AI as an Interface to the Game: A Round-Up by Johannes Pfeffer | Devcon SEA

[Music] hello my name is uh Johannes feffer and I founded a company a project called hyro that builds human Centric AI agent systems and I'd like to talk about a few aspects of AI as an interface to blockchain so so in February this year there was this article article by vitalic which spurred this uh presentation it's called the promise and challenges of crypto and AI applications who of you have read the article hands up please okay only I see I see like four or five right well in the article um vitalic has identified four main categories of AI and crypto intersection which with with which I agree a bit and sometimes I disagree but let's go through them real quick um so he sees AI as a player of the game as an interface to the game as the rules of the game and as the objective of the game right that is these four uh categories I will have to speed through the initial ones because this is a lightning talk but yeah well AI as the objective of the game it means that the goal is to create models that can compete with anthropic and that can compete with open AI right and AI needs a lot of money to develop and crypts startups have shown to be able to raise a lot of money and large sums and this could actually be a chance to create a competitor that is better aligned with you with Humanity right and there are great projects doing this uh for example the sentian foundation but also bit tensor and more projects are coming up and I I'm like really looking forward because we all need these models then uh there's AI as a rules of the game this is something I will just only very briefly touch on the goal here is to be able to run and call into a trusted Ai and so that can use AI for example as judge because we can prove the the computation we can keep things secret we can prove the training data and we can trust it right we're not very far on that in that area yet but uh yeah hopes are up and the last two categories or the first two ones uh are these AI as an interface to the game and AI as a player in the game and the goal here is to act on the blockchain but how right so first you need to give the AI tools you give them research tools access the web analytics maybe scam detection maybe even transaction simulation lots of apis and then you also need to derive onchain actions from intentions of the user and that results in basically a unified interface to blockchain apps um but why are we excited about this it is because there is no way that humans can manage the complexity of our ecosystem on our own let me illustrate this a little why do we even need AI as an interface well let's say you're just a random dude or a random girl in this world and you have heard there must be an infinite Garden somewhere and you go looking for it right well you dive a bit deeper and there it is ethereum Bitcoin salana goatsy 1,000 blockchains 400 al2s 10,000 apps or more and you're like wow but what happens next is kind of sad because most of the people that join this way they either get scammed they buy a meme coin and then they leave or they get confused because it's all so much and they don't understand it and they leave or maybe they get rich and happy but unfortunately the last category is rather a minor percentage the complexity is just huge the human brain cannot comprehend it fully even after many years in the space and definitely not when you're new and the space of change is only increasing the speed of change is only increasing most people that are uh new don't even have a concrete intention yet right they don't want to bridge from optimism to fuel but they do smell the opportunity they might have a notion of being part of a dow or discover the infinite Garden or say maybe I want to get rich but they don't know how to get there and here comes the thing autonomous agents are in exactly in the same position the agent and we have the same problems and likely the same goals we and the agents are using the same tools the same apis the same everything to achieve these goals so I argue that player in the game and interface to the game is converging with each other from the perspective of the blockchain there is no difference from the perspective of the blockchain there is no difference between a human player and an uh an agentic player the blockchain is agnostic to that it is a neutral layer but there is a difference on another Dimension and that is autonomy let's illustrate these differences there let's Place existing projects on the Spectrum so terminal of Truth is not really autonomous Andy is curating the tweets before they get posted but it's still some of the stuff is foring autonomously higho the product I build is humancentric we're trying to give control over the agents to humans and coinbase based agent system that's just an SDK you can build anything on that Spectrum so let's take a step back and see what this means if agent systems and interfaces are converging and the main differenti differentiator is the degree of autonomy in what type of a world will we live so here's my first take in any complex system with thousands of tools and especially in a system that is neutral and uncensorable we cannot compete with agents agents never get tired they scale well and they you'll never see a wor a worse version of them than now the only option we have if we don't want to give up our own ageny is to make use of Agents we will most mostly use agents and agentic interfaces to do stuff on ethereum and on on all other corners of the internet too and our best bet is that the combination of humans and agents is more effective than agents alone if that's not the case then humanity is a b in big trouble so let's build agents for humans and let's let's make sure that humans are not sidelined in permissionless systems such as ethereum I want humans to in the infinite garden and that's what I'm building towards and with this my time is up and I thank you for listening if you're interested in blockchain versus AI agents follow me on Twitter and check out the human Centric a agent system that we're building at hyro thank you thank you I love the way you concluded it it's all about human and agents together uh any burning question from our audience oh marry and there's a yes hello uh great presentation I just wanted to ask that if we as humans give uh agents the whole permission is there a way to control them because if we give them the whole like transaction power they can do transactions that we do not want as humans to be executed and then we might be in losses and then we might wake up and see our whole account to zero I see yeah of course very good good question and I'm not just saying that because it is one of the central questions when uh we're trying to make humans use AI agent systems right if the agent is autonomous then we don't care right if they lose their money and if they do something well they are autonomous but if we want to be in control there has to be a lot of checks that what the agent wants to do for us on chain is actually what we want to do on chain so our approach is uh to have another model that uh basically tries to match the intent of the user and uh the transaction being suggested and score that right so that but yeah the problem is not solved with that it's just the first step I think we have a long ways to be can to be able to be 100% sure that what the what an agent suggests and wants to do for us is really what we want to do I mean if you say that the agent is autonomous it means that it can take actions on its own right and it's also responsible for its own actions right that's the fully autonomous version okay we have one over here yes um my question is essentially uh could you explain a little bit more about how AI agents are an improvement over something like robotic process Automation and those kinds of thoughts right so process automation is uh something that we've done to to tackle complexity too right but process automation is not creative it turns out that agents are really good at something that we thought would be a domain of humans and that's creativity coming up with new ideas iterations of what has already been done and uh that's like if you to automate a system in robotics or in process control you actually have to do a lot of work right you are doing the work and then after that you you you you gather the fruits from that and you uh you have lots of efficiency but uh agents because they are creative they can actually do single tasks that you do only a single time right and they automate specialized tasks and that's a big big difference uh one last question a quick one uh are there any eips or ercs that are like based on agents or are there any eips like uh propose proposals oh um I'm not really sure that if there are yet any any eips because what that I know there are several prod projects are trying to solve the problem of using AIS from the blockchain calling into but I'm not sure sorry I don't know no wor thank you and that's it for our question session I guess we can reach out to yanus yes after this and let's give him another round of applause