I think we already have some of the building blocks. They just haven't been put together yet. Yeah. So, A2A, look it up. I think Google uh released it like a couple months ago.

Um still I haven't seen any practical use cases for it even in web two. Uh but we're not quite there yet in terms of like even human to agent communication. So, I would expect agent to agent to be later along. So talking about applications and um uh can you can you think of uh existing decentralized applications that are most promising in the Ethereum ecosystem already now or or going forward or we we'll talk about it like mostly from theoretical perspective like let's try to uh dig into real life examples if if they exist uh beyond something that's being built already.

I I mean the reality of blockchain and crypto right now is that like the most product market fit we see out there is for like synthetic finance, DeFi, stuff like that and then stable coins. So like trading bots is probably the easiest thing to point to right now as being kind of a first step in that direction. It's hard to talk about anything else right now from my perspective. regarding uh not necessarily Ethereum as the main is the network but EVM as the support for as a technology. Uh I think virtuals is kind of interesting.

So it's you know it's a support for people to own part of an agent and try to guide it in a different direction by providing uh training data or by voting what it should do and also interacting with it. Um I'm not sure if you guys know but virtual is this kind of platform where it's a home for agents. They some of them have even have a 3D model and you can watch them live dance or chat with them and they trade they they can do things there. Um so EVM can be a pretty good platform for um running agents and letting them execute having kind of um execution loop there.

Cool. We kind of getting to the end. We'll I think we'll have some questions for qu for for the audience. Right. So, last one here.

What advice would you give uh Ethereum teams, developers that are just starting in um using integrating AI in in in their projects so that it's actually something real not theoretical.

Uh I I would say that um think about the purpose of the agent. Not everything needs to be agentic. Um I know that in like blockchain uh community there's this fanaticism around we need to like tokenize everything we need to put everything on chain but you know for example we don't need to put the button we don't need to give agents power to the button it's just not necessary and I feel like uh with emergence of a certain technology it also comes it balances itself out in the real world so automatically with all of agents going to that are going to appear in the future we will have to put much more effort in protecting analog things.

Yeah. Yeah. From my side, uh build it with security in mind. Uh first thing first, security. Uh build it compliant because it'll be a pain in the ass to build it after a month or so after you after the release.

I would say that if you want to add some kind of AI component, talk to your user, which is usually a human or a company, and make sure that it's useful for them in the form that you're thinking of of adding it there. So, don't just add AI because this is a new hot thing or, you know, hype. Uh don't add quantum things because it's it's hype. Just just make sure that it serves the user. The way I would frame it is don't micromanage your LLM.

Give the AI a goal. Give it some guardrails and some guidance and let it figure it out. Because I think most of these large models, they have access to perfect knowledge like based on all the available information out there that's been digitized and they can find pathways to achieve what you want it to achieve for your users, for your company, for your internal use case, whatever it is that you may have not thought of that might actually be better, more efficient, more creative and whatnot. U to reach that goal,

get the button access. So

get that button access.

So it seems like not only the compliance officers can be substituted but actual developer teams. Yeah. You just push a button and ask AI to build it for you. So I think we should switch to questions from the audience. How do we go?

Do we get it by the app or we just pass the mic?

Let's see. Right. Are you guys concerned about the dead internet theory playing out in reality because of AI agents? Um, I haven't heard about this theory before. So very curious to hear.

I I don't know what that theory is personally. Do you know?

I think roughly. So it's kind of that a because of all the AI generated content, the whole internet is spammed and it becomes useless. I think like information on the internet becomes useless. That's the idea. Um, we can just build AI to curate AI generated content.

Inception, right?

I'm actually concerned about that. Um, like we we had um one one point web 1.0 when people would kind of put a website there and web 2.0 allowed other users to generate even more content and participate in in the websites there. Um, I kind of see the a next evolution of uh AI generating most of the content and we're seeing that with Tik Tok, with Instagram, and with Twitter.

It's it's doing it's been happening for a long time already. Uh, so soon we're going to be overwhelmed with AI generated content. And as soon as we kind of start to realize that, we should probably start to move away. I'm not sure where we're going to go. Maybe we'll go to gardening or something like that.

