AI x Ethereum: Infrastructure for Human–Machine Coordination

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:10

   Hi everyone. Uh I'm David. I lead the AI team at the fume foundation. So this is not properly an AI talk or although we do a lot of uh work in AI. This is about like uh how we can use uh blockchains and Ethereum to coordinate AI and also to uh keep uh humans kind of relevant and uh in control in this like AGI world that we're approaching.

2. 0:36

   So the vision we have is that uh we want Ethereum to be like the decentralized settlement and coordination backbone uh for machines with humans at the center. So Ethereum for AI uh really means uh Ethereum for humans. So essentially like the idea is that like we want to uh keep giving agency to the users uh control uh and then ownership of their uh data of their memories and of the assets and I'm going to show you like uh what protocols and what technologies we are building for this.

3. 1:13

   So the motivation for us is that like uh Ethereum so far has been like uh uh one of the main settlement layers for like uh this new wave of finance like digital finance. Uh but like uh a settlement layer is only as valuable as the amount of economic activity that it settles and like more and more of this is being intermediated by AI. Uh so we feel that this is kind of like a key area where

4. 1:36

   like um we should invest and um it's funny because like uh I think two years ago I wrote this post called the internet of agents and like people were kind of not seeing this especially like uh even within the blockchain ecosystem they were like okay how are blockchains even relevant for AI but today like there is a lot of problems uh in AI uh that like uh people are talking about like how do we govern uh these things um um how do we like um secure these things and like that's

5. 2:08

   where like the work we've been doing becomes relevant so uh we have like a standard on ethereum which is for agent coordination that I'll describe uh and then um basically with the raise of like autonomous AI like we just seen the beginning but like all these things uh become real and like we are starting to see more usage already So the work we do there is uh really two pillars. One is decentralized coordination as I said. So we want to use like uh blockchains as like a public

6. 2:40

   computer for discovery and trust uh for securing API calls. Uh bots are in bots uh staking uh negotiation and commitment. So like cooperation between different AIs for different manufacturers u even and then like of course all the financial uh uh primitives that like uh the chain can offer.

7. 3:03

   The other area is like um uh this I'm using this acronym crops uh it means censorship resistant, open, private and secure. So these are uh the values that uh um we follow when building like uh the Ethereum platform and these are values that like u AI today completely lacks. So we kind of want to export them uh in AI and this means like investing in local AI, local agents like to interact with the chain. um security auditors like the security industry has

8. 3:35

   been kind of subsidized by uh uh blockchain in the past 10 years and like basically we want to reorient uh uh this to like u um essentially make progress in like AI secured by cryptography and secured by blockchains um protocol agents governance uh and then eventually you can envision this world in which like also the AI production uh chain uh is not just one vertically integrated company but a lot of protocols that are working together to produce like verifiable uh artifacts which are the models.

9. 4:11

   Yeah, maybe I'll go fast here. I don't know how relevant for this audience is but like the team foundation now like is investing like uh quite heavily on this. We have a dedicated team. Uh we have um active collaboration with like privacy and security teams across the ecosystem.

10. 4:26

    And then like in the Ethereum ecosystem there is a lot of like uh builders that are focused on like private AI um and like on uh kind of new applications for agent interperability. Some of them uh you've heard about or you will hear about uh today as well.

11. 4:44

    So here is a road map uh for our team. So as I said we have two areas. one is AI coordination and there is this new standard called ERC804 which is basically like governance for AI agents. So it's like two decentralized registry one as like agent identity uh and the other one um essentially allows people to leave uh and and bots to leave feedback to other bots. Uh so the idea is that like these

12. 5:11

    are primitives that like um as um AI services scale uh will allow um like discovery of AI services similar to what's happening on the internet uh today, right? Like um um you don't have like centralized registry. You have like an open protocol where like these agents that are economic entities are registered and anyone can discover like people can build Google on top of this.

13. 5:37

    uh people cannot build Google on top of like private MCP registries or private A2A registries. Um the other area is privacy and security. Uh we have a bunch of projects uh related to like um uh verifying properties of AI system and of their input data. I'll talk briefly about a couple of them.

14. 5:58

    Um just an overview of like what this ERC 80004 standard like past five months have been like. uh we kind of um published the spec as an open source u uh repo uh 5 months ago. Uh since then it's been adopted by uh every blockchain uh all the existing blockchain networks uh not just Ethereum and not just Ethereum compatible chains. Um we have like hundreds of projects that are building on these standards and like kind of uh reg uh registering agents and

15. 6:31

    using this for agent identities and um there is a lot of primitives that people are building on top of the standards which is very exciting like we feel that kind of like this is kind of a new um uh small network that is kind of like growing in capabilities and size.

16. 6:51

    Hopefully this will become the place where like you go to like discover like new agent or like trust in the future. Um people are building like um new primitives on top as I said. So like uh there is a cool project called sign in with agent which essentially allows um uh agent to like uh prove properties that have been certified uh on the registry in order to access like uh uh different services and APIs. Um and this has become like one of the uh main

17. 7:21

    standards on um the blockchain and beyond. The other thing we're doing is like as Ilia was saying before uh it's kind of important to like uh bring together like researchers uh and like um builders industry practitioners and especially like uh uh for us u since it's like very early in like using blockchains to govern AI is like very important to like u essentially bridge the gap with like

18. 7:50

    the AI research community. uh we've been publish publishing a new ICLR for the past two or three years but we need to continue this so we have like a couple of grants which are focused on like um AI related u blockchain research um I'm just like giving them here um in case there is anyone interested to apply you can also talk to me later uh the other main area uh is um essentially like building uh uh digital

19. 8:23

    infrastructure that is robust and secure in the age of AI. So this is an example uh protocol that we just released. Uh it's called ZK API credits. So essentially like the idea here is that um uh we not only want um uh agents and AIS to be able to like pay for services in a very seamless and efficient way uh which they can already do with like blockchain based payments. So they can pay for like um uh receiving uh your agent can pay for receiving AI tokens.

20. 8:55

    Um uh uh each uh single time it sends a prompt to a cloud LLM doesn't need a subscription. Uh but in this work we felt it's really important to to endow this with privacy. And so the idea is that now you can um uh using some modern cryptography you can anonymously uh pay for uh intelligence or other services.

21. 9:18

    Um this is really cool. Uh we build this protocol that is also like postquantum resistance. Like we feel that essentially like uh in the future uh Ethereum should be uh the most most secure most secure place where you want to store your um assets. you want to store like um uh your memory encrypted.

22. 9:40

    uh is essentially the place where like in um a situation where AI is like uh running wild in the world and like uh hacks are increasing like uh you see Ethereum as kind of like the secure place where the human stays in control and um we have a team there is my direct team and then uh we also have like the ERC4 team and uh we have many other hundreds of builders in the ecosystem as I've said And um yeah, a bunch of

23. 10:11

    references if you uh are interested in this work or in related work we are doing like to train AI agents that can negotiate using escros on chain. Uh please come and talk to