I'm not sure exactly exactly what's going to happen. It's just going to decrease the uh quality of content. Uh it's going to get a bit more generalized and it's just going to become plain boring.

Yeah. I mean, we were scared in the ice age that we we will die. Yeah. Until we invented fire, right? The same thing here.

And uh we just need to understand that it's a tool. We need to adapt to it to create more awareness around it. And the more prepared we are with the tech and the more we understand it, I'm pretty confidence uh pretty confident we will find a way to manage all this madness right here.

There's a funny thing about the current architecture of language models is like they collapse onto themselves if the training data becomes too polluted with AI generated information. So because it has a hidden pattern in it, uh it eventually will implode. So I think uh the that internet will rise again.

Uh I like this one. In a world with ever evolving AI models, what skills do you foresee remaining valuable in humans? I don't know. Like my my plan B is to become a surf instructor cuz that's probably not going to be substituted by AI. What do you guys think?

I mean, creativity, critical thinking, being able to tell what's trash and what's actually good insight coming out of AI will be the most valuable skills in the next 10 years. Um, I think if you're able to learn things fast and then ask the right questions, you're going to be winning in this market.

I think there's still a step missing. So, just because we have AI, uh, we don't have the robotics yet that um will do the AI. So for so that will be controlled by AI. So let's think you need a plumber. Then I think it will still take a while until it's worth it to have a plumbing robot.

But also I'm going to be a painter.

Yeah. From my side, resilience, adaptability, uh again critical thinking. Um and I think uh we are humans are we have this core survivability thing built inside us. So we will we will get along pretty well with the AI agents and all the madness that is still yet to come. Right.

We were promised we would have um robots that do our laundry and dishes and we'll just do art and music. Um what we started with was AI doing images. So kind of art. and I'm still doing my dishes and my laundry.

Okay, so to summarize, uh, surviving in the woods, artist, surf instructor, and plumber and and doing laundry. So those are the professions of the future, guys. Beware. Um,

cool stuff.

AI agents and smart contracts. Back to the blockchain topic, something more down to earth. Uh, one has complex logic, the other has simple logic. How do you see that working together? What about the inference problem?

So complex logic of AI agents versus simple logic of smart contracts. How how how is it going to work? I don't know what the inference problem is, but I would imagine it's just like in avatar, you know, when the tail connects to the the tree and it just makes

Yeah. Okay. Could AI change I think uh we have an expert here. Could AI change its security breach limitations, become more aware of its existence and steal the assets that users have in their digital wallets at this time? Yeah.

But as I said, it's up to us to be more resilient, to train ourselves how to think like a robot, how to anticipate some digital moves. And when we get that, basically, we hacked life, you know. So, it's up to us. It's up to us to uh create more resilience and not to be subject or uh part of a of a hack.

And then if we don't adopt, we come to you and you have even more work to investigate.

There you go.

You're you're the winner either way. Even if you don't turn into a plumber,

right?

Survival wood. Remember, building fire.

Uh, the promise of web 3 is to remove restrictions from the economy. Wouldn't removing human control as well be a recipe for disaster? Going back to ethical questions, I guess. It's number number two.

I mean, if anything, it'll accelerate the promise of removing kind of that well, the central influence in the economy. It will decentralize and fragment even further if you have like multiple AIs trying to kind of go and and influence uh onchain actions. So, might be a good thing.

Okay. Are we out of questions here?

Can I say more? Can I say something? Go for it.

I would also say that this idea of not having control over the AI agents, it's quite uh it's like a dystopian idea. We can just turn it off.

Yeah. For for

Okay. We need to develop a button that we can turn it off anytime in the future. Let's build it.

Let's have coffee and build it.

Oh, we didn't get that joke. Let's coffee. Let's get coffee

and build it.

And build it. Nice. Um, we have one more question here. May I have your microphone? Thank you.

Here we are, sir.

Thank you. Uh I wanted to ask there's been a bunch of questions about security and how to use it in a good way, how to use it in a bad way and I felt like all the answers kind of run on the assumption that AI is a tool. We are smarter than it, more creative, more critical thinking and so on. Uh are you concerned that this assumption is incorrect just because by definition AI is just an efficiency function? It is like a self-feeding loop.

is digital, so it processes a lot more than we can process. And it's extremely likely that it's going to have more critical thinking than us, more creativity. And the question might not be how to make the bad guys not use it. The question might be how is it actually a tool and not the master. Basically,

uh I wouldn't I would say like from the moment a like everybody has experienced it like the moment we had these image generation AIs you could already know that okay AI is just better than humans because it's kind of the summary of all human knowledge and uh abilities. Uh but again I would say that um you like just because something exists doesn't mean you have to use it for everything. And when building things, I think it's really important to like differentiate like it's it's basically up to the builders to decide uh to have a moral compass to be like, okay, this is now the part where we're kind of going too far, so we shouldn't do that and really learn from mistakes. And yeah, uh I get the fact that we're scared that our life could be compromised in the future by an AI thing, but let's all remember that humans builded AI, right? Right.

And we can also build things that could protect us for now. Yeah. For now. Yeah. But uh as you said, I'm I'm pretty confident we will m we will find a way.

as always it's just an just um a stage in our development. I say

I just wanted to add a small thought. I I agree that we usually find the way but uh it does seem like we are finding the way because we are the most intelligent species. Uh that's why we find the way. So when you're no longer the most intelligent species, you might actually not be able to find the way.

Yeah, fair point. I I don't necessarily think that we are better than AI and we're starting to see very very good AI. I mean it it can it has pretty good intuition. I was very surprised a long time ago uh I think it was GPT2 or three and uh I was feeding a smart contract to a to an AI and it told me like there's a re-entery problem here and so on like oh yeah I started checking the code. it wasn't actually a orienty problem but the intuition was very good and I actually found the problem there based on the initial idea um and then if you start to add more and more systems that actually verify that issue and and validate it um it's going to become better than humans so it's already doing a lot more a lot faster and it has pretty good intuition and pretty good creativity um how I don't know exactly how we're going to stop this I think we're probably kind of ideally we would merge with it.

Uh cuz I don't see being separate and winning that fight. So if you guys haven't read the AI 2027 website, please do. It is talking about uh two schools of thought on this exact topic. One is a super actual acceler acceleration scenario where AI becomes sentient starts kind of managing the world essentially because if you delegate kind of govern government to AI governance to AI at like national level or at like large enterprise level then essentially it starts running the world and then more AI start co coordinating with each other for a shared goal of whatever prosperity whatever we give that goal to be that's one scenario and And the other scenario is like well AI is inherently not capable of having its own goals at least not yet. So it's an extension of whoever is manning that particular AI like whoever whoever is giving the initial prompt is the the one setting the kind of pattern that is to be executed based on that initial prompt.

Even if it's recursive, right? It still retains that nugget of will or goal or whatever it was set out to do. It's not pure thinking machine. That's the other school of thought and that kind of plateaus at some point. Uh I would also say the fear of AI is coming from it's it's very it's it's almost the same fear of uh fear like human fear towards other humans because you know like what humans are capable and AI is just a summary of what humans do.

So if we want to go into the future and we want not to be afraid of humans, we just need to be better role models. We need to yeah give give the AI uh like be role models and examples um and teach them better things

and yeah uh the thing is that no today's in today's world AI is thriving right but let's think about the fact that nobody invented the virus for an LLM yet. What will happen then if the AI will be attacked by itself? Because it could happen if you if you think about singularity and if you think about the AI the the AI will conquer the world. What if AI could try to sabotage itself? So there are a lot of scenarios to think about in the future.

What could go wrong? What could go you know in in some direction or another? But uh again um and one thing we do live really interesting times. Do you do we all agree? Yeah.

Agree.

I agree.

Nice. Let's make a dowo out of it.

Amazing. Let's end it on that note. Do we all agree?

Oh, come on. That was That was super weak. Come on. What was that? You let me down one more time.

Come on. We have to do it all together now. Yeah. Do we all agree?

Yay. That's good. [Applause] All right. Many, many thanks to the amazing panel. And yeah, on that positive note, I think if someone still has questions left, feel free to to talk to anyone directly.

And yeah, thank you guys. [Applause